Home
Creators
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register Favorites Keys Review DMs Logout
Community
Telegram ThePornDude
Home Creators Posts Import Register Favorites Logout

LTX-2.3 THE BEST VIDEO AI KING WITH VIDEO EXTEND & MULTI-FRAME! (Patreon)

Published:
2026-03-21 17:54:24
Edited:
2026-06-15 15:56:20
Imported:
2026-06
Tags:

Downloads

Content

EDIT 15 june 2026: right now the ltx node ComfyUI-LTXVideo doesn't work well with the newest comfyui updates. IF YOU HAVE ISSUES, you can use the new FIX-LTXVIDEO-KORNIA.bat file and put it in your ComfyUI_windows_portable folder, next to your run_nvidia_gpu.bat file, run it and it should fix the issues with the node at least until the node receives an update.Hey everyone! I've created a 1-click installer for the BEST FREE VIDEO AI KING, “LTX-2.3”. It’s an absolutely insane open-source model that can generate videos from text, images and even other videos, and now with VIDEO EXTEND AND MULTI-FRAME control, it’s even more powerful than before. This is easily one of the craziest and most fun video models I have ever tried!

You can check out the video right here: https://youtu.be/nOCMsqVujBI

The workflow: https://www.patreon.com/posts/153598145

The installer of course automates the entire install process for comfyUI and the models downloads!

LOCAL INSTALL!!! RECOMMENDED TO START FROM SCRATCH

Download the LTX-2-3-ULTRA-COMFYUI-MANAGER_AUTO_INSTALL.bat

Run the bat file

Wait for the install to finish

The webui will launch, load the LTX-2-3_ULTRA_WORKFLOW.json file and there you go😎

DON'T FORGET TO GO UPDATE COMFYUI BY GOING INSIDE THE UPDATE FOLDER AND RUNNING THE COMFYUI UPDATE BAT FILE THEN CHOOSING THE CORRECT LAUNCHER FOR YOUR GPU!!

IF YOU ALREADY HAVE COMFYUI INSTALLED USING A PREVIOUS INSTALLER:

Download the LTX-2-3-MODELS-NODES_INSTALL.bat and place it in the ComfyUI_windows_portable\ComfyUI folder

Run the bat file

Wait for the install to finish

???

Profit 😎

IF YOU GET AN ERROR WHEN LAUNCHING COMFYUI AFTER USING THIS INSTALLER SAYING "AssertionError: Torch not compiled with CUDA enabled" it's ok don't panic.

Just go inside the python embedded folder, click on the folder path, type cmd press enter, this will bring the command prompt window and there type:

python.exe -m pip uninstall torch

python.exe -s -m pip install --upgrade torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128

--->replace cu128 by your cuda version (if you don't know your cuda version just open a new cmd window and type nvcc --version it will then tell you at the end Cuda compilation tools, release 12.8, V12.8.93 or something different like 12.6, in that case just use cu126 instead of cu128)IF YOU ARE USING RUNPOD:

Create an account if you haven't already: Runpod

Click on Pod (on the left side) then click deploy

Choose a GPU with at least 24gb of VRAM (4090 is best), click here for the comfyui template (aitrepreneur/comfyui), then edit the template and choose 100gb for both the container and volume disk, then deploy on demand

Go to my pods, wait for everything to finish and then click "connect", then "Connect to HTTP SERVICE port 3000" and click on the manager to update comfyui and restart it

Then go to the port 8888, go inside the comfyui folder and then drag and drop the LTX-2-3-AUTO_INSTALL-RUNPOD.sh file on the left side of the UI then click on the "Terminal" icon on the right side on the UI

Copy and paste these two lines then press enter:

chmod +x LTX-2-3-AUTO_INSTALL-RUNPOD.sh

./LTX-2-3-AUTO_INSTALL-RUNPOD.sh

Wait for everything to be installed

Go back to the port 3000 and click on the manager and restart it

Load the LTX-2-3_ULTRA_WORKFLOW.json file and there you go😎

Also use the command:

tail -f comfyui.log

inside the workspace/logs to get the real time logs so that you always know when the install is done. It will then ask you to restart, so click on that restart button, wait for the webui to reboot, click refresh once it's done and there you go.IF YOU GET OOM error :Click on the little 3 dots button, arguments variables then enter

EXTRA_ARGS --reserve-vram 10 --cache-nonethen restart the pod

As always, supporting me on Patreon allows me to keep creating helpful resources like this for the community. Thank you for your support - now go have some fun😉!

Files

Comments

H-BOX

You rox dude 2nd comment edit:btw I gotta wait for more .. interesting loras before I go for it..

todd beebe

i am getting this error with the audio vae I got one with video too but i updated and that fixed the video one but still getting with audio this is error: RuntimeError: ERROR: VAE is invalid: None If the VAE is from a checkpoint loader node your checkpoint does not contain a valid VAE. File "C:\ai-software-stuff\comphyui_01_17_26\ComfyUI_windows_portable\ComfyUI\execution.py", line 525, in execute output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\ai-software-stuff\comphyui_01_17_26\ComfyUI_windows_portable\ComfyUI\execution.py", line 334, in get_output_data return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\ai-software-stuff\comphyui_01_17_26\ComfyUI_windows_portable\ComfyUI\execution.py", line 308, in _async_map_node_over_list await process_inputs(input_dict, i) File "C:\ai-software-stuff\comphyui_01_17_26\ComfyUI_windows_portable\ComfyUI\execution.py", line 296, in process_inputs result = f(**inputs) File "C:\ai-software-stuff\comphyui_01_17_26\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-KJNodes\nodes\nodes.py", line 2427, in load_vae vae.throw_exception_if_invalid() ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^ File "C:\ai-software-stuff\comphyui_01_17_26\ComfyUI_windows_portable\ComfyUI\comfy\sd.py", line 847, in throw_exception_if_invalid raise RuntimeError("ERROR: VAE is invalid: None\n\nIf the VAE is from a checkpoint loader node your checkpoint does not contain a valid VAE.")

todd beebe

fixed it by updating all custom nodes

cool1

I get errors like: ...\comfy_extras\nodes_lt.py", line 671, in execute latents = av_latent["samples"].unbind() ~~~~~~~~~^^^^^^^^^^^ TypeError: 'NoneType' object is not subscriptable Prompt executed in 1.07 seconds I did "update all"/ the .bat update. Looking at the comfyui nodes in the workflow, quite a few have conflicts.

Manic

Is there any plans to add sage attention to these?

Skippy

Make sure to enable "No Custom Audio IMG2VID" if you don't have an audio file uploaded into the Load Audio node. That fixed it for me.

brian

does this break the LTX-2.0 he released a few months ago if i install it in the same portable comfy file?

CHRIS MCEVOY

I was trying this with 720p but got: RuntimeError: Sizes of tensors must match except in dimension 1. Expected size 4096 but got size 3840 for tensor number 1 in the list. So i assumed it was because I set it to 720 but put it back to 1080p and still get the same error when rendering.

Tyler

HI there, I'm getting thie invalid VAE error also. Do you know which custom node in the workflow was causing the issue? Updating all of them seems like a lot.

Tim M

Same here. The audio VAE is throwing the error. I'm using the medium model, Q5_K_S if that makes any difference. I've updated Comfy, and I updated every custom node, yet the error persists.

Tim M

AUDIO VAE FIX: Many of us were erroring on the audio VAE even after updating all nodes. Claude found the solution for me. Here's the fix. Update your ComfyUI-KJNodes to the latest version. Specifically: 1. Navigate to your ComfyUI/custom_nodes/ComfyUI-KJNodes folder 2. Run git pull 3. Restart ComfyUI Cheers everyone!

Ash

Kerfelox

I get this issue that using RunPod and using 300gb network storage. i install everything with the runpod.sh file and once it is done i update comfyui i will not be able to use it even after creating new pod it simply will not load. And if i dont update it cannot locate the still missing custom nodes which leads to not working workflow. Any solutions or someone who also face this problem? This problem came up since a month or so, before that i was able to use every workflow and installations with no problem.

DrHans Brewery

F YOU ALREADY HAVE COMFYUI INSTALLED and are using the model nodes installer .bat file Fix for “Could not find python… System32” error If you get an error like: Could not find python at "C:\Windows\System32\..\python_embeded\python.exe" The issue is not the install — it’s how the .bat file is launched. Solution: Go to: ComfyUI_windows_portable\ComfyUI Click the address bar and type: cmd Press Enter (this opens Command Prompt in the correct folder) Type: LTX-2-3-MODELS-NODES_INSTALL.bat and press Enter Running the script this way ensures it uses the correct relative path to Python.

Ash

Okay I've found the folder where its kept, and upon relaunching it says workflow uses a missing node "Frame Slicer", first run pip install "-U --pre comfyui-manager" in your python environment to install Node Manager, then restart ComfyUI with the "--enable-manager" flag. What does it mean by python environment? Can I run it in CMD? sorry if this is a noob question

cool1

I had an error message mentioning the audio previously and then i disabled the audio buttons that were enabled by default. The error message in the cmd window had mentioned not finding the audio file. This new error I think is for something different. I can re-check later but as far as I know it was when the audio options were already disabled and it didn't mention not finding the audio.

Ian Solo

Hi Guys, everything works fine except that when I use Image to video with audio, the character does not speaks despite mentionning "perfect lip sync" in the prompt.. I just got a video with the sound of my audio but no lip sync. Any clue ?

Ash

hey, can I ask what do you mean by run git pull? I cant see any file or option to do that, thanks in advance

Tim M

Sure. You will need Git installed for this command to work. Git records the history of code changes and allows you to load snapshots of code called "commits". This is version control. To install Git: - Download the latest version from the official Git for Windows website. - Run the downloaded executable file. - Follow the prompts in the setup wizard. The default settings are generally recommended for most users, though you can select your preferred text editor (e.g., Notepad++ or VS Code instead of the default Vim). Ensure that the option to "Use Git from the Windows Command Prompt" is selected for our purposes. - Complete the installation. - Verify the installation by opening the Command Prompt or PowerShell (this is an advanced command prompt with more functionality than the basic one - they both work) and typing "git --version" Once you've done this, you should be able to run "git pull" through your command prompt. Use the commands inside the quotes: - Type "cmd" in your computer's search bar to run the command prompt (or, optionally, search for and run Windows PowerShell) - Once in the command prompt, you can use certain commands (command prompt/DOS commands) to navigate within the directories of your computer's drives. - "dir" will show you the current directory contents. If the directory is large, "dir /w /p" will list the contents in a wide format and pause each time the window is filled so you can see every file and directory. - When using "dir" you will see the word DIR on the list next to each directory name within your current directory. Listings without DIR are usually just files. - "cd directory-name" will move you to the listed directory. For example, if you're in the C: root directory and want to go to C:/users, you would type "cd users" to move down into that directory. - You can move down multiple directories at once as long as you spell the directory names correctly and use forward slashes between them without spaces, i.e. if you're in C:, typing "cd users/me/downloads" would take you to C:/users/me/downloads in one command as long as all those directory names were correct - "cd .." will move you back up one directory. If you're in C:/users, this would return you to C: - "cd \" That's a backslash. This will move you back to the root directory for anywhere, i.e. if you're in C:/users/me/desktop, typing "cd \" will take you to C: - To change drives, type the drive letter with a colon, like: "D:" or "C:", which will take you to that drive's root directory. - Those commands should be enough to allow you to find your way to the "ComfyUI/custom_nodes/ComfyUI-KJNodes" directory in your comfy installation Once you're in that ComfyUI-JKNodes directory, in the command prompt, with Git installed, you then type "git pull" in the command prompt to update the directory contents with the latest version saved in their repository. Now that I've explained it, I realize how complicated it must sound if you don't know DOS/command prompt commands. Once you learn how to navigate your drives using those basic commands, it gets easier!

Roma SoS

Hello everyone, I'm a bit of a beginner and I'd like to know if it's possible to install all of this, including ComfyUI and its future dependencies, in a directory other than the root directory, C:\. Currently, I'm using Pinokio, which does this for me; it installs everything in my d:\pinokio directory without affecting my C:\ drive. Any ideas on how to overcome my lack of experience?

Alan Smile

The script is designed to install exactly where it is run from.. so if you want to install it to "d:\ai-stuff" then: 1. Open your D: drive and create a new folder named ai-stuff 2. Drag and drop (or copy and paste) your LTX-2-3-ULTRA-COMFYUI-MANAGER_AUTO_INSTALL.bat file into that new d:\ai-stuff folder. 3. Run it: Double-click the .bat file from inside d:\ai-stuff The script will detect its new home and download and install everything right there, inside d:\ai-stuff\ComfyUI_windows_portable .

todd beebe

I am not sure, for me it only took like a minute or two to update them.

todd beebe

Try installing missing nodes to see if that works. If not after installing missing nodes go back and update custom nodes again.

todd beebe

Yes you can install it in any drive you like. Just download his .bat file into the drive you want to install it in, and run it in there. Everything should remain localized in the directory and drive you install it in.

BuAuch

I had a few good tries without any issues, but then my GPU (5090) starts running full fans, screen goes black, nothing responds, and I have to force restart. This has been happening more often. My PSU is 1200. It's only happening with the LTX 2.3, didn't have this problem running Wan. Has anyone else come across this issue? I've updated by GPU drivers, did a fresh install using the one-click installer.

todd beebe

Did you try in your prompt, let's say it's a woman in a coffee shop. Try first describing the image like: the dark hair woman wearing the blue sun dress sitting in the dimly lit coffee shop looks up as she sips her coffee and softly says, "Hey I would like a doughnut please."

todd beebe

It shouldn't as long as you still have the models for the ltx-2 and the workflow

PJ

That would be nice to have multi frame generation feature, like eg. 10 frames and can generate >1 minute long video.

Roma SoS

Oh, thank you for the quick reply. So if I do that, nothing will go to C:\? Not even future updates?

RavenWT

Hey ! Thank your for the workflow and the video! The example with Asmongold is .... waou... but be careful, his community could shit storm you for this :s I'm a bit worried....

Benjamin GULMEZ

Yop, how to add more steps for the complex scene ?

BadgerWi Johnson

Runs fine starts generating then I am getting the following error "ValueError: not enough values to unpack (expected 4, got 3)" I have the log if anyone wants to see it.

Akind

thnx for this. Is it NSFW? I tried oe prompts with no success. thnx in advance

Josh Atencio

Runpod constantly times out and is always trying to reconnect, until it just dies. Anyone else have that issue?

brian

couple issues with this. 1st the json file uploaded if tried to use straight out of the download doesn't like the 2.3 VAE video and audio files provided. it give me this error: "Error(s) in loading state_dict for VideoVAE: size mismatch for decoder.conv_in.conv.weight: copying a param with shape torch.Size([1024, 128, 3, 3, 3]) from checkpoint, the shape in current model is torch.Size([256, 128, 3, 3, 3]). size mismatch for decoder.conv_in.conv.bias: copying a param with shape torch.Size([1024]) from checkpoint, the shape in current model is torch.Size([256])." the error goes away after i switch it to the "LTX2_video_vae_bf16" and "LTX2_audio_vae_bf16" but then it gives the error: 'Linear' object has no attribute 'weight' this is on the comfy installer he made for LTX2 3 months ago. EDIT: updated comfyui and the VAE are passing at least initially now but the Checkpoint "ltx-2.3-22b-dev-Q8_0.gguf" doesn't work and gives me RuntimeError: Error(s) in loading state_dict for LTXAVModel: size mismatch for audio_embeddings_connector.learnable_registers: copying a param with shape torch.Size([128, 2048]) from checkpoint, the shape in current model is torch.Size([128, 3840]). size mismatch for audio_embeddings_connector.transformer_1d_blocks.0.attn1.q_norm.weight: copying a param with shape torch.Size([2048]) from checkpoint, the shape in current model is torch.Size([3840]). size mismatch for audio_embeddings_connector.transformer_1d_blocks.0.attn1.k_norm.weight: copying a param with shape torch.Size([2048]) from checkpoint, the shape in current model is torch.Size([3840]).

Aitrepreneur

Ltx 2 loras are compatible with the 2.3 btw

Aitrepreneur

The error is coming from VAELoaderKJ on the audio VAE. Try updating KJNodes or delete the folder from the ComfyUI_windows_portable\ComfyUI\custom_nodes folder, then delete the LTX23_audio_vae_bf16.safetensors from the ComfyUI_windows_portable\ComfyUI\models\vae fodler. Then use the LTX-2-3-MODELS-NODES_INSTALL.bat put that installer into the ComfyUI_windows_portable\ComfyUI folder and run it, it will reinstall all the missing nodes and models.

Aitrepreneur

This seems like either a corrupt ltx 2.3 model or you selected an incorrect model

Aitrepreneur

Btw no need to use git pull, you can just go inside the update folder and run the update comfyui bat file

Aitrepreneur

Yes, unfortunately I have noticed the similar issues as well since the ltx 2.0 model came out, it's the only model where I have to use a separate bat launcher with memory offload. Not even counting the fact that comfyui still has memory leakage issue....the joy of open source... but yeah this indeed only happens with ltx models. The only way I found is that sometimes after a few generations, I'll close comfyui then restart it, to avoid and of these issues

Aitrepreneur

You can actually, there is a special node to do that, I wanted to implement it but after multiple tests, there are too many flickers and bad transitions between frames, 3 frames is often the limit for correct frames transition, at least for now

Aitrepreneur

Yeah it's fine, as I said it's just a bit of banter, just a random example I found of a low quality video with bad lighting and a character that ltx doesn't know so that the voice cloning would work fine

Aitrepreneur

No need for that, I've changed the way the previous workflow works, if you want better quality, you can just change the sampler from euler to something like euler_ancestral_cfg_pp but it will take longer to generate

Aitrepreneur

yes, go to civitai and filter for lora for ltx 2.0 and 2.3, 2.0 loras work for 2.3 as well

Aitrepreneur

The error is coming from VAELoaderKJ on the audio VAE. Try updating KJNodes or delete the folder from the ComfyUI_windows_portable\ComfyUI\custom_nodes folder, then use the LTX-2-3-MODELS-NODES_INSTALL.bat put that installer into the ComfyUI_windows_portable\ComfyUI folder and run it, it will reinstall all the missing node

Tim M

I actually updated ComfyUI, and then updated custom nodes, twice, yet the error persisted. I needed to run git pull before it would work. This is how Claude explained it (and I'm glad I didn't have to spend an hour researching this): "This is a well-known version mismatch issue between the LTX 2.3 audio VAE model and the ComfyUI-KJNodes custom node. Several people have run into exactly this. What's happening: The audio VAE has different layer names than what older versions of the KJNodes loader expect, so the loader doesn't recognize it Hugging Face and throws a "VAE is invalid: None" error. Multiple users confirmed that doing a git pull in the KJNodes folder resolved the issue. If that doesn't work, try uninstalling ComfyUI-KJNodes entirely, reinstalling it, and restarting."

brian

same result. RuntimeError: Error(s) in loading state_dict for LTXAVModel: size mismatch for audio_embeddings_connector.learnable_registers: copying a param with shape torch.Size([128, 2048]) from checkpoint, the shape in current model is torch.Size([128, 3840]). I'm using all the base models and settings from your json

Pizza

Amazing videos, but question. The more I use this the slower it gets. Now it's stuck using image to video after like 20 videos. Is there a cache to clear or something?

Instruments of Destruction

I've been getting the following error when I run it: Node 'ID #1537:1530' has no class_type. The workflow may be corrupted or a custom node is missing.

H-BOX

It is ??! That rox

brian

finally got it working by just installing it from scratch on it's own portable. Word of advice, don't even bother trying to install on an existing comfy install. not worth the headache.

BuAuch

I noticed it just happened on my WAN install too. What's the memory offload that you set up? I've tried adding -reserve-vram 2 --lowvram to the launch. This allowed me to do about 3 or 4 generations before the GPU crash.

Mark

yeah his community might actually be part of this community and thought it was funny too

Bob

EXACTLY!!!!! trust this guy

Peter J Walker

I'm finding censorship, a slight glimpse of panty and i get a grid of purple dot over the image! why?

Ramjet

installed in on my existing install. worked no problems. just lucky i guess.

Ramjet

i tried a couple of nsfw lora's and got some results that looked like crack addicts. frightening :-)

Bob

where how can i edit steps?

Ken L. Black

Hi there. I am getting this error when running the "Image to Video" workflow. Any ideas? (And yes; I updated Comfy AND custom nodes...): TypeError: 'NoneType' object is not subscriptable File "C:\ComfyUI_windows_portable\ComfyUI\execution.py", line 525, in execute output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\ComfyUI_windows_portable\ComfyUI\execution.py", line 334, in get_output_data return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\ComfyUI_windows_portable\ComfyUI\execution.py", line 308, in _async_map_node_over_list await process_inputs(input_dict, i) File "C:\ComfyUI_windows_portable\ComfyUI\execution.py", line 296, in process_inputs result = f(**inputs) File "C:\ComfyUI_windows_portable\ComfyUI\comfy_api\internal\__init__.py", line 149, in wrapped_func return method(locked_class, **inputs) File "C:\ComfyUI_windows_portable\ComfyUI\comfy_api\latest\_io.py", line 1764, in EXECUTE_NORMALIZED to_return = cls.execute(*args, **kwargs) File "C:\ComfyUI_windows_portable\ComfyUI\comfy_extras\nodes_lt.py", line 671, in execute latents = av_latent["samples"].unbind() ~~~~~~~~~^^^^^^^^^^^

Cheese Plant

I received an Import Error when trying to use the VIdeo Extender. ImportError 1537:1525 NormalizeAudioLoudness In the terminal it reads:- ImportError("pyloudnorm package is not installed")

Bob

100% WORKING 1. create new folder name something "LTX 2026" 2. move LTX-2-3-ULTRA-COMFYUI-MANAGER_AUTO_INSTALL.bat into that folder 3. run it in the folder (it will fresh install all new comfyui with all nodes and models) some models you may want to find your self and add into clips or models or loras. 4. forget your old comfyui you've been using WORKS

THEO-Ed animation

Hello, there – Not sure if I'm doing something wrong, but this is what I get once I've got everything installed and I try to fire it up (sorry about the tl/dr, but I don't know what's important and what isn't): E:\LTX\ComfyUI_windows_portable>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build E:\LTX\ComfyUI_windows_portable\python_embeded\Lib\site-packages\requests\__init__.py:113: RequestsDependencyWarning: urllib3 (2.6.3) or chardet (7.2.0)/charset_normalizer (3.4.5) doesn't match a supported version! warnings.warn( [START] Security scan [DONE] Security scan ## ComfyUI-Manager: installing dependencies done. ** ComfyUI startup time: 2026-03-23 13:05:17.717 ** Platform: Windows ** Python version: 3.13.11 (tags/v3.13.11:6278944, Dec 5 2025, 16:26:58) [MSC v.1944 64 bit (AMD64)] ** Python executable: E:\LTX\ComfyUI_windows_portable\python_embeded\python.exe ** ComfyUI Path: E:\LTX\ComfyUI_windows_portable\ComfyUI ** ComfyUI Base Folder Path: E:\LTX\ComfyUI_windows_portable\ComfyUI ** User directory: E:\LTX\ComfyUI_windows_portable\ComfyUI\user ** ComfyUI-Manager config path: E:\LTX\ComfyUI_windows_portable\ComfyUI\user\__manager\config.ini ** Log path: E:\LTX\ComfyUI_windows_portable\ComfyUI\user\comfyui.log Prestartup times for custom nodes: 0.0 seconds: E:\LTX\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Easy-Use 0.0 seconds: E:\LTX\ComfyUI_windows_portable\ComfyUI\custom_nodes\rgthree-comfy 1.7 seconds: E:\LTX\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Manager E:\LTX\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\cuda\__init__.py:184: UserWarning: cudaGetDeviceCount() returned cudaErrorNotSupported, likely using older driver or on CPU machine (Triggered internally at C:\actions-runner\_work\pytorch\pytorch\pytorch\c10\cuda\CUDAFunctions.cpp:88.) return torch._C._cuda_getDeviceCount() > 0 Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_mxfp8', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_mxfp8', 'scaled_mm_nvfp4']} Found comfy_kitchen backend triton: {'available': False, 'disabled': True, 'unavailable_reason': "ImportError: No module named 'triton'", 'capabilities': []} Found comfy_kitchen backend cuda: {'available': False, 'disabled': False, 'unavailable_reason': 'CUDA not available on this system', 'capabilities': []} Checkpoint files will always be loaded safely. Windows fatal exception: access violation Stack (most recent call first): File "E:\LTX\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\cuda\__init__.py", line 424 in _lazy_init File "E:\LTX\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\cuda\__init__.py", line 1094 in current_device File "E:\LTX\ComfyUI_windows_portable\ComfyUI\comfy\model_management.py", line 206 in get_torch_device File "E:\LTX\ComfyUI_windows_portable\ComfyUI\comfy\model_management.py", line 256 in File "", line 488 in _call_with_frames_removed File "", line 1023 in exec_module File "", line 935 in _load_unlocked File "", line 1331 in _find_and_load_unlocked File "", line 1360 in _find_and_load File "E:\LTX\ComfyUI_windows_portable\ComfyUI\execution.py", line 17 in File "", line 488 in _call_with_frames_removed File "", line 1023 in exec_module File "", line 935 in _load_unlocked File "", line 1331 in _find_and_load_unlocked File "", line 1360 in _find_and_load File "E:\LTX\ComfyUI_windows_portable\ComfyUI\main.py", line 197 in E:\LTX\ComfyUI_windows_portable>echo If you see this and ComfyUI did not start try updating your Nvidia Drivers to the latest. If you get a c10.dll error you need to install vc redist that you can find: https://aka.ms/vc14/vc_redist.x64.exe If you see this and ComfyUI did not start try updating your Nvidia Drivers to the latest. If you get a c10.dll error you need to install vc redist that you can find: https://aka.ms/vc14/vc_redist.x64.exe E:\LTX\ComfyUI_windows_portable>pause Press any key to continue . . . There seems to be a "fatal exception error," but I don't know what I did here. Any ideas, K?

Brian

Same error after doing what was suggested too.

brian

i would just do the fresh comfy install if you have the space. Comfy is too finicky for it to handle the LTX-2-3-MODELS-NODES on an existing comfy installer reliably. when installed the one from scratch it works fine now.

Kurt Claes

Getting the same error...

Kurt Claes

I found what it was on my end. I turned of the IMG2VID audio too, and it did not like it...

Shap

Hi there. I tried the "image to video" workflow to make an avatar talk with a custom audio file, but there is no lip sync at all. Any guess what's wrong?

Ian Solo

I copied the way of prompting from the video but just get a voice over and no lip sinc. I use the I2V with a pre recorded audio and the following prompt: a young Asian woman with long, straight, dark brown hair and fair skin, sitting cross-legged on a textured, gray, velvet chair. She wears a white, short-sleeved, cropped t-shirt that reveals a sliver of her midriff, and light blue, high-waisted jeans. Her white scrunchie adorns her right wrist, and she has white sneakers on her feet. She rests her chin on her right hand, gazing directly at the camera with a neutral expression. On the chair to her left, there's a small black purse and a plush toy resembling a pink and white rabbit. She looks at the camera and is talking the whole time with perfect lip sync to the audio, keeping direct eyes contact with the camera and reacting with slight eyebrow raises and understated facial expressions as the shot holds on her

jonescrusher1

I'm finding it crushingly slow to generate anything, seems to be ram bottleneck even though im running a pod with 60gb ram, takiing 15 -20 minute per generation, assuming it doesn't stall completely

Aaron Aa

I am not able to get any of these workflows working on my new 4090 computer. but they do work on my older computer.

Aaron Aa

sorry I'm new at this. what software and programs have to be installed on my new computer that the workflow depends upon?

Baris Kilic

it has autoclear vram inside the subworkflow

Pizza

Thanks, I reloaded it and it's working fine now. But for some reason most of my videos have subtitles lol, and it looks like it's in another language that doesn't exist. Any clue on that one?

Pizza

Any idea why the videos get subtitles in another weird language?

Baris Kilic

Somehow with img to video, the settings for width and height don't get applied, is there a fix?

Christoph Bauer

Hy, sorry im new here. I cant install cause it says i need a 16Bit Computer. What im doing wrong

gomez abel

Hey — first off, thanks for the LTX 2.3 workflow, I’ve been working through your setup. I’m running into a blocking issue and wanted to sanity check what exact stack you’re using. Current state on my side: * Using your JSON workflow (latest video) * Models: * ltx-2.3-22b-dev-Q4_K_S.gguf * gemma_3_12B_it_fp4_mixed.safetensors * ltx-2.3_text_projection_bf16.safetensors (2.31GB, SHA256 verified) * LTX23_video_vae_bf16 + audio VAE * ComfyUI updated to latest master * Nodes present: * ComfyUI-LTXVideo * ComfyUI-KJNodes * rgthree-comfy Issue: * Workflow loads, but CLIP load crashes with: ValueError: buffer length must be a multiple of element size (2) (triggered inside load_clip) * If I remove KJNodes → workflow breaks (missing VAELoaderKJ) * If I keep KJNodes → runtime CLIP/VAE conflicts Questions: 1. Are you still using KJNodes in your current setup, or did you move fully to LTXVideo nodes? 2. What exact ComfyUI commit or installer build are you on? 3. Are you using the fp4_mixed Gemma or a different variant? 4. Did you modify the workflow to remove VAELoaderKJ? Feels like a version mismatch between node implementations vs model expectations. Any guidance on the exact working combo would help a ton — I’m very close but stuck at the final load_clip stage. Appreciate the work you put into this.

Peter J Walker

why am I getting a grid of purple dots flashing on screen for a second or so about half way, It like water marking or some kind of censor shit???

CHRIS MCEVOY

I just used you bat to do a clean install as I was using comfy EZi before, It now works perfectly and with out changing anything a T2V took only 223 seconds on my 5090. I recommend people dont run the bat file on the desktop as I did at first and onedrive got involved and it was a mess, I put my bat file on my D drive and it works great.

Aitrepreneur

nope, never saw any of that could just be your prompt

Aitrepreneur

I would need the full error log to be sure. It could be a corrupt clip model so maybe delete it and use the installer to redownload the correct model once again. The workflow does use kjnodes throughout and also for loading audio + video vaes

Aitrepreneur

you need to download and install 7zip, it usually does it automatically but for some people if you have a prior incorrect install it doesn't work well

Aitrepreneur

It's the model unfortunately, change your prompt and use another random seed

Aitrepreneur

Just use the LTX-2-3-ULTRA-COMFYUI-MANAGER_AUTO_INSTALL.bat it will install everything you need from scratch

Aitrepreneur

ram doesn't matter if you got enough vram though, what GPU have you rented?

Aitrepreneur

This seems like a GPU issue, what gpu do you have?

John Lee

Hi, thanks so much for this workflow. In the extend video section, can I use custom audio? The generated voice doesn't quite match the original video. Thanks in advance!

Peter J Walker

ok an update, it happens on clips of 30 seconds or longer. 10 sec or 20 no problem but after that 30 sec 60 sec you get this weird grid of dots like a disco lights for a second. you get it at 25 seconds as well but not so much. It happens in the last second.

Aaron Aa

I have run this bat file, everything seems to load up properly but ComfyUI does not launch and I get these errors. 0/24 [pywin32] WARNING: The scripts pywin32_postinstall.exe and pywin32_testall.exe are installed in 'G:\AITrepreneur\ComfyUI_windows_portable\python_embeded\Scripts' which is not on PATH. Consider adding this directory to PATH or, if you prefer to suppress this warning, use --no-warn-script-location. ━━━╺━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 2/24 [yapf] WARNING: The scripts yapf-diff.exe and yapf.exe are installed in 'G:\AITrepreneur\ComfyUI_windows_portable\python_embeded\Scripts' which is not on PATH. Consider adding this directory to PATH or, if you prefer to suppress this warning, use --no-warn-script-location. ━━━━━━━━━━╺━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 6/24 [trimesh] WARNING: The script trimesh.exe is installed in 'G:\AITrepreneur\ComfyUI_windows_portable\python_embeded\Scripts' which is not on PATH. Consider adding this directory to PATH or, if you prefer to suppress this warning, use --no-warn-script-location. WARNING: The script tabulate.exe is installed in 'G:\AITrepreneur\ComfyUI_windows_portable\python_embeded\Scripts' which is not on PATH. Consider adding this directory to PATH or, if you prefer to suppress this warning, use --no-warn-script-location. Checkpoint files will always be loaded safely. G:\AITrepreneur\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\cuda\__init__.py:184: UserWarning: cudaGetDeviceCount() returned cudaErrorNotSupported, likely using older driver or on CPU machine (Triggered internally at C:\actions-runner\_work\pytorch\pytorch\pytorch\c10\cuda\CUDAFunctions.cpp:88.) return torch._C._cuda_getDeviceCount() > 0 Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_nvfp4']} Found comfy_kitchen backend cuda: {'available': False, 'disabled': False, 'unavailable_reason': 'CUDA not available on this system', 'capabilities': []} Found comfy_kitchen backend triton: {'available': False, 'disabled': True, 'unavailable_reason': "ImportError: No module named 'triton'", 'capabilities': []} If you see this and ComfyUI did not start try updating your Nvidia Drivers to the latest. If you get a c10.dll error you need to install vc redist that you can find: https://aka.ms/vc14/vc_redist.x64.exe G:\AITrepreneur\ComfyUI_windows_portable>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build G:\AITrepreneur\ComfyUI_windows_portable\python_embeded\Lib\site-packages\requests\__init__.py:113: RequestsDependencyWarning: urllib3 (2.6.3) or chardet (7.3.0)/charset_normalizer (3.4.4) doesn't match a supported version! warnings.warn( [START] Security scan [DONE] Security scan ## ComfyUI-Manager: installing dependencies done. ** ComfyUI startup time: 2026-03-24 10:17:26.440 ** Platform: Windows ** Python version: 3.13.11 (tags/v3.13.11:6278944, Dec 5 2025, 16:26:58) [MSC v.1944 64 bit (AMD64)] ** Python executable: G:\AITrepreneur\ComfyUI_windows_portable\python_embeded\python.exe ** ComfyUI Path: G:\AITrepreneur\ComfyUI_windows_portable\ComfyUI ** ComfyUI Base Folder Path: G:\AITrepreneur\ComfyUI_windows_portable\ComfyUI ** User directory: G:\AITrepreneur\ComfyUI_windows_portable\ComfyUI\user ** ComfyUI-Manager config path: G:\AITrepreneur\ComfyUI_windows_portable\ComfyUI\user\__manager\config.ini ** Log path: G:\AITrepreneur\ComfyUI_windows_portable\ComfyUI\user\comfyui.log Prestartup times for custom nodes: 0.0 seconds: G:\AITrepreneur\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Easy-Use 0.0 seconds: G:\AITrepreneur\ComfyUI_windows_portable\ComfyUI\custom_nodes\rgthree-comfy 1.8 seconds: G:\AITrepreneur\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Manager Checkpoint files will always be loaded safely. G:\AITrepreneur\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\cuda\__init__.py:184: UserWarning: cudaGetDeviceCount() returned cudaErrorNotSupported, likely using older driver or on CPU machine (Triggered internally at C:\actions-runner\_work\pytorch\pytorch\pytorch\c10\cuda\CUDAFunctions.cpp:88.) return torch._C._cuda_getDeviceCount() > 0 Found comfy_kitchen backend triton: {'available': False, 'disabled': True, 'unavailable_reason': "ImportError: No module named 'triton'", 'capabilities': []} Found comfy_kitchen backend cuda: {'available': False, 'disabled': False, 'unavailable_reason': 'CUDA not available on this system', 'capabilities': []} Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['apply_rope', 'apply_rope1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'scaled_mm_nvfp4']} G:\AITrepreneur\ComfyUI_windows_portable>echo If you see this and ComfyUI did not start try updating your Nvidia Drivers to the latest. If you get a c10.dll error you need to install vc redist that you can find: https://aka.ms/vc14/vc_redist.x64.exe If you see this and ComfyUI did not start try updating your Nvidia Drivers to the latest. If you get a c10.dll error you need to install vc redist that you can find: https://aka.ms/vc14/vc_redist.x64.exe G:\AITrepreneur\ComfyUI_windows_portable>pause

Aitrepreneur

for this you can use the frame slice node and increase the amount of frame to be cut at the end

Aitrepreneur

not for now, need to change the workflow for that, the whole point of this workflow is to just use the video as a reference only to generate the rest

Alpha Ghost 47

What settings are best to keep Facial detail integrity. When i try extended the face instantly changes. How did you get the Asmon clip to maintain the same Facial details? Was there any specific settings?

John Lee

Thanks for the quick response. Consistency still seems to be the biggest hurdle in AI, great visuals often get ruined by mismatched audio. If LTX 2.3 allows custom audio in extended videos, it would be a big breakthrough for anyone scaling AI video with open source tools without compromising quality. Thanks for the workflow, it's the best one I've tried yet.

Aitrepreneur

Yes you can use custom audio for extended videos, just need to modify the existing workflow. I'll be making a v2 version of the workflow soon anyway

Aitrepreneur

The problem is not settings, ltx doesn't know who that character is so it's trying its best to approximate what the character really looks like when talking, if you got enough seconds of video references it can do a pretty good job but it will never be perfect. For that you'll need to train a video lora of the character to get the face and voice right. For now you can maybe increase the number of seconds for the reference video and see if it's better

Alpha Ghost 47

Thanks ill try that. Will there be any other face swap software for images released again. I used to use 1.5 reactor until it was closed and the flux one is very shiny and plastic. Is there any you recommend. Thanks for all your hard work

Brian Spiker

If ComfyUI desktop is already installed, does the auto batch file install in there or does it install in its own portable version?

John Lee

Got it, thanks. I’ve looked everywhere for a custom audio workflow for LTX 2.3 video extension but haven’t found one yet. Looking forward to your v2.

John Lee

It installs its own portable version in the same directory as the bat file.

Kris Miller

I can't get the installer to run. It says that the 7-Zip install has failed. I have tried installing 7-Zip manually, but can't get it to recognize it.

Kris Miller

I was able to get it to work. I had to download 7-Zip, and add it to path. Just to the directory not the application itself.

Peter J Walker

i wouldn't have the slightest idea how to add the slice node where how , not a clue. There is a reason I subscribe to get ready made workflows, because I cant do it.

Balledk1

I got the same problem, but local and on Runpod. Any got a solution?

new page

getting a LTXVSeparateAVLatent error. TypeError: 'NoneType' object is not subscriptable I tried the nodes installer first and got this error then tried doing the full fresh install but it's still giving me the error. Anyone have an idea of what I'm doing wrong??

gomez abel

I clocked in over 30 hours via ChatGPT with two open powershells to run commands. part of the struggle were corrupted bad models but once I got past that, the major part was the internal upscaler that the json keep calling for, so I had to edit the json and make a patch. I finally got it to generate and not crash, but that's the background story. ---> Hi Aitrepreneur, I’ve been working through your LTX ComfyUI workflow in depth and wanted to share what I’ve learned so far, along with a few targeted questions to better align with your intended design. First — really appreciate the work you’ve put into this. After stabilizing a few areas, the system’s potential is very clear. --- ## 🔍 What We’ve Confirmed ### 1. GGUF + Distilled LoRA Path (Correct Choice) I stayed on the Q4 GGUF model (ltx-2.3-22b-dev) with the distilled LoRA, and this combination produced more stable results on an 8GB setup — consistent with your notes. --- ### 2. Two-Stage Pipeline (Key Insight) From testing and reviewing your explanation: * **Stage 1**: Base generation (structure, motion, composition) * **Stage 2**: Refinement / upscale (adds detail, sometimes artifacts) Running Stage 1 only produced cleaner, more predictable results during debugging. --- ### 3. Sampler Outputs Matter SamplerCustomAdvanced exposes: * `output` → refinement / second-stage path * `denoised_output` → clean forward path Using `denoised_output` for the main latent flow significantly improved stability and removed unintended feedback behavior. --- ### 4. LoRA is Required for Structure Running without the distilled LoRA resulted in weak conditioning and poor prompt adherence. Once added, outputs improved immediately. --- ### 5. Latent Scaling Controls Performance Using reduced latent size (0.25–0.5) dramatically improved speed while still producing usable results — consistent with your examples. --- ## ⚠️ Observations During Stabilization ### 1. Latent Upscaler Behavior During testing, we temporarily bypassed the `LTXVLatentUpsampler` node because: * it appeared to introduce instability in the latent path * it made it harder to isolate Stage 1 behavior * it sometimes contributed to unintended feedback loops when connected upstream With the upscaler bypassed, Stage 1 generation became significantly more stable. 👉 Question: Is the latent upscaler intended to be used strictly in Stage 2 (refinement), rather than in the primary generation path? --- ### 2. Latent Flow / Concat Nodes There appear to be multiple `LTXVConcatAVLatent` nodes: * one involved in refinement/looping behavior * one used for final forward decode It wasn’t immediately obvious which one is intended as the canonical output path. --- ### 3. Feedback Loop vs Intended Design At one point, the graph effectively formed: Sampler → Separate → Crop → Concat → back into Sampler Breaking this loop improved: * shape stability * prompt adherence * runtime 👉 Question: Is any latent feedback loop intentional, or should the base pipeline be strictly forward-only? --- ### 4. Sampler Output Usage Should the main generation path always use: * `denoised_output` (Stage 1) and reserve: * `output` for Stage 2 processing only? --- ### 5. Low-VRAM Targeting Since GGUF + offloading is used: 👉 Is the workflow primarily optimized for: * 8–12GB VRAM systems (with tradeoffs) or * higher VRAM setups with fewer compromises? --- ## 🎯 Goal I’m aiming to: * preserve your architecture * stabilize the pipeline (no unintended loops) * clearly separate Stage 1 vs Stage 2 * improve predictability and iteration speed --- If you can clarify the intended latent path and where the upscaler fits into the pipeline, that would help tremendously. Thanks again — this is very close to a highly powerful and reliable workflow. Thank you in advance.

Anton Karpuzikov

Hey guys! I want to switch to RX 9070 XT are here any AMD users to provide some info on that?

THEO-Ed animation

I've got an NVIDIA GeForce RTX 4080 Super with apparently 16GB of dedicated GPU memory, which I take to be the VRAM portion of its capabilities. I'm starting to wonder if I'd be better off with RunPod after all…

Frédy Nativel

hello, does it possible to create a worklow using LTX 2.3 like WAN ANIMATE 2.2 ultra Workflow ?

Joel o' St. Pete

Thanks for the excellent workflow! Everything's working for me except for custom audio. I've tried with both text-to-video and image-to-video. I rewatched your video multiple times to make sure I'm not missing any settings, and I've got everything set the way you do in your video (including making sure that the spoken words are part of the prompt), but lip sync is simply not happening. I do hear the custom audio in the output video, but the character does not lip sync to it. I've tried with both MP3 and FLAC audio clips. Is there some trick to get lip-sync working?

Joel o' St. Pete

Nevermind because I just got a successful lip synced render. I made the prompt more explicit about the fact that this was a video of someone speaking and that seems to have made the difference. I also made sure that the video length and audio clip length were the same number of seconds. Thanks again for this fantastic workflow! Here's the relevant part of the prompt, in case anyone runs into the same issue: He is clearly and visibly speaking directly to camera as a single on-screen speaker, delivering the exact line aloud with natural mouth shapes, realistic jaw movement, and precise synchronized lip movement for every word. His lips, tongue, and facial muscles match the spoken dialogue in real time, with subtle head motion and natural conversational expression while he talks.

Aaron Aa

Hello Overlord, this dumb human subject of yours is still having problems getting your workflows to work on my new RTX 4090. I have tried several times running the 1 Click install bat and I keep on getting the same error that I sent in the previous post. This is confusing me because the 1 Click installers work on my older less capable computer and they do not work on my new computer. There must be something I have missed in this setup that is preventing these workflows from working.

James Stewart

Great vid. Saw this on Reddit, could it be built into the workflow? https://www.reddit.com/r/comfyui/comments/1s36ag4/generate_face_swaping_video_with_ltx_23_lora/?rdt=59804 With your LTX video the voice continued on using a 3 second clip. Is there another way a voice can be plugged in eg through TTS etc? When it comes to LoRAs I'm guessing its better to create them using Z-IMAGE TURBO (or similar), create an image and then use the LTX workflow for image to video?

Pak Chung

i got this error not sure what causes it : File "D:\AIExperiments\ComfyUI\resources\ComfyUI\execution.py", line 525, in execute output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\AIExperiments\ComfyUI\resources\ComfyUI\execution.py", line 334, in get_output_data return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\AIExperiments\ComfyUI\resources\ComfyUI\execution.py", line 308, in _async_map_node_over_list await process_inputs(input_dict, i) File "D:\AIExperiments\ComfyUI\resources\ComfyUI\execution.py", line 296, in process_inputs result = f(**inputs) ^^^^^^^^^^^ File "D:\AIExperiments\ComfyUI\resources\ComfyUI\comfy_api\internal\__init__.py", line 149, in wrapped_func return method(locked_class, **inputs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\AIExperiments\ComfyUI\resources\ComfyUI\comfy_api\latest\_io.py", line 1764, in EXECUTE_NORMALIZED to_return = cls.execute(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\AIExperiments\ComfyUI\resources\ComfyUI\comfy_extras\nodes_lt_audio.py", line 153, in execute audio_freq = audio_vae.latent_frequency_bins ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Timeless1175

so anyone who also encounters the error, "VAE is not correct," simply load the LTX2 VAE, which worked for me...gl hf

Ken L. Black

Even after doing a completely NEW installation of ComfyUI from scratch, I continue to receive the same error as I reported before: "TypeError: 'NoneType; object is not subscriptable. Once again, here is the entire error message: TypeError: 'NoneType' object is not subscriptable File "C:\ComfyUI_windows_portable\ComfyUI\execution.py", line 525, in execute output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\ComfyUI_windows_portable\ComfyUI\execution.py", line 334, in get_output_data return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\ComfyUI_windows_portable\ComfyUI\execution.py", line 308, in _async_map_node_over_list await process_inputs(input_dict, i) File "C:\ComfyUI_windows_portable\ComfyUI\execution.py", line 296, in process_inputs result = f(**inputs) File "C:\ComfyUI_windows_portable\ComfyUI\comfy_api\internal\__init__.py", line 149, in wrapped_func return method(locked_class, **inputs) File "C:\ComfyUI_windows_portable\ComfyUI\comfy_api\latest\_io.py", line 1764, in EXECUTE_NORMALIZED to_return = cls.execute(*args, **kwargs) File "C:\ComfyUI_windows_portable\ComfyUI\comfy_extras\nodes_lt.py", line 671, in execute latents = av_latent["samples"].unbind() ~~~~~~~~~^^^^^^^^^^^

Matthew Burgess

Hey man, I have been messing with this for a few days now. I'm running on Ubuntu, the ...MANAGER_AUTO_INSTALL.bat dose not run because it is a windows file. I tried importing the .json and installing everything independently, no dice. I gave $10 to RUNPOD and I have no idea what I am doing, your directions do not match what is on my screen. How dose one make this work on Linux (ubuntu). Thanks.

Adam Quinn

why is it saying it's missing models? "ltx-2.3-22b-dev-Q8_0.gguf (1)"

Snuff McLoud

Hello! I was able to start the workflow, but it seems some matrix mismatch? I was trying to troubleshoot with help of AI, what I got is that I should not use GGUF? RuntimeError: mat1 and mat2 shapes cannot be multiplied (1024x4096 and 32x4096) Full Error: RuntimeError: mat1 and mat2 shapes cannot be multiplied (1024x4096 and 32x4096) TIPS: If you have any "Load CLIP" or "*CLIP Loader" nodes in your workflow connected to this sampler node make sure the correct file(s) and type is selected. File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\execution.py", line 525, in execute output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\execution.py", line 334, in get_output_data return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\execution.py", line 308, in _async_map_node_over_list await process_inputs(input_dict, i) File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\execution.py", line 296, in process_inputs result = f(**inputs) File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\comfy_api\internal\__init__.py", line 149, in wrapped_func return method(locked_class, **inputs) File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\comfy_api\latest\_io.py", line 1764, in EXECUTE_NORMALIZED to_return = cls.execute(*args, **kwargs) File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\comfy_extras\nodes_custom_sampler.py", line 963, in execute samples = guider.sample(noise.generate_noise(latent), latent_image, sampler, sigmas, denoise_mask=noise_mask, callback=callback, disable_pbar=disable_pbar, seed=noise.seed) File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1051, in sample output = executor.execute(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes) File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 112, in execute return self.original(*args, **kwargs) ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 995, in outer_sample output = self.inner_sample(noise, latent_image, device, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes) File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 970, in inner_sample self.conds = process_conds(self.inner_model, noise, self.conds, device, latent_image, denoise_mask, seed, latent_shapes=latent_shapes) ~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 794, in process_conds conds[k] = encode_model_conds(model.extra_conds, conds[k], noise, device, k, latent_image=latent_image, denoise_mask=denoise_mask, seed=seed, latent_shapes=latent_shapes) ~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 704, in encode_model_conds out = model_function(**params) File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\comfy\model_base.py", line 1031, in extra_conds cross_attn = self.diffusion_model.preprocess_text_embeds(cross_attn.to(device=device, dtype=self.get_dtype_inference()), unprocessed=kwargs.get("unprocessed_ltxav_embeds", False)) File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\comfy\ldm\lightricks\av_model.py", line 578, in preprocess_text_embeds out_vid = self.video_embeddings_connector(context_vid)[0] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^ File "F:\ComfyUI portable\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "F:\ComfyUI portable\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl return forward_call(*args, **kwargs) File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\comfy\ldm\lightricks\embeddings_connector.py", line 297, in forward hidden_states = block( hidden_states, attention_mask=attention_mask, pe=freqs_cis ) File "F:\ComfyUI portable\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "F:\ComfyUI portable\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl return forward_call(*args, **kwargs) File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\comfy\ldm\lightricks\embeddings_connector.py", line 93, in forward attn_output = self.attn1(norm_hidden_states, mask=attention_mask, pe=pe) File "F:\ComfyUI portable\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "F:\ComfyUI portable\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl return forward_call(*args, **kwargs) File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\comfy\ldm\lightricks\model.py", line 420, in forward gate_logits = self.to_gate_logits(x) # (B, T, H) File "F:\ComfyUI portable\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1776, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "F:\ComfyUI portable\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1787, in _call_impl return forward_call(*args, **kwargs) File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\comfy\ops.py", line 392, in forward return self.forward_comfy_cast_weights(*args, **kwargs) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^ File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-GGUF\ops.py", line 217, in forward_comfy_cast_weights out = super().forward_comfy_cast_weights(input, *args, **kwargs) File "F:\ComfyUI portable\ComfyUI_windows_portable\ComfyUI\comfy\ops.py", line 385, in forward_comfy_cast_weights x = torch.nn.functional.linear(input, weight, bias)

Aitrepreneur

the frame slice node is already there, you just need to change the values, you can dm if you need

Aitrepreneur

yes, make sure to select either custom audio or no custom audio, if you leave them both enabled it will give you this error

Aitrepreneur

don't do that, you need the 2.3 vae not 2.0, if you got errors you just need to update kjnodes

Aitrepreneur

make sure to select either custom audio or no custom audio, if you leave them both enabled it will give you this error

Aitrepreneur

what is the issue for runpod? I don't use Linus but the runpod installer could be adapted to a local install I suppose

Aitrepreneur

probably because you are selecting the wrong model in the model selection area

Aitrepreneur

This either looks like a corrupted model or you have loaded an incorrect ltx 2.3 model. make sure ti's the 2.3 not the 2.0

Adam Quinn

I found out what the problem was I just had to tap on it even though it looked like it was already selected 🤪

Anton Karpuzikov

"--cache-none --reserve-vram 7" Why? for 16GB it cuts vram in half, why no cache?

AI Princess Artstation (edited)

Comment edits

2026-09-08 20:00 Hey :) THX for the workflow - it´s awesome. But it would be cool if there we´re options for the Sampler: CFG, Steps etc. - is that possible to provide? THX for your effort and your amazing work!
2026-03-29 10:20

Snuff McLoud

I have review all folders and what models need to be in those, based on pastebin manual guide you have provided. I removed what models are not uses, so it will not interfere. I have started 12GB VRAM bat file. Now it just complete Jobs in 0 second without any result.

Ken L. Black

Oh, my goodness....such a simple solution! I had both disabled. (Why can't error messages be in plain English?!?!) Thank you SO much, my AI Overlord. Best of health to you.

Snuff McLoud

I just did fresh install from scratch. Same issue as before. matrix multiplier do not match. I have tried T2V, I2V, same issue: RuntimeError: mat1 and mat2 shapes cannot be multiplied (1024x4096 and 32x4096) TIPS: If you have any "Load CLIP" or "*CLIP Loader" nodes in your workflow connected to this sampler node make sure the correct file(s) and type is selected. Fresh new installation of ComfyUI portable from provided LTX-2-3-ULTRA-COMFYUI-MANAGER_AUTO_INSTALL.bat I have applied the following solutions to fix the original workflow(in case I am not the only one): 1. Removed 19b - old LTX 2 (6 matrix) lora 2. Download new(fixed new version) of unet model: ltx-2.3-22b-dev-Q8_0.gguf 3. Removed custom node called ComfyUI-GGUF_Forked

hyfe solo

I get an error when trying to run the ComfyUI isntaller. it says its not compatible with 64bit windows? is there a version for 64 bit?

Aitrepreneur

Ltx models are known to have issues crashing comfyui, I found these parameters to be the most stable

Aitrepreneur

you can already change the sample, steps are done automatically with the manual sigmas node which are programed for best results and cfg isn't required since we are using the distill lora

Aitrepreneur

seems like you got the wrong custom node installed, if you did a fresh install though, everything should have been fixed automatically with the installer

Aitrepreneur

You need to download and install 7zip, the installer usually do it automatically if you don’t have it installed but for some people it doesn’t work well, so you can just do a simple manual install with this installer: (64-bit Windows x64) https://www.7-zip.org/download.html

Jacob

Hey, using your workflow and trying the video extend part and having problems. I’m using a run pod with a 4090, and it keeps getting frozen/stuck even if it’s the first generation attempt of the day on it. I’ve even tried downgrading my model to the Q5_K_M in the hopes it would work with less RAM needs, still freezes up. Am I doing something wrong or should I give up trying 4090’s?

Aitrepreneur

No it works fine, in my video I show you how to solve the vram issue on runpod by modifying the pod, in the post as well: "IF YOU GET OOM error : Click on the little 3 dots button, arguments variables then enter EXTRA_ARGS --reserve-vram 10 --cache-none""

Snuff McLoud

I did fresh install twice. Same result. With the following 3 steps, it solved the issue. Thank you anyway and thank you for the workflow!!

A Dante

This is really fun to play with - most animations using text to video are pretty twisted though, as in the bodies go into weird shapes and forms sometimes. :D I also noticed that this is not really for NSFW generations? It's mostly censored from the attempts that I've done - should I use a different model for this purpose or just a Lora?

hyfe solo

Hi, sorry im talking about the auto installer ; LTX-2-3-ULTRA-COMFYUI-MANAGER_AUTO_INSTALL.bat

Jacob

I thought that might have been it after I posted my question, and it did work of course. Once again, thank you.

ChudJones

Thanks for this. Is there any way to use the extender on a video without audio? I get a a NormalizeAudioLoudness error for any video that doesn't have an audio track, even if I upscale the video into the LTX format using the workflow.

Bob

can you please make video generation with wan and audio with ltx, workflow or where wan with audio, because ltx video generation just not as good as wan, even using first last frame sometimes identity blurs and just changes in some parts.

Kitty Doyle

I am currently running this on my linux laptop which has a mobile RTX 4060 with 8gb of vram and it works well. Additionally, I have integrated the Text to Video workflow into a ComfyUI skill I created for my OpenClaw and will soon be working on integrating the other four. Thank you for these work flows. They are amazing.

Dazzastrous

when i try the Extend Video workflow i get " Loop Detected 1546,1545" ?

michael morris

RuntimeError: shape '[384, 4096]' is invalid for input of size 366403 File "C:\AI Apps\Aitrepreneur\ComfyUI_windows_portable\ComfyUI\execution.py", line 525, in execute output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\AI Apps\Aitrepreneur\ComfyUI_windows_portable\ComfyUI\execution.py", line 334, in get_output_data return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\AI Apps\Aitrepreneur\ComfyUI_windows_portable\ComfyUI\execution.py", line 308, in _async_map_node_over_list await process_inputs(input_dict, i) File "C:\AI Apps\Aitrepreneur\ComfyUI_windows_portable\ComfyUI\execution.py", line 296, in process_inputs result = f(**inputs) File "C:\AI Apps\Aitrepreneur\ComfyUI_windows_portable\ComfyUI\custom_nodes\rgthree-comfy\py\power_lora_loader.py", line 56, in load_loras model, clip = LoraLoader().load_lora(model, clip, lora, strength_model, strength_clip) ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\AI Apps\Aitrepreneur\ComfyUI_windows_portable\ComfyUI\nodes.py", line 710, in load_lora lora = comfy.utils.load_torch_file(lora_path, safe_load=True) File "C:\AI Apps\Aitrepreneur\ComfyUI_windows_portable\ComfyUI\comfy\utils.py", line 149, in load_torch_file raise e File "C:\AI Apps\Aitrepreneur\ComfyUI_windows_portable\ComfyUI\comfy\utils.py", line 129, in load_torch_file sd, metadata = load_safetensors(ckpt) ~~~~~~~~~~~~~~~~^^^^^^ File "C:\AI Apps\Aitrepreneur\ComfyUI_windows_portable\ComfyUI\comfy\utils.py", line 110, in load_safetensors tensor = torch.frombuffer(mv[start:end], dtype=_TYPES[info["dtype"]]).view(info["shape"])

Midnight

I am reading online and there seems to be negative prompts in a lot of generated stuff for LTX 2.3? Is there a way to easily add that to these work flows?

ChudJones

I'm getting significant ghosting in high contrast movement (light character on dark background) with the upscaler and I can't track down what's causing it. I've adjusted both the detailer lora and the distilled lora and neither seem to be the issue. Any ideas?

Morbid

TypeError: 'NoneType' object is not subscriptable i am getting this

Venus

Hope this was not asked before. Getting this error: ValueError: cannot reshape array of size 6181148 into shape (8192,1408).

Aitrepreneur

Well yeah it's a 22B model, altough even the most powerful models right now can't get every body movement

Aitrepreneur

The goal of that particular workflow is audio+video cloning, if you don't have audio it's not gonna work well

Aitrepreneur

Depends what you mean, you can't make audio with ltx and have lipsync with wan 2.2 because wan2.2 wasn't trained on that. But you can make some animations with wan and add audio with ltx but the result is really not great tbh.

Aitrepreneur

The negative prompts are not that great and also for this workflow we are using the distilled lora so this will influence the negative prompts as well

Aitrepreneur

not sure what you mean, you can dm me and show me show examples

Aitrepreneur

make sure you select either custom audio OR no audio, if you leave both enabled it won't work and will give you this error

Aitrepreneur

this either looks like a corrupted model or you have selected an incorrect ltx 2.3 model

Midnight

Alright great thanks or answering that makes sense.

LouNokoAI

In RUNPOD after updating confyui I get and error, what should I do? Access to djuj32lxho9g9e-3000.proxy.runpod.net was denied

LouNokoAI

Logs say that update and restart was successful

Venus

Thank you for the response, started from scratch and it went through, appriacate the work.

Aitrepreneur

It's on runpod's side, they've had some issues the last few days, I had that issue as well, try again now

LouNokoAI

Still getting access denied I don't know what to do

Aitrepreneur

Yes, I found the issue, the problem seems to be the browser and/or some extensions you got. If you take the url and open it in a private window or in another browser, it will work. Not quite sure what is causing this

LouNokoAI

Okay I will try that now, you know in RUNPOD there are 2 options when you choose a gpu, network volume and volume disk which one should I use?

Ricky Joynes Jr.

THIS! Thanks for posting this, I was having problems no matter what I tried before creating a new folder.

Anthony

Hi, I already have comfyui installed. When I placed the LTX-2-3-MODELS-NODES_INSTALL.bat in the comfyui folder, it said "[ERROR] Could not find python at "C:\Users\Anthony\AppData\Local\Programs\ComfyUI\..\python_embeded\python.exe" Please run this script from: ...\ComfyUI_windows_portable\ComfyUI\". Is this works only for comfyui portable version?

Aitrepreneur

yes, all installers are for comfyui portable versions, I highly recommend the portable one, much easier to use, you can have multiple installs and you can move each installs independantly

Anthony

Hi, thanks for the inputs. But will AMD gpu of 16gb run ltx 2.3?

Aitrepreneur

Yes but you need the comfyui portable amd version, other than that it's good

Anthony

Thank you sir 👍🏻

Kynkie

If you may, is comfyui portable a different version of comfyui? or something I can install in app through comfui? Sorry! I'm also running into this problem but with an RTX 4050 laptop.

Anthony

Same software, one is desktop, another one is portable.

Michael Young

Heads up — the nodes install option for adding this to your existing 2.2 workflow is currently broken. When added to the previous install, it conflicts with the VAELoaderKJ node. I would suggest don't even provide these unless tested..

Aitrepreneur

Not sure what you mean, the VAELoaderKJ node is part of kjnodes, just having it updated to the latest version is enough

Michael Young

I tried that multiple times, even manually and it never worked. I simply installed from scratch again. Really wish these upgrade methods you listed would work though, but I guess it’s also Comfy being annoying with updates as well.

Aitrepreneur

Not sure it's my fault tbh, even I've had my fair share of "Comfyuifuckery" lately just trying to install simple nodes. Comfyui really became extremely unintuitive with newer updates breaking more and more stuff. Everyone in the community is complaining about it and they know it....

Matthew Burgess

No, I finally figured it out. It was a configuration problem on my side. In order to run ComfyUI on Linux with out issues you have to build and install into a Python container. Otherwise, things will break. The runpod installs it globally. I had to put the .json into ComfyUI and download all the nodes and models manually, and now... IT'S AMAZING! So much so that I now have a liquid cooled 5090 in the mail. It should be here Wednesday. This has changed everything. I have a list of ideas to monetize my work, and I don't now where to start! Any tips on how to make money with this? I have a Freelancer.com account. Thanks again. And thanks for replying, so many people set these things up and never do.

Dazzastrous

Managed to get it working Misouir

Adam

Hello, I tried to run on runpod using your shell script but I was stopped by the following problems: When loading the graph, the following node types were not found: Label (rgthree) SetNode INTConstant GetNode VHS_VideoCombine UnetLoaderGGUF LatentUpscaleModelLoader VAELoaderKJ Power Lora Loader (rgthree) 7d91fb0e-1bf7-4fb3-8eb8-8fd7d498d870 Fast Groups Muter (rgthree) Fast Groups Bypasser (rgthree) VHS_LoadVideo af5e88c2-c6bc-4370-ad3d-103c5d79aeb7 17762657-8014-4b35-b636-d574cadc4b8d 940d3192-3df1-4f4d-810e-f68de8a3cd72 8a523b6d-86b1-4332-8a56-2c625798825d 3f6cda67-43ea-4bab-bb53-cae754c4908e Nodes that have failed to load will show as red on the graph. I tried to update everything I could but some nodes have conflicts. How can I fix that?

Adam

I also see that updating Comfy Manager breaks the installation. I cannot reconnect after that update.

Aitrepreneur

is this locally or runpod? you can dm me for the rest

Myeisha Greenwood

torch.AcceleratorError: CUDA error: no kernel image is available for execution on the device keep getting this error. Need Help! Nvidia GTX 1080 16gb card

Adam

Runpod. I will try to reproduce and collect logs.

Timothy Danforth

I am having the same issue. As soon as I updated comfyUI and refreshed, the page wouldn't load and it gave a generic "you don't have permission to view this page" error.

todd beebe

I have two questions, nothing to do with the operation of it. I have it working fine. My first question is there a way to increase the steps to like 12 or even 20? Question two when I use the 3 frames part, first frame, middle, last when the video is ready and I play it it hesitates for a less then a second maybe where the middle frame was inserted. Its just like a tiny stutter. Is this normal and is there a way to fix this. Its not that bad but its not perfectly smooth like I would prefer.

Timothy Danforth

I am having an issue on RunPod where after restarting ComfyUI and then refreshing the page, I get one of two error pages: "You Don't Have Permission" error, or "This Site Can't be Reached" error. I did a little poking around and this is what I have managed to dig up. root@858ab6fa19ce:/workspace# grep -r "start_comfyui\|restart\|subprocess" /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/cm_global.py 2>/dev/null | head -20 root@858ab6fa19ce:/workspace# grep -rl "restart" /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/ 2>/dev/null /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/github-stats.json /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/extension-node-map.json /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/custom-node-list.json /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/README.md /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/__pycache__/prestartup_script.cpython-311.pyc /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/node_db/new/extension-node-map.json /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/node_db/new/custom-node-list.json /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/node_db/dev/custom-node-list.json /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/js/custom-nodes-manager.js /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/js/custom-nodes-manager.css /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/js/common.js /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/js/comfyui-manager.js /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/glob/manager_util.py /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/glob/manager_server.py /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/glob/manager_core.py /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/glob/__pycache__/manager_server.cpython-311.pyc /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/docs/en/use_aria2.md /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/.git/index /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/.git/objects/pack/pack-d52aafeb4fbc42503c1aa462bc8eb42d5055b3b2.pack /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/.git/objects/pack/pack-9f40e0eaa8f29432ed2e975a0c4b4438e4c0c9eb.pack /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/.cache/746607195_github-stats.json /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/.cache/1742899825_extension-node-map.json /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/.cache/1514988643_custom-node-list.json root@858ab6fa19ce:/workspace# grep -n "restart\|os.exec\|subprocess\|sys.exit" /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/glob/manager_server.py 5:import subprocess # don't remove this 162: process = subprocess.Popen(cmd, cwd=cwd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, bufsize=1, env=core.get_script_env()) 641: logging.info("\nAfter restarting ComfyUI, please refresh the browser.") 1367: logging.info("\nAfter restarting ComfyUI, please refresh the browser.") 1704:def restart(self): 1737: return os.execv(sys.executable, cmds) root@858ab6fa19ce:/workspace# sed -n '1700,1740p' /workspace/ComfyUI/custom_nodes/ComfyUI-Manager/glob/manager_server.py return web.json_response(alerts) @routes.get("/manager/reboot") def restart(self): if not is_allowed_security_level('middle'): logging.error(SECURITY_MESSAGE_MIDDLE_OR_BELOW) return security_403_response() try: sys.stdout.close_log() except Exception: pass if '__COMFY_CLI_SESSION__' in os.environ: with open(os.path.join(os.environ['__COMFY_CLI_SESSION__'] + '.reboot'), 'w'): pass print("\nRestarting...\n\n") # This printing should not be logging - that will be ugly exit(0) print("\nRestarting... [Legacy Mode]\n\n") # This printing should not be logging - that will be ugly sys_argv = sys.argv.copy() if '--windows-standalone-build' in sys_argv: sys_argv.remove('--windows-standalone-build') if sys_argv[0].endswith("__main__.py"): # this is a python module module_name = os.path.basename(os.path.dirname(sys_argv[0])) cmds = [sys.executable, '-m', module_name] + sys_argv[1:] elif sys.platform.startswith('win32'): cmds = ['"' + sys.executable + '"', '"' + sys_argv[0] + '"'] + sys_argv[1:] else: cmds = [sys.executable] + sys_argv print(f"Command: {cmds}", flush=True) return os.execv(sys.executable, cmds) @routes.post("/manager/component/save") root@858ab6fa19ce:/workspace# ps aux | grep python root 112 0.2 0.0 391532 88628 ? Sl 04:52 0:12 /usr/bin/python3 /usr/local/bin/jupyter-lab --allow-root --no-browser --port=8888 --ip=* --FileContentsManager.delete_to_trash= root 151 0.3 0.1 13762348 762968 ? SLl 04:52 0:14 python3 main.py --listen 0.0.0.0 --port 3001 root 2986 0.0 0.0 3852 2020 pts/4 S+ 06:08 0:00 grep --color=auto python root@858ab6fa19ce:/workspace#

Timothy Danforth

From what I can tell, port 8000 is "too slow" on initial startup and so when ComfyUI is refreshed manually, port 3000 throws the error. By opening port 8000 and manually stopping and starting ComfyUI, then opening port 3000 ComfyUI launches successfully.

Alex Iannelli

Running the complete install LTX-2-3-ULTRA-COMFYUI-MANAGER_AUTO_INSTALL. I placed the bat file on the drive I wanted it installed on, but it installed it into my C:\Windows\System32 folder. How do I get it to install on desired drive.? Can I just move the entire ComfyUI_windows_portable folder?

Kerfelox

Same issue, gave up using runpod it simply wont work even after recreating the network volume over and over again, after the installs it wont find the missing nodes or simply will not work after that. I tried restarting, refreshing, drag and dropping the workflow to new page.

Aitrepreneur

the problem seems to be the browser and/or some extensions you got. If you take the url and open it in a private window or in another browser, it will work. Not quite sure what is causing this

Aitrepreneur

the problem seems to be the browser and/or some extensions you got. If you take the url and open it in a private window or in another browser, it will work. Not quite sure what is causing this

Aitrepreneur

It's because you are running it as admin, run the installer normally

Alex Iannelli

Thanks. its running now, However, when I try to run the video extender, it returns a message 'Loop Detected 1537,1418'. I am using the ultra workflow via the download json file. I tried the no- audio workflow and got the exact same message.

tri sunarso

hello. i got those error yesterday. for the "access denied", the solution is unexpected : just add trailing slash "/" on browser. it works for me

tri sunarso

thank you very much. i use this in my PC and it works. the quality can be improved more by training and prompt engineering. my PC : windows 11, CPU i7 14700f , 64GB RAM, RTX 3060 12GB. currently testing just 640x480, trying the prompts.

tri sunarso

i just add trailing slash '/' and it works for me

Kerfelox

Thank you so much this solved my problem!

Showa

I have kind of a weird issue. Used to work, but after updating comfyUI I am now getting this error even after trying to uninstall everything and re-installing with just the auto install: Exception: An error occured in the ffmpeg subprocess: [aac @ 0000023511307740] Input contains (near) NaN/+-Inf [aost#0:1/aac @ 00000235113074c0] Error submitting audio frame to the encoder [aost#0:1/aac @ 00000235113074c0] Error encoding a frame: Invalid argument [aost#0:1/aac @ 00000235113074c0] Task finished with error code: -22 (Invalid argument) [aost#0:1/aac @ 00000235113074c0] Terminating thread with return code -22 (Invalid argument) File "D:\Work\Comfi\ComfyUI_windows_portable\ComfyUI\execution.py", line 525, in execute output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\Work\Comfi\ComfyUI_windows_portable\ComfyUI\execution.py", line 334, in get_output_data return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\Work\Comfi\ComfyUI_windows_portable\ComfyUI\execution.py", line 308, in _async_map_node_over_list await process_inputs(input_dict, i) File "D:\Work\Comfi\ComfyUI_windows_portable\ComfyUI\execution.py", line 296, in process_inputs result = f(**inputs) File "D:\Work\Comfi\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-VideoHelperSuite\videohelpersuite\nodes.py", line 609, in combine_video raise Exception("An error occured in the ffmpeg subprocess:\n" \ + e.stderr.decode(*ENCODE_ARGS))

Sploosh Live

I have custom audio off. But I have no sound with my videos. its generating the video okay but just no sound output. ive tried i2v and t2v

Son Goku

Does anyone else have issues with workflow data being saved in the video? usually I can just drag a video into comfy and the workflow is there. and yes, the save meta data is ticked. Any thoughts?

Aitrepreneur

it's because you get 2 files. 1 video with audio and 1 video without audio. Something like: LTX2_00106-audio and LTX2_00107 the audio one does not retain metadata, only the non audio one does

Brian Spiker

is this a one and done install or are there updates from time-to-time?

Brian Spiker

Well I just did another install of the latest and for the n-th time I still have no "run" button. I see in the YouTube video it should be at the bottom right but I just have the options for the resize, etc there. At the top right I have the "Manager" but again no run. What am I missing?

Brian Spiker

ok there was a pop-up that was covering the run bar. yep my bad. I need to click all the "X" to close out random errors. but When I still get run time errors. The sad part for the portable for me is I do not get red boxes around the offending module. This goes back to my question, how do I ensure that the install is complete and there are no updates? Just run it again? I wanted to upload a screen snap of what I see, but I don't see a means to upload.

Aitrepreneur

you can dm me and use something like imgur for the image. The updates are done through the update comfyui bat file located in the update folder and by using the manager button and clicking on update all (both methods needs to be used)

Kyle Williams

Quick question: I followed all steps for running LTX 2.3 for video generation and generated a few videos, but when I restart my pod (rented GPU from RunPod) and clicked on port 3000 again it won't work anymore... I'm very new to this so I'm pretty sure I'm missing a very basic step. Please help. The video I'm following is your video titled "INSANE FREE NSFW VIDEO AI KING! VIDEO EXTEND & VOICE CLONE!" on YouTube. By the way thinks for all tutorials.

Aitrepreneur

It's not your fault, it started happening for a few people as well, including me. The fix is very simple, just add a / at the end of the url and press enter, the UI will reappear

Dvox

what does: The workflow does not contain any output nodes (e.g. Save Image, Preview Image) to produce a result. mean? Total beginner here

Dvox

figured out the problem but now i got a new error: \Downloads\ComfyUI_windows_portable\ComfyUI\comfy_extras\nodes_lt.py", line 671, in execute latents = av_latent["samples"].unbind() ~~~~~~~~~^^^^^^^^^^^ TypeError: 'NoneType' object is not subscriptable

Aitrepreneur

make sure you have correctly selected in the option no custom audio or yes custom audio if you leave them both selected or unselected it won't work

michael delgrosso

got the workflow working but everything is super blurry with the settings you have. im so confused. messing with settings right now.

The Alchemist

Hey everyone, how's it going? I have a question: is there a setting I can change to make the video 10, 15, or 20 seconds long? Right now it's only generating 5 seconds.

Aitrepreneur

yes, it's in the parameters, open your eyes man xD

lurkergao

I'm trying to run this on runpod, I followed the directions in the video to use the comfyui aientrepreneur template but the pod keeps provisioning endlessly. how long is that supposed to take? It's been about a day now of it sucking up credits. system log just repeats this endlessly: create container aitrepreneur/comfyui:2.3.5 create container: still fetching image aitrepreneur/comfyui:2.3.5

Aitrepreneur

That is not normal att all just delete that pod and try again with a new one 10 mins max is what it should take to start

Patrick Smith

Hi, I can't seem to track down the LTX-2-3_ULTRA_WORKFLOW.json file. Any ideas?

Leo Werndly

For anyone getting this bug recently - I had to downgrade KJNodes to version 1.39 to get it working for me

ZenNate

Hi, I'm using Runpod and seemingly got everything installed correctly, but when I add my custom audio, select custom audio, enter my prompt and select run, I get this messaging: Return type mismatch between linked nodes images, received_type(VAE) mismatch input_type(IMAGE)

ZenNate

I didn't adjust anything to the premade work flow.

Ryan Joel

Is there any way to use this on an AMD GPU?

Aitrepreneur

Yes you just need to dl the comfyui portable amd version and use the models nodes installer to install the models and nodes you are missing

Ryan Joel

You are a gentleman and a scholar

ZenNate

Is anyone else who is using Runpod able to get this to work? No matter what I try, I can't get this to work. I'm new to Runpod and would greatly appreciate any assistance.

Scofiled Yeh

I got VAE error as below, please help RuntimeError: ERROR: VAE is invalid: None If the VAE is from a checkpoint loader node your checkpoint does not contain a valid VAE. File "D:\AI\ComfyUI_windows_portable\ComfyUI\execution.py", line 525, in execute output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\AI\ComfyUI_windows_portable\ComfyUI\execution.py", line 334, in get_output_data return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\AI\ComfyUI_windows_portable\ComfyUI\execution.py", line 308, in _async_map_node_over_list await process_inputs(input_dict, i) File "D:\AI\ComfyUI_windows_portable\ComfyUI\execution.py", line 296, in process_inputs result = f(**inputs) File "D:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-KJNodes\nodes\nodes.py", line 2426, in load_vae vae.throw_exception_if_invalid() ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^ File "D:\AI\ComfyUI_windows_portable\ComfyUI\comfy\sd.py", line 847, in throw_exception_if_invalid raise RuntimeError("ERROR: VAE is invalid: None\n\nIf the VAE is from a checkpoint loader node your checkpoint does not contain a valid VAE.")

Aitrepreneur

I think you are just using an older comfyui version + non updated comfyui nodes. To solve this, first go inside the ComfyUI_windows_portable\update folder and run the update_comfyu.bat file. Then launch comfyui, click the manager button and click on “update all”. Then once everything is updated click restart then try again

CemeteryPumpkin

Have you looked into LTX2.3 Video Director? Theres a youtube video out there about it and its wild. I would love a workflow by you incorporating ggufs. The tool is awesome, but it seems it only works on the reg models, for now.

Aitrepreneur

Don't worry, the updated video workflow should be coming this weekend ;) (probably more on Sunday than Saturday)

Matt

I was going to install the last wan again after having failed at it earlier on, but then this came out and I guess I'm installing it!

Antes

How do I start it in RunPod?

Psokratease

Hello! An error after install (the comfyui with ltx) I get "runtime error vae is invalid none

Psokratease

apologies! I did use the kjnode batch file but had forgotten what you always say after install, "update your comfyui!". So, albeit slow on my 3090, it is running.

Oscar Wolf

I spent hours and around 10$ in runpod trying to get this running, it didn't worked. Installed comfy ui, updated comfy ui, installed the "auto install for runpod", dragged the workflow and there was many missing nodes, installed missing nodes except one. The models were downloaded and stored somewhere but not in the model folders, so I had to re install them one by one using wget. I got an out of memory error and I gave up. I think something got outdated or broken, because this isn't working anymore.

Aitrepreneur

I can tell you it still works, I use it fairly often myself when i don't want to do it locally, watch my video i even explain at the end what to do to avoid the out of memory error

8SteelBeans

I'm also looking for the answer to whether a new model, or lora, is needed to achieve nsfw results. Have you figured anything out?

Reijer den Hartog

Did run git pull but got the message already up to date but still not working.

Jerald

Does this work if i am using RunPod? I see you said their is a fix on Windows machines? I use wan 2.1 right now for img to video /=.

Aitrepreneur

Yes and yes. Try ltx 2.3, much better than wan 2.1....

Moro Moro

running into dependency issues when trying to install on rinstances with a 5090. so I can't complete the install...

Frog Prince

Is there any chance you'd make a workflow for LTX specifically with Director? I can't see anything out there better than that Director node... Wish there was a Director node for Krea 2 Love your workflows, so tidy :) thanks!