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

1-Click INSTALL WAN 2.2 ULTRA VIDEO & IMAGE KING ComfyUI WebUI! (Patreon)

Published:
2025-08-04 21:39:11
Edited:
2026-09-03 01:02:40
Imported:
Tags:

Downloads

Content

EDIT 3 September:NEW RUNPOD CUDA 13 TEMPLATE!

I’ve spend the last few days, completely rebuilding my RunPod ComfyUI template from scratch.... The previous template was getting very outdated and was starting to cause problems with newer GPUs, ComfyUI versions and recent workflows, so please use this new template from now on.

The new template uses CUDA 13, supports newer GPUs like the RTX 50 series, comes with SageAttention preinstalled, automatically keeps ComfyUI updated and works with my new RunPod installers.

HOW TO USE THE NEW TEMPLATE

I made a quick video showing exactly how to deploy it and use the installers:

👉 https://youtu.be/9wlUXXyy9mE

👉 RUNPOD TEMPLATE LINK

IMPORTANT: When choosing your GPU, use the CUDA 13.x filter on RunPod

The old RunPod template is now completely retired, so use this one for all new installs going forward.To use the installer use this command:bash "WAN2_2-ULTRA-AUTO_INSTALL-RUNPOD-NEW.sh"I WILL SPEND THE NEXT FEW DAYS UPDATING ALL PREVIOUS RUNPOD INSTALLERS TO THE NEW TEMPLATE SO IF THERE IS AN INSTALLER THAT I FORGOT LET ME KNOW.Hey everyone! I've created a 1-click installer for the NEW WAN 2.2 MODELS! Simply the best Text to Video, Image to Video AND Text to image m

Hey everyone! I've created a 1-click installer for the NEW WAN 2.2 MODELS! Simply the best Text to Video, Image to Video AND Text to image models ever released to date. Better motion, better style, better quality, just better in every single way! Capable of creating lifelike videos and images with incredible motion in only a few minutes. The age of movie creation at home as started, jump on the hype train right now while you still can!There is just no competition for Wan right now....it wins on everything...

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

The workflow: https://www.patreon.com/posts/135728889

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

LOCAL INSTALL!!! RECOMMENDED TO START FROM SCRATCH SINCE YOU NEED THE LATEST COMFYUI TORCH VERSION!!!

Download the WAN2_2-ULTRA-COMFYUI-MANAGER_AUTO_INSTALL.bat

Run the bat file and select either option 1, 2 or 3 depending on your GPU vram and then determine if you want the Original Wan model (recommended) .

Wait for the install to finish

The webui will launch, load the WAN2_2-ULTRA_WORKFLOW.json file and there you go😎

IF YOU ALREADY HAVE COMFYUI INSTALLED USING A PREVIOUS INSTALLER:

Download the WAN2_2-ULTRA-MODELS-NODES_INSTALL.bat and place it in the ComfyUI_windows_portable\ComfyUI folder

Run the bat file and select either option 1, 2 or 3 depending on your GPU vram and then determine if you want the Original Wan model (recommended) .

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

When selecting your GPU on RunPod, make sure you click on Additional Filters → CUDA Versions → CUDA 13.0Then choose a GPU, click here for the comfyui template (aitrepreneur/comfyui:3.0.1-cuda13)

Then click deploy, wait for everything to finish and then click on port 8888 and then drag and drop the WAN2_2-ULTRA-AUTO_INSTALL-RUNPOD-NEW.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 this line then press enter:

bash "WAN2_2-ULTRA-AUTO_INSTALL-RUNPOD-NEW.sh"

Wait for everything to be installed

Then go to the port 8188

Load the WAN2_2-ULTRA_WORKFLOW.json workflow and there you go😎

Also use the command:

tail -f comfyui.log

inside the workspace/logs to get the real time logs

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

TrentD

Hi. I tried doing img2vid in the provided workflow, and ComfyUI gave this message when trying to run it: KSamplerAdvanced Given groups=1, weight of size [5120, 36, 1, 2, 2], expected input[1, 32, 31, 90, 90] to have 36 channels, but got 32 channels instead So this is Greek to me. Any ideas?

julian brown

Not sure if im missing something obvious but i get an error message saying it is not compatable with a 64bit version of windows when i try to run the installer that includes comfy ui. any help would be appreciated.

Bassplasma

Hi ! i got this error , followed everything , updated comfy too , any chance to get the source of the problem ? Given groups=1, weight of size [5120, 36, 1, 2, 2], expected input[1, 32, 21, 90, 160] to have 36 channels, but got 32 channels instead

julian brown

Thanks for the help and quick response

Aitrepreneur

What comfyui version you see in the manager? It's better to use the update comfyui bat file inside the update folder

FILM426

Hey! so excited for this been waiting for your awesome work! ive got it installed however when testing T2I Im getting this error KSamplerAdvanced 'NoneType' object has no attribute 'to'

Aitrepreneur

use the update comfyui bat file inside the update folder instead of the manager

Aitrepreneur

Make sure comfyui is updated and try again, for some reason I remember having this error as well, simply restarting comfyui after an update worked

Dauron Robinson

I've updated everything and still getting the error

FILM426

unfortunately an update and refresh/relaunch didnt work, im on V3.35

Hemdroid

hi could use a little help. Got and error in the Ksampler Advanced "KSamplerAdvanced 'NoneType' object has no attribute 'to'

Aitrepreneur

No that's the manager version, for the comfyui version you need to scroll down on the right. Also are you using the magcache nodes? Because as i said you shouldn't use it until it gets updated for wan 2.2

Aitrepreneur

DM me screenshots of the comfyui version and parameters before the generation, also apparently if you already have a previous comfyui install go inside the custom nodes folder and look for the flow2-wan-video folder if you have it, if you do, delete it

FILM426

Ooh that could be it. I didn't remember to disable them. I'll give that a try. Thank you

Aitrepreneur

Update comfyui and don't use the mag cache nodes as i explained in the video until it's updated for wan 2.2

Dauron Robinson

Its finally working, i had to run the update bat 3 times

Aitrepreneur

Yeah...unfortunately that happens, sometimes correctly updating comfyui is hard for no reason..

Goblin

I'm having to look in the outputs folder to get an output, and even then, it looks mosaic

Aitrepreneur

yeah It's my bad, I just forgot to show the node for the video, just right click on the final node and click "show preview". For the mosaic that is weird and you might be using magcache maybe so don't

Goblin

I have that turned off, but I'll keep doing research. Thank you Aitrepreneur! big fan!

Darkhitman66

Allocation on device This error means you ran out of memory on your GPU. TIPS: If the workflow worked before you might have accidentally set the batch_size to a large number.

Aitrepreneur

could you stop spamming here and send me a dm instead

gregoryallen caldwell

i have plenty of vram and 32 gig ram and huge swap file, but i have AMD no gpu! will it work?

gregoryallen caldwell

i have SD 1.5 working good on same machine , will it even install without a GPU to try it ? ryzen 7 5700, 41 gigs vram available out of 58 total,page file 27 gigs..radeon graphics, can i optimize for cpu use ?

Leigh Woods

I'm trying to find the WAN2.2 Runpod Template, did you forget to include the link or is the install-runpod.sh the file that needs to bu uploaded to runpod?

Aitrepreneur

same answer as down below, I can't say for certain

Aitrepreneur

the template is called "aitrepreneur comfyui", the installer is the WAN2_2-ULTRA-AUTO_INSTALL-RUNPOD.sh and the workflow is the WAN2_2-ULTRA_WORKFLOW.json everything is uploaded

RB

Nice!

Goblin

Replying for anyone else who might be having this issue. All I did was download a checkpoint. Don't be me - read instructions

Ron Harris-White II

So if I want to do use FLUX Kontext, Nunchaku and WAN 2.2, what should be the method of installation? Install WAN 2.2 and then install MODELS-NODES bat files for Nunchaku and Kontext?

Russ Ader

First, I am so excited to try this. Second, is there any way to incorporate a looping feature into the workflow to make longer videos like this one... https://civitai.com/models/1836348/wan-22-i2v-simple-multiple-promptvideo-loop?modelVersionId=2078089&fbclid=IwY2xjawL-PlpleHRuA2FlbQIxMQABHmiVHhexnvLdB1jvuIL-z0otKzrIH-JU-Ak7RwKY6u4x-__CV8y2S5NNOXS7_aem_0cdaz84Iz1Emy9xz_3tW5g I used it and it worked wonderfully. I made a 20 second video just fine. Figured you could easily do a better one than it to make longer videos. I have always liked your workflows better and the ability to open a single workflow vs multiples. You also use better nodes. And third, allot of people are using sage attention. What ever that is. Does it help and is it included? If not, is there any chance you will include it in the next update?

Onyx N

is it impossible to generate longer video?

Pete

Anyone getting this error? I keep updating but still get this error: Prompt execution failed Prompt outputs failed validation: KSamplerAdvanced: - Value not in list: sampler_name: 'res_2s' not in (list of length 40) - Value not in list: scheduler: 'bong_tangent' not in ['simple', 'sgm_uniform', 'karras', 'exponential', 'ddim_uniform', 'beta', 'normal', 'linear_quadratic', 'kl_optimal'] KSamplerAdvanced: - Value not in list: sampler_name: 'res_2s' not in (list of length 40) - Value not in list: scheduler: 'bong_tangent' not in ['simple', 'sgm_uniform', 'karras', 'exponential', 'ddim_uniform', 'beta', 'normal', 'linear_quadratic', 'kl_optimal']

Aitrepreneur

Everything is possible...you just need to...Believe!... Yeah no I can do it, might make an update or something once we get more stuff like the image to video loras or something

Aitrepreneur

Or use a separate special workflow for.long videos

Aitrepreneur

Have you used the installer? You need to have the RES4LYF node installed

Onyx N

is it possible if i use lower model of gguf and lower frame rate?

Pete

using "WAN2_2-ULTRA-MODELS-NODES_INSTALL" breaks my comfui so careful anyone using that

Aitrepreneur

Absolutely be careful, all the installers i make break all comfyui installs...........OR......MAYBE....JJUST MAYBE.......you should read the description. Just a hint y'a know

Paul Mazurkiewicz

Thanks again for another great installer!! Question, when choosing 1-3 at the start, my GPU has 24 gb ram. So do I choose 12-24 or 24+? Both say 24 in them.

Pete

Using the "WAN2_2-ULTRA-MODELS-NODES_INSTALL" broke my ComfyUI, so I restored it from a backup. I then copied everything from the broken ComfyUI’s models and custom_nodes folders. Now it’s working. I'm not sure what else installed that broke it so maybe it just corrupted when downloading or something. Got to love comfyui.

Pete

Also yeah fair enough <3

Russ Ader

Whoa! Things are moving fast. They just figured out how to do a first last frame generation with Wan 2.2 If you can incorporate that and the looping into your workflow, you will have created a whole video studio suite that surpasses Framepack studio. Using your on off toggles you can generate your image, then produce a 10 second video looping style, then take the last frame of the looped video and insert it into the first last frame video section, generate another image as the last frame and run the first last frame generation to create an awesome transition. Take the looped video and the transition video and splice them together then input that into your interpolation and upscalers. Oh, there is also the ability to upscale using the 5b model. All in one workflow. You are the only one I can see knowing or willing to do this.

Pizza

I did the Comfyui install and the Image2image is working, but the Text2image is giving me this error. I didn't make any changes: KSamplerAdvanced 'NoneType' object has no attribute 'to'

Pizza

Full error: # ComfyUI Error Report ## Error Details - **Node ID:** 66 - **Node Type:** KSamplerAdvanced - **Exception Type:** AttributeError - **Exception Message:** 'NoneType' object has no attribute 'to' ## Stack Trace ``` File "C:\ComfyUI2 All inOne\ComfyUI_windows_portable\ComfyUI\execution.py", line 496, 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, hidden_inputs=hidden_inputs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\ComfyUI2 All inOne\ComfyUI_windows_portable\ComfyUI\execution.py", line 315, 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, hidden_inputs=hidden_inputs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "C:\ComfyUI2 All inOne\ComfyUI_windows_portable\ComfyUI\execution.py", line 289, in _async_map_node_over_list await process_inputs(input_dict, i) File "C:\ComfyUI2 All inOne\ComfyUI_windows_portable\ComfyUI\execution.py", line 277, in process_inputs result = f(**inputs) ^^^^^^^^^^^

aroke

any idea what this error means? 'NoneType' object has no attribute 'to'

BD

having this issue when using the runpod install script

TrentD

That worked, it's running and saving videos to the hard drive. Are they supposed to be displayed in the Video Combine node when they are done? Cause they don't here, just seeing them saved to disk.

John Horton

disable MagCache to make it work, problem is there

kaudlol

Try turning off Mag Cache

Serguei Nevarko

I'm getting the same issue on local PC. I tried running it on GPU and CPU and had the same error (fresh install)

kaudlol

Hey @Aitrepreneur, any way I can resolve this on my own for the video preview to show once the I2V fully completes? Instead of right clicking and using the "Show Preview" every time a generation is done.

KaraMoot

Do the runpod directions not work with a 5090? I chose your template and followed the instructions but got the error: "CUDA error: no kernel image is available for execution on the device CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1 Compile with TORCH_USE_CUDA_DSA to enable device-side assertions."

kaudlol

Tried Runpod with the 5090 and ran into this issue as well. Switched to the 4090 and worked fine, so not exactly sure what the issue is, but might have to use the 4090 for the mean time.

Zoran Kanti-Paul

Getting this error mate - Given groups=1, weight of size [5120, 36, 1, 2, 2], expected input[1, 32, 21, 90, 160] to have 36 channels, but got 32 channels instead

Aim Corp

sounds like you got like sdxl mixed with flux somehow, some model you got selected weather it be a .ae or clip, something is selected wrong

Pizza

How do you do that exactly? I toggled off the loras and same error.

Pizza

I disabled MagCache one at a time and both and get same error. Are you supposed to be using both the high and lownoise? So confused, I am new to Comfyui

Robert

followed the instructions for runpod. also chose the recommended 4090. got the following error after installing all missing nodes from the workflow: After restarting ComfyUI, please refresh the browser. Restarting... [Legacy Mode] Command: ['/workspace/ComfyUI/venv/bin/python3', 'main.py', '--listen', '0.0.0.0', '--port', '3001'] Using Python 3.11.9 environment at: venv [START] Security scan [DONE] Security scan ## ComfyUI-Manager: installing dependencies done. ** ComfyUI startup time: 2025-08-05 08:23:48.626 ** Platform: Linux ** Python version: 3.11.9 (main, Apr 6 2024, 17:59:24) [GCC 11.4.0] ** Python executable: /workspace/ComfyUI/venv/bin/python3 ** ComfyUI Path: /workspace/ComfyUI ** ComfyUI Base Folder Path: /workspace/ComfyUI ** User directory: /workspace/ComfyUI/user ** ComfyUI-Manager config path: /workspace/ComfyUI/user/default/ComfyUI-Manager/config.ini ** Log path: /workspace/ComfyUI/user/comfyui.log Using Python 3.11.9 environment at: venv Using Python 3.11.9 environment at: venv [ComfyUI-Manager] PyTorch is not installed Prestartup times for custom nodes: 0.0 seconds: /workspace/ComfyUI/custom_nodes/rgthree-comfy 0.0 seconds: /workspace/ComfyUI/custom_nodes/ComfyUI-Easy-Use 0.0 seconds: /workspace/ComfyUI/custom_nodes/comfyui-easy-use 2.7 seconds: /workspace/ComfyUI/custom_nodes/ComfyUI-Manager Checkpoint files will always be loaded safely. Total VRAM 24210 MB, total RAM 515790 MB pytorch version: 2.7.1+cu126 WARNING[XFORMERS]: xFormers can't load C++/CUDA extensions. xFormers was built for: PyTorch 2.4.0+cu121 with CUDA 1201 (you have 2.7.1+cu126) Python 3.11.9 (you have 3.11.9) Please reinstall xformers (see https://github.com/facebookresearch/xformers#installing-xformers) Memory-efficient attention, SwiGLU, sparse and more won't be available. Set XFORMERS_MORE_DETAILS=1 for more details Set vram state to: NORMAL_VRAM Device: cuda:0 NVIDIA GeForce RTX 4090 : cudaMallocAsync Using pytorch attention torchaudio missing, ACE model will be broken torchaudio missing, ACE model will be broken Traceback (most recent call last): File "/workspace/ComfyUI/main.py", line 147, in import execution File "/workspace/ComfyUI/execution.py", line 16, in import nodes File "/workspace/ComfyUI/nodes.py", line 32, in from comfy_api.version_list import supported_versions File "/workspace/ComfyUI/comfy_api/version_list.py", line 1, in from comfy_api.latest import ComfyAPI_latest File "/workspace/ComfyUI/comfy_api/latest/__init__.py", line 12, in from comfy_api.latest._ui import _UI as ui #noqa: F401 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/workspace/ComfyUI/comfy_api/latest/_ui.py", line 12, in import torchaudio File "/usr/local/lib/python3.11/dist-packages/torchaudio/__init__.py", line 2, in from . import _extension # noqa # usort: skip ^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/torchaudio/_extension/__init__.py", line 38, in _load_lib("libtorchaudio") File "/usr/local/lib/python3.11/dist-packages/torchaudio/_extension/utils.py", line 60, in _load_lib torch.ops.load_library(path) File "/workspace/ComfyUI/venv/lib/python3.11/site-packages/torch/_ops.py", line 1392, in load_library ctypes.CDLL(path) File "/usr/lib/python3.11/ctypes/__init__.py", line 376, in __init__ self._handle = _dlopen(self._name, mode) ^^^^^^^^^^^^^^^^^^^^^^^^^ OSError: /usr/local/lib/python3.11/dist-packages/torchaudio/lib/libtorchaudio.so: undefined symbol: _ZNK5torch8autograd4Node4nameEv

Xyzzx

i am the only one getting completely red nodes after building this on runpod??? Not urgent, but I'll try again tomorrow

Zoran Kanti-Paul

No that is not it. I looked and there is nothing like that. ANy other ideas? So far I have not had one complete generation of an image to video!!

Remowylliams .

I used your runpod install in my ComfyUI last updated with the wan2.1 workflow. Updated ComfyUI with update all and am getting a node error for the Ksampler (Advanced) saying: 'NoneType' object has no attribute 'to' previous problem said 'no bong_tangent' I've tried updating and fully restarting ComfyUI. with no luck. Any ideas?

Mike tester

try disabling the TXT2IMG MAG option, it worked for me

Brandon Waters

i'm doing the wan 2.2 text to image workflow, and i'm getting double bodies when i try to do portrait 1920 x 1088. i'm using fusionx at 1.30 for high and low, then wan 2.2 lightning 4 step at 1.00. Any ideas?

fjui

Thank you, this worked for me

Adam

Same fault here: KSamplerAdvanced Given groups=1, weight of size [5120, 36, 1, 2, 2], expected input[1, 32, 21, 90, 160] to have 36 channels, but got 32 channels instead I don't know enough about this stuff to diagnose, but I'm using all the models and VAEs that were installed via the auto installer.

Inbound_Titan

I2V not working. It's getting hung up on the clip. CLIPLoaderGGUF cannot reshape array of size 15547983 into shape (4096,7040) Edit: So I just deleted the original install and reinstalled everything and it's working. No clue what happened. I did download through vpn originally so it may have lost some files in translation.

Zoran Kanti-Paul

I did a complete 'update all' from comfyui manager and got it to work.. Try that

Adam

I ended up updating ComfyUI via the .bat in the update folder. Then the error changed to: AttributeError: 'NoneType' object has no attribute 'to' Then I turned off MAG and it worked.

Walker4k

Same error on runpod 5090

KaraMoot

I think you're right! I think it is an issue with the CUDA/Torch version on the template. There are other runpod templates for ComfyUI that are updated for 5090 that work better, and I think the same installation process otherwise.

Zoran Kanti-Paul

yeah apparently MAG is not updated for 2.2 yet

LW

Got the txt2image and img2image working fine and oh my, it's fast and detailed. This tech is developing fast, I'm glad you're here, K, to keep us up to date. You will let us know if Skynet is about to nuke us all, right? When I tried img2vid, I got this error: UnetLoaderGGUF cannot reshape array of size 418344 into shape (5120,3520) I did a complete fresh install using WAN2_2-ULTRA-COMFYUI-MANAGER_AUTO_INSTALL.bat I am using the Q5_K_S models and I have img2vid Mag switched off. I have tried reducing the size of the actual input image as low as 480 x 480 and tried reducing the video size output as low as 128 x 128 x 48 frames. I have a 3090 with 24gb vram I saw that Inbound_Titan did a complete re-install that fixed a similar issue. Do I need to reinstall or is there a quicker fix?

Aitrepreneur

the first last frame gen is already in the workflow, for the rest wait and see

Aitrepreneur

there is always a limit, wan was trained on 81 frames at 16fps so it makes 5s videos, everything above that in a single normal gen not only takes a lot of vram but also decreases the coherence of the video but I heard that you can go up to 10s with the wan 2.2 with pretty good results

Aitrepreneur

Watch the video man I explain everything, with 24 you should probably choose option 1 or 2, it's up to you, quality will be pretty much the same really, very little difference

Aitrepreneur

Redownload the new workflow, I removed magcache for now since so many people had issues with it

Aitrepreneur

Redownload the new workflow, I removed magcache for now since so many people had issues with it

Aitrepreneur

Yup always prefer using the update comfyui bat file from the update folder. For the magcache issue, you can redownload the new workflow, I removed magcache for now since so many people had issues with it, I'll reupdate it once it's compatible with wan 2.2

Aitrepreneur

Things have changed, you need to update comfyui first then use the installer, watch the video or read the instructions in the post

Aitrepreneur

Things have changed, you need to update comfyui first then use the installer, watch the video or read the instructions in the post

Aitrepreneur

1. The no bong tangent, you need to use the installer or manually install the https://github.com/ClownsharkBatwing/RES4LYF node 2.'NoneType' object has no attribute 'to' Redownload the new workflow, I removed magcache for now since so many people had issues with it

Aitrepreneur

yeah it's a common issue with Ai models even back to the first stable diffusions, depending on the prompts and resolution, it might make double bodies or elongated bodies to fit the image. You can try using some negative prompts to negate the effect but most of the time, changing the aspect ratio or resolution is the key

Aitrepreneur

We have some "pseudo vace" available but it's better to wait until we get the real model , the real deal

Aitrepreneur

Yes I'll let you know if Skynet comes for us all, for now we can just play with creating images at home lol. For the error, are you using the correct image to video model? and not the text to video or something? Screenshots always help. Make sure that you are rocking the latest comfyui version by using the update comfyui bat file inside the update folder

Aitrepreneur

You can redownload the workflow, i should have fixed it now

Pizza

Thanks!

Robert

i've read your instructions and tried several times. the solution in the end was.. 1st. installer THEN update comfy xD...

Klaus The.

Same for the newer RTX Pro 6000 at RunPod. "CUDA error: no kernel image is available for execution on the device"

Klaus The.

it seems to be the PyTorch version: Checkpoint files will always be loaded safely. /usr/local/lib/python3.11/dist-packages/torch/cuda/__init__.py:230: UserWarning: NVIDIA RTX PRO 6000 Blackwell Workstation Edition with CUDA capability sm_120 is not compatible with the current PyTorch installation. The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_70 sm_75 sm_80 sm_86 sm_90. If you want to use the NVIDIA RTX PRO 6000 Blackwell Workstation Edition GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

Aitrepreneur

Yeah unfortunately for now my template doesn't work with blackwell gpu....we can thank nvidia for that....

Bob Bobenko

Used your installer for runpod. Getting this error...

Bob Bobenko

Used your installer for runpod. Getting this error after running: Value not in list: scheduler: 'bong_tangent' not in ['simple', 'sgm_uniform', 'karras', 'exponential', 'ddim_uniform', 'beta', 'normal', 'linear_quadratic', 'kl_optimal'] - Value not in list: sampler_name: 'res_2s' not in (list of length 40) I updated everything first, then run your instalation script. Getting this error.

Aitrepreneur

that is weird, I can't reproduce the issue. The error is from a missing node. Go into the manger, click node manager and search for RES4LYF then install it, and then click restart

Bob Bobenko

Solved: In JupyterLab activating venv, then cd to custom_nodes/RES4LYF, and then pip install -r requirements.txt

Klaus The.

Go to manager and click restart. after that reload the comfyui page. Then close the workflow and dragndrop the .json workfile again.

Matti Mustajärvi

Would there be any benefit to using sageAttention?

Klaus The.

In the two MagCache Nodes, select the correct nodetype for wan2.1 I2V 720p - the wrong T2V are preselected. That fixed it for me.

kaudlol

Thanks brother, much appreciated ^^

Devendra Singh

Same error I2V not working. It's getting hung up on the clip. CLIPLoaderGGUF cannot reshape array of size 15547983 into shape (4096,7040)

joaquin me llamo

curl: (23) client returned ERROR on write of 866 bytes Download failed.

Pizza

Unfortunately I get the same errors even after installing the updated fresh one. I don't think this workflow works at all. After fresh install I get errors: Prompt outputs failed validation: UnetLoaderGGUF: - Value not in list: unet_name: 'wan2.2_t2v_low_noise_14B_Q8_0.gguf' not in ['Wan2.2-I2V-A14B-HighNoise-Q4_K_S.gguf', 'Wan2.2-I2V-A14B-LowNoise-Q4_K_S.gguf', 'Wan2.2-T2V-A14B-HighNoise-Q4_K_S.gguf', 'Wan2.2-T2V-A14B-LowNoise-Q4_K_S.gguf'] UnetLoaderGGUF: - Value not in list: unet_name: 'wan2.2_t2v_high_noise_14B_Q8_0.gguf' not in ['Wan2.2-I2V-A14B-HighNoise-Q4_K_S.gguf', 'Wan2.2-I2V-A14B-LowNoise-Q4_K_S.gguf', 'Wan2.2-T2V-A14B-HighNoise-Q4_K_S.gguf', 'Wan2.2-T2V-A14B-LowNoise-Q4_K_S.gguf']

Aitrepreneur

This means that you don't have or are not choosing the correct models for the generation, it's really not the workflow...

Aitrepreneur

You might be using a vpn or don't have curl installed on your os

Jaina Shop

working perfectly out of the box with fantastic results! i'm using a 3090. one Question only, is it possible to set the frames at 15fps, instead 24fps? so will be more easy generate videos, picked up the good one, and interpolate only the one choose? will be much more quickly and easy instead run all of them in 24fps, thank you

Aitrepreneur

You need to update comfyui using the update comfyui bat file in the update folder

Aitrepreneur

Sure, it should speed up the generation but might make the video a bit worse in quality

Aitrepreneur

You can, wan 14b is usually trained on 16fps videos but 24 fps works as well with wan 2.2 but what matters since the number of frames not the fps number

Pizza

Ok, ima give up on the text to image one. I can get it to spit out a yellow box, but that is it.

Stephen Dobby

I had the same thing on a fresh install. the workflow had lower case "t2v". and i had the above error. K's response prompted me to re-select the model in the UnetLoaderGGUF node, it went to capitalised "T2V" (which matches the file names in my unet folder) and I got past the error. Love your work K.

Aitrepreneur

Man just send me a dm with screenshots instead of writting here trying to solve this one your own...

Bassplasma

hey , thanks for aswering me , i solved the problem ( and it's quite weird ) : i simply deleted a custom nodes called " flo wan 2 " and the generation started . Not sure why it was happening but im on the last comfy version anyways ..

Bassplasma

if that can help , i had the exact same issue last night , withj everything updated ( comfy and python dependencies ) , look up in your custom nodes is you have " flow2-wan-video " installed . Simply remove it , reload comfy . Solved the problem for me , and im not aware of the origin of the problem

Walker4k

Could you create a 1 click install of this for Massed Compute?

Robert

Hi guys, having this: curl: (7) Failed to connect to github.com port 443 after 31 ms: Couldn't connect to server Download failed. -Solved it by downloading latest version of comfyui from github and used models_nodes.bat

joaquin me llamo

Failed to open the file ComfyUI.7z: Permission denied

Maurice

I have the same issue on a fresh install on run_pod via the script. I also magaged to fix the model IDs. However, he also complained about the scheduler (bong_tangent) and the sampler name (res_2s), which are not available in the dropdown. I chose simple and euler instead. It does something, but the prompt is not being respected, especially for image 2 image.

James Stewart

Hi K and friends, from previous advice I'm doing a fresh Comfy install. Selected WAN2.2 Quantization option 2 Q5_K_S as I have 16gb GPU and am getting this error only about 5-10 mins into the install. This has happened twice now: -------- Downloading ComfyUI -------- % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0 97 2038M 97 1990M 0 0 5923k 0 0:05:52 0:05:44 0:00:08 5784k curl: (18) end of response with 50643857 bytes missing Download failed. Press any key to continue . . .

Michael Balzer

Crashing on installing ComfyUI "git-setup.exe" cannot start or run due to imcompatibilty with 64-bit versions of windows. --- UPDATE: installing Git manully resolved this

Aitrepreneur

I managed to find the error, I updated the runpod installer

Aitrepreneur

I updated the runpod installer, redownload it

Aitrepreneur

you might have some antivirus running or you are installing in a drive with limited permission

Aitrepreneur

seems like a internet connection issue maybe

Aitrepreneur

I think there might have been an issue with your connection or huggingface, basically the download got stopped. Try again now

Aitrepreneur

Me neither, i saw that solution as well. I don't even know what that node even is

Len

Did a clean install (on RTX 4090). WanVideoNAG node doesn't load (this node is part to imge2imge workflow. But even when I use TXT2IMG part, I just get noise, no actual picture... What am I doing wrong? https://ibb.co/Rk0FZg5m

Aitrepreneur

because you are using the Image to video model on the text to video workflow

Testtn

Runpod 5090, I do not see this asked but is it possible to edit the runpod Pod to use a 5090, or could you upload a new comfyui pod that we can use ?

Aitrepreneur

Foe now my template does not work with blackwell gpu

James Stewart

Have done while logged in to huggingface, looks like its working now. Crossing fingers, thanks again K.

Paul Mazurkiewicz

Im getting this error: KSamplerAdvanced Given groups=1, weight of size [5120, 36, 1, 2, 2], expected input[1, 32, 21, 90, 160] to have 36 channels, but got 32 channels instead

Aitrepreneur

You need to update comfyui using the comfyui update bat file in the update folder

Zoran Kanti-Paul

Does anyone know what makes the video output so jumpy and fast moving? I am trying to get normal smooth movement and it is impossible so far. Also, 81 length is giving my only 3 seconds. How to I gen 5 seconds?

Anvil

Can you choose to have multiple 4090 GPUs to speed up generation and make videos at a higher resolution?

Robert

Thanks man! Excellent tools! Solved everything just needed right loras, took me a whole night to get it to work.

Ash

Runpod 5090 is working, use these steps: cd ComfyUI source venv/bin/activate pip uninstall torch pip install --upgrade torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128 deactivate Then restart ComfyUI. Remember do not use his autoupdate script after this, else you will have to redo the above steps

Klaus The.

Can confirm this works for the RTX 6000 Pro (96 GB VRAM is very nice for longer videos!), too

Zoran Kanti-Paul

OK I am really getting nowhere here. Who can decipher this note?? <---HIGH NOISE MODEL LORAS EITHER USE lightx2V between 1 and 3 strength and 1 CFG 10 steps 0-5 OR NO LORA and 3.5 CFG 15 steps 0-7 Add additional lora with the button (only 2.2 compatible) That second part.. makes no sense to me. 3.5 CFG...then15...then step 0-7?? What does that mean? I see 3 numbers mentioned and there isn't 3 places to put numbers??

Jordan Goodfellow

Hey, do you know if this workflow will support vace controlnet img to video with the motion processor from video reference, similar to the WAN2.1 workflow you posted not too long ago?

Divinshield Entertainment

The first download keeps failing for me at 83% for some reason. I have tried 3 times. I will try again, and if it still fails, I will try again later today.

Dauron Robinson

Using the modeling for 12gb of veam with the same settings for image to video and running out of RAM 😕

Allan

is this compatible with CONTROLNET somehow?

TrentD

Perfect. Excellent work. :)

Wojtek Mazur

Installer just stops downloading and says download failed. tried 3 times, it crashes at 180 Mb downloaded. My internet connection is stable and I usualy get downloads in 8-12 Mb speeds, but the downloader doesnt go higher than 2 mbs and then it fails after downloading 180megabytes. Any clues?

Testtn

Fair enough. I was using an H100 SXM GPU with 80 GB VRAM last night.. 10 second i2v was about 2 - 3 minutes. Insane quality

Testtn

The H100 series GPUs work at blinding speeds as well. Going to check the 5090 though it'd be much cheaper

Steve Sermons

I'm also getting the failed download issue. Looks like a lot of people are. Any idea what the problem is?

RB

How much does that cost you?

Aitrepreneur

you can change from 24fps to 16fps for 5s, 81frames/16fps=+-5seconds. Usually that jumpy movement can be explained by either the prompt, the resolution or the use of loras for the high noise model

Aitrepreneur

maybe but you'll need to use the special multi gpu model nodes for that

Aitrepreneur

so this is only for the image to video and frame to frame since there is no wan 2.2 lightx loras. wan21_lightx2v works fairly well for the wan 2.2 image to video models. So you can either: -Use the wan21_lightx2v lora with the lora strength value between 1 & 3 (I recommend just leaving it at 3 by default like in the workflow) but if you do, you need to setup the cfg value for the high noise ksampler node to 1.0 and the steps at 10 with the start_at_step value 0 to end_at_step value to 5 (but the workflow's default 6 0-2 works really well too so you can just leave it like that) -Not use the wan21_lightx2v lora BUT IF YOU DO, you need to setup the cfg value for the high noise ksampler node to 3.5 and the steps at 15 with the start_at_step value 0 to end_at_step value to 7 BUT if you want my advice, if you have still no idea what I just explained and you are a complete beginner, just leave everything by default with the default values provided with the workflow. That's it.

Aitrepreneur

I mean it's not up to me, if the Vace wan 2.2 model comes out then yeah

Aitrepreneur

install the latest nvidia drivers and enable sysmem fallback, at least it will avoid getting out of memory errors.

Aitrepreneur

We need the Vace wan 2.2 model for this, which hasn't come out yet

Aitrepreneur

I mean...it could be on huggingface's server side or you have a vpn activated or some other connection issue here

Aitrepreneur

Yeah I mean if it's suddenly global might have been an issue on huggingface's server side, it happened to other people's yesterday too, then today it worked 🤷‍♂️

Aitrepreneur

Probably an issue on huggingface server side

Jordan Goodfellow

Gotcha. Sorry, I'm very new to the space and will have dumb questions

Testtn

$2 - $2.35 an hour for the H100 series. The 5090 is $0.75/hr. Which is a big difference, however I was just using a 5090 for about 2 hours and generated 50% of what an H100 series will do in the same time.

Tore van Kooten

ok so I am pretty new at this.. I followed the guide and got the "AssertionError" after updating comfyUI to newest version which was fixed by changing CUDA to 12.8. There was no JSON in the file list for this post, so I assumed it was the one from the post "WAN 2.2 ULTRA VIDEO & IMAGE WORKFLOW!". But now I get this error when trying to make an image: Prompt execution failed Prompt outputs failed validation: UnetLoaderGGUF: - Value not in list: unet_name: 'Wan2.2-T2V-A14B-LowNoise-Q8_0.gguf' not in ['Wan2.2-I2V-A14B-HighNoise-Q4_K_S.gguf', 'Wan2.2-I2V-A14B-LowNoise-Q4_K_S.gguf', 'Wan2.2-T2V-A14B-HighNoise-Q4_K_S.gguf', 'Wan2.2-T2V-A14B-LowNoise-Q4_K_S.gguf'] UnetLoaderGGUF: - Value not in list: unet_name: 'Wan2.2-T2V-A14B-HighNoise-Q8_0.gguf' not in ['Wan2.2-I2V-A14B-HighNoise-Q4_K_S.gguf', 'Wan2.2-I2V-A14B-LowNoise-Q4_K_S.gguf', 'Wan2.2-T2V-A14B-HighNoise-Q4_K_S.gguf', 'Wan2.2-T2V-A14B-LowNoise-Q4_K_S.gguf']

Ilir

Hi there, I do have the same issue. For some reason the workflow is using other filenames for the models. However if you use the arrows next to the models that are being used, you will be able to select the correct Models, that actually have the sames names, but they differ a little bit.

Tore van Kooten

thanks buddy :), That solved it! also at first I choose V2V and wondered why I just got noise... easy to tell I am new at this.

Rahul Rabha

memory limit triggered for generating first video t2v in runpod with 4090. Op said 4090 is best. do i need to use higher GPU, or there any other way.

Aitrepreneur

You need to use the run_nvidia_gpu.bat file

Aitrepreneur

What resolution were you using? I never had oom with a 4090

PTMarks

Hi, I ran the models downloader to get the new loras and now comfyui doesn't launch, getting this error: Checkpoint files will always be loaded safely. Traceback (most recent call last): File "D:\WAN2_2\ComfyUI_windows_portable\ComfyUI\main.py", line 147, in import execution File "D:\WAN2_2\ComfyUI_windows_portable\ComfyUI\execution.py", line 15, in import comfy.model_management File "D:\WAN2_2\ComfyUI_windows_portable\ComfyUI\comfy\model_management.py", line 233, in total_vram = get_total_memory(get_torch_device()) / (1024 * 1024) ^^^^^^^^^^^^^^^^^^ File "D:\WAN2_2\ComfyUI_windows_portable\ComfyUI\comfy\model_management.py", line 183, in get_torch_device return torch.device(torch.cuda.current_device()) ^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "D:\WAN2_2\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\cuda\__init__.py", line 1071, in current_device _lazy_init() File "D:\WAN2_2\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\cuda\__init__.py", line 403, in _lazy_init raise AssertionError("Torch not compiled with CUDA enabled") AssertionError: Torch not compiled with CUDA enabled D:\WAN2_2\ComfyUI_windows_portable>pause Press any key to continue . . .

Part Time Poker Pro

LMAO dude I'm an idiot. I downloaded your JSON but forgot to load it up, i was looking at a different workflow. YOUR workflow seems to be working now. thanks dude

Pak26

I just swap the 2x T2v loras for the new 4 steps i2v loras, in your i2v workflow, change for 4 steps in both Ksampler, 1 .0 for both Lora strength and it takes a very long time + the output is very blurry ! I'm I missing something ? Thanks

J.ruthless

im having this same issue, if i use the original T2Vlight lora non high/low it works fine in the I2V workflow provided, But using the new I2V high/low loras just released causes blurs, also wanna mention it have updated in the manager and also with the update bat. When I am getting these blur issues

B Reaper

I'm currently experiencing this after a fresh install, renders out on text to image, very over contrast images or a mess, or for txt to video it renders a very snowy/mosaic looking video, you can see the video underneath but its all messy. Just a fresh install no changes on a 3090 strix

Feindbold

Good to know im not the only one, also having strong blur with the new I2V lora.

Len

Just a question. Is 720p maximum resolution I can do with 4090 card? I tried 1080p and got out of memory error...

Aitrepreneur

Yeah pretty much, if you want higher res you can either rent a bigger gpu on runpod or just upscale the video with the upscale workflow

Aitrepreneur

What are you talking about? Why did you swap the text to video loras with the image to video ones?? It's not meant for that. Just download the updated workflow and use the default parameters

Aitrepreneur

Read the post, i explain this issue and how to solve it

Aitrepreneur

You are probably using the wrong models like i2v (image to video) instead of t2v (text to video)

Cristo

I have just formated my computer, If I download this.. everything will work fine or do I have to download python,git and other stuff first?

Aerofire

Which models do you use then? The only options are i2v and t2v it comes this way out of the box

PTMarks

That worked. Thanks. Sorry I panicked....

Rahul Rabha

Didn't touch anything, just default. Let me check with settings. I'll tell.

Feindbold

Okay my bad, found out why it was blurry for me after updating. I updated the workflow but that obviously didn't magically download the new 4 step I2V lora. Downloaded the lora and now its working. The 4 step lora is for sure faster BUT: After some testing I would say there is quite a quality drop in comparison to the old light2v workflow with 6 steps. Also the 4 step I2V lora does not work as well with other loras as the old light2v.

Anvil

Really wish I was smart enough to understand what this all meant :/

mspi51

I have 2 NVIDIA GPUs. Is there a way to tell which one to use?

Testtn

img2vid Runpod, today W2.2, W2.1 models and Vae are loading in between each video generation taking incredibly long time. The past couple days have been smooth sailing but then today in a span of a few hours I barely pushed out 5 x 6 second clips. I was also trying to troubleshoot, started from scratch a couple times, tried various GPUs but Nada. I noticed only the GPU and the GPU memory was being used. The CPU and CPU memory were sitting idle all day for every separate pod I deployed. And I'm not sure why but every time it went to generative video it loaded the 2.2 model, did its thing, then loaded the 2.1 model that did its thing and somewhere in there it also loaded the vae. The generations were also really crappy compared to how they having the last couple days. Curious if anybody else has had this issue or if it's just me?

Anvil

Yep, reconnects between each fucking step, super annoying

Anvil

actually, solved my issue by using a 5090 and someone else's suggestion on how to run it, bout twice as fast now and no more restarting

Zoltan Vegh

heey, thank you, amazing work! what shall people do excacly who installed your first version? everything is working now, so I try to do the right thing to install the faster loras and the new workflow. I just have to ran the models downloader and update the workflow? Or do I have to do something with the installer too?

Ash

The runpod(cloud gpu) 5090 is running an older version of pytorch, so we update the version of pytorch so comfyui supports it. chatgpt it next time my friend, devs aren't that smart, but don't let it reduce your morale

Len

Disregard my question. Yeah T2V does image generation too. Sorry I was confused thinking that there should be T2I model, but there's none

Jeff Smith

I keep getting this error when trying the auto installer. Is there a solution to this problem yet? Seems as though its been an issue for many people. -------- Downloading ComfyUI -------- % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0 42 2038M 42 870M 0 0 2865k 0 0:12:08 0:05:10 0:06:58 2748k curl: (18) end of response with 1225048977 bytes missing Download failed. Press any key to continue . . .

Wojtek Mazur

Hi Man. I had this too. I downloaded stuff manually in the end. The Bat file seems to time out.

Tohnbon

I tried the one click installer and got. Cannot start or run due to incompatibility with 64 bit versions of windows? What do I do?

gabbledub

Do you have Git installed? I had the same error and installing Git did the trick.

Wojtek Mazur

"Unet LoaderGGUF Cannot Reshape Array of Size 43403491 into shape (5120, 9504)" That comes up when I try I2V with new loras. Any clue K boii?

Jeff Smith

Still working on finding everything. I have Comfy up and running, looking for the model files now.

Tohnbon

No is there a link for that? How do you install Git? Are there instructions or tutorial. I can't find anything here or on Aitrepreneur youtube channel. Or least I can't seem to find it?

Tohnbon

I installed Git but I'm still getting the same error.

GlobalGrooveWORLD

I2B does not work. I uploaded a 1280 x 720 image, and it is the same size, immediately displays an error message, and highlights this model in purple. UnetLoaderGGUF cannot reshape array of size 29328717 into shape (13824,5440)

Robert Zinke

RUNPOD QUESTION/ERROR: Whenever I restart after successfully installing AND running everything I get an error. It was working flawlessly before but then I restart after uploading some loras and get this: Restarting... [Legacy Mode] Command: ['/workspace/ComfyUI/venv/bin/python3', 'main.py', '--listen', '0.0.0.0', '--port', '3001'] Using Python 3.11.9 environment at: venv [START] Security scan [DONE] Security scan ## ComfyUI-Manager: installing dependencies done. ** ComfyUI startup time: 2025-08-08 22:43:17.722 ** Platform: Linux ** Python version: 3.11.9 (main, Apr 6 2024, 17:59:24) [GCC 11.4.0] ** Python executable: /workspace/ComfyUI/venv/bin/python3 ** ComfyUI Path: /workspace/ComfyUI ** ComfyUI Base Folder Path: /workspace/ComfyUI ** User directory: /workspace/ComfyUI/user ** ComfyUI-Manager config path: /workspace/ComfyUI/user/default/ComfyUI-Manager/config.ini ** Log path: /workspace/ComfyUI/user/comfyui.log Using Python 3.11.9 environment at: venv Using Python 3.11.9 environment at: venv [ComfyUI-Manager] PyTorch is not installed Prestartup times for custom nodes: 0.0 seconds: /workspace/ComfyUI/custom_nodes/rgthree-comfy 0.0 seconds: /workspace/ComfyUI/custom_nodes/ComfyUI-Easy-Use 2.1 seconds: /workspace/ComfyUI/custom_nodes/ComfyUI-Manager Checkpoint files will always be loaded safely. Total VRAM 24107 MB, total RAM 902891 MB pytorch version: 2.8.0+cu128 WARNING[XFORMERS]: xFormers can't load C++/CUDA extensions. xFormers was built for: PyTorch 2.4.0+cu121 with CUDA 1201 (you have 2.8.0+cu128) Python 3.11.9 (you have 3.11.9) Please reinstall xformers (see https://github.com/facebookresearch/xformers#installing-xformers) Memory-efficient attention, SwiGLU, sparse and more won't be available. Set XFORMERS_MORE_DETAILS=1 for more details Set vram state to: NORMAL_VRAM Device: cuda:0 NVIDIA GeForce RTX 4090 : cudaMallocAsync Traceback (most recent call last): File "/workspace/ComfyUI/main.py", line 147, in import execution File "/workspace/ComfyUI/execution.py", line 16, in import nodes File "/workspace/ComfyUI/nodes.py", line 24, in import comfy.diffusers_load File "/workspace/ComfyUI/comfy/diffusers_load.py", line 3, in import comfy.sd File "/workspace/ComfyUI/comfy/sd.py", line 11, in from .ldm.cascade.stage_c_coder import StageC_coder File "/workspace/ComfyUI/comfy/ldm/cascade/stage_c_coder.py", line 19, in import torchvision File "/usr/local/lib/python3.11/dist-packages/torchvision/__init__.py", line 10, in from torchvision import _meta_registrations, datasets, io, models, ops, transforms, utils # usort:skip ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/torchvision/_meta_registrations.py", line 163, in @torch.library.register_fake("torchvision::nms") ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/workspace/ComfyUI/venv/lib/python3.11/site-packages/torch/library.py", line 1069, in register use_lib._register_fake( File "/workspace/ComfyUI/venv/lib/python3.11/site-packages/torch/library.py", line 219, in _register_fake handle = entry.fake_impl.register( ^^^^^^^^^^^^^^^^^^^^^^^^^ File "/workspace/ComfyUI/venv/lib/python3.11/site-packages/torch/_library/fake_impl.py", line 50, in register if torch._C._dispatch_has_kernel_for_dispatch_key(self.qualname, "Meta"): ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ RuntimeError: operator torchvision::nms does not exist

Roni Bliss

Great setup and workflows! Thanks for your work! ... The only thing I was disappointed with was it downloaded GGUF models of 2.2 which are just not as good quality at all... I use H200's for video generation, I have 141gb VRAM . So I want full red blooded OG models for proper quality and capabilities of the models| If anythiing I would go for fp8's over GGUF's I have yet to see any of the perform better or faster than fp8's, there is a false sense of BS with the quantized models. Easily fixed in the .sh file though.

Vicente Varela

Good Morning, I feel like I'm always installing the bat. file in the wrong location. Is there a way to uninstall everything and start fresh? What is the correct folder I should always use? Thanks,

RG

@Krush Oldwin: Hi, I'm also running an H200. But my doesn't work, cannot generate anything. Getting this error: CLIPTextEncode CUDA error: operation not supported CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1 Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions. --> on Wan2.2 or Kontext. Same on SwarmUI. I'm on W11Pro, I also have a SwarmUI setup that works on Linux/Ubuntu, but my W11Pro H200 setup doesn't want to work, get crashes and H200 doesn't even appear sometimes once booted. How did you get yours to work at all? or did you have similar issues? Maybe you modded Aitrepreneur setup to work on Linux (you mentioned a .sh fix)? Would appreciate any input, thanks.

Alessa

Is is possible that you may add firstframelastframe to this for longer videos?

uniqornate

how can we make continuous interpolations with last frame and combine them in a single video? if there is possibility can you update with that please?

Xaro

Boxertwin

Hi, AI Toolkit has the WAN2.2 ti2v 5b model for training WAN2.2 LORAS but they dont work on your WAN2.2 workflow. Would you possibly create a workflow to work with this model (and these LoRAs)?

Zoltan Vegh

maybe try with some prompts like gazing into the camera, or give specific face expression

Testtn

I think I know whose advice you're talking about I'll give that a try this evening

m16akm

Tip: Including Patch Sage Attention improves generation speed by 1.5-2x

傑生 尚

Great workflow. Thanks for your hard work. Apologies if this was already covered, but any reason why in the i2v section of the workflow it always reverts to the starting image after following the prompt? Any way to get it to be a through-composed video without going back to the first image?

RavenLand

Lame question but, is there a way to make the image-to-video generated longer than the default 3 seconds?

Xaro

Very simple: change the *length* for example to 81 for ~5 seconds

Xaro

+1 this would be awesome

UNX

For some reason the Image to Video is not working for me. The output is totally different than the input image.

Nicolas Giarrusso

How can I apply a mask so that it only affects a portion of the image when I do img2img in Imagen with WAN 2.2?

Tark Morris

does this not work for amd graphics cards? I didn't see any mention of that anywhere...

Sean Rowe

Is there a prompt generator workflow/model we can use to generate NSFW prompts? It would be awesome to have a creative prompt generator that uses WAN prompt styles to help generate better images and video.

Kris Raves

When I try to generate an image or video on a 4090 on RunPod the Pod Utilization Memory maxes out and comfyui progress freezes on Clean VRAM used. I have to rent a 5000 ADA to generate anything. Am I doing something wrong?

Aitrepreneur

yeah, probably the resolution or length that is too high

Aitrepreneur

maybe use some uncensored LLM and feed a bunch of prompts examples and ask for what you want. I think venice ai can do that

Aitrepreneur

make sure that you are using the image to video model, not the text to video

dougadam

Hello, since this morning I have an issue with Runpod. When I connect to a H200SXM pod like I usually do, I can only access to port 8888, I do not have other choices like usually. So I can not connect to the ComfyUI page (3000). I tested on a 4090 pod, it's the same. On the logs of the pod it seems OK : Status: Image is up to date for aitrepreneur/comfyui:2.3.5 On the comfyui.log, it says : To see the GUI go to: http://0.0.0.0:3001 but when I go to https://xxx-3001.proxy.runpod.net I have a 404 error. Is it an issue on Runpod side or with the template?

dougadam

Nevermind, after several attempts it seems to work now.

Cactopus

To elaborate if anyone's interested, I followed https://github.com/loscrossos/helper_comfyUI_accel to get it working for this workflow and it basically doubled my speed. Scroll to the issues section of the repo or https://github.com/loscrossos/helper_comfyUI_accel/issues/10 to circumvent one issue I encountered. I feel like Aitrepreneur should test it out and make a video guide down the line since 2x generation speed (from 320s/it to slightly more than half at 175s/it in my case) is really quite incredible.

Cactopus

I observed this behavior as well. The image to video workflow/model (idk where it originates from) often tries hard to match the last frame of the short video to the starting frame (base image) and creating a semi-looping effect, though not matching the starting frame exactly. e.g. I would prompt for an action/change to an object in the base image (like an object turning blue or a person reaching out their hand) and near the end it would likely revert back close to the original state unprompted. I wonder what is causing it.

Bradford

After generating dozens of videos with this on my own computer, I came to the conclusion that this is not the best video generator and that Hunyuan video is much better. That's because all of the people in the videos generated with WAN text to video have stiff and expressionless faces! In fact it's so bad that it's like the people's faces are basically just still images throughout the whole video.

Aitrepreneur

100% not true, probably because you are not crafting correct prompts, read the prompting guide

Xaro

Im trying to extend the Videos on my own and use the last frame of the created Video for the next Video. Even after upscaling the last frame (Image) before creating the 2nd video, quality gets badder and badder for next Videos. Any idea for a better way to do it? Thx ❤️✨

jhon

i think you might have already installed comfy ui and are trying to install again in seperate folder. this worked for me IF YOU ALREADY HAVE COMFYUI INSTALLED USING A PREVIOUS INSTALLER: Download the WAN2_2-ULTRA-MODELS-NODES_INSTALL.bat and place it in the ComfyUI_windows_portable\ComfyUI folder Run the bat file and select either option 1, 2 or 3 depending on your GPU vram and then determine if you want the Original Wan model (recommended) . Wait for the install to finish. i got cuda error after that but fixed following instructions on how to fix on wan 2.2 ultra video instruction download post...ater that i could not generate videos until i turned my windows paging file size set to let windows decide.

jhon

121 FOR 5 SECONDS, 81 FOR 3

jhon

i think you might have already installed comfy ui and are trying to install again in seperate folder. this worked for me IF YOU ALREADY HAVE COMFYUI INSTALLED USING A PREVIOUS INSTALLER: Download the WAN2_2-ULTRA-MODELS-NODES_INSTALL.bat and place it in the ComfyUI_windows_portable\ComfyUI folder Run the bat file and select either option 1, 2 or 3 depending on your GPU vram and then determine if you want the Original Wan model (recommended) . Wait for the install to finish. i got cuda error after that but fixed following instructions on how to fix on wan 2.2 ultra video instruction download post...ater that i could not generate videos until i turned my windows paging file size set to let windows decide.

jhon

i think you might have already installed comfy ui and are trying to install again in seperate folder. this worked for me IF YOU ALREADY HAVE COMFYUI INSTALLED USING A PREVIOUS INSTALLER: Download the WAN2_2-ULTRA-MODELS-NODES_INSTALL.bat and place it in the ComfyUI_windows_portable\ComfyUI folder Run the bat file and select either option 1, 2 or 3 depending on your GPU vram and then determine if you want the Original Wan model (recommended) . Wait for the install to finish. i got cuda error after that but fixed following instructions on how to fix on wan 2.2 ultra video instruction download post...ater that i could not generate videos until i turned my windows paging file size set to let windows decide.

jhon

i think you might have already installed comfy ui and are trying to install again in seperate folder. this worked for me IF YOU ALREADY HAVE COMFYUI INSTALLED USING A PREVIOUS INSTALLER: Download the WAN2_2-ULTRA-MODELS-NODES_INSTALL.bat and place it in the ComfyUI_windows_portable\ComfyUI folder Run the bat file and select either option 1, 2 or 3 depending on your GPU vram and then determine if you want the Original Wan model (recommended) . Wait for the install to finish. i got cuda error after that but fixed following instructions on how to fix on wan 2.2 ultra video instruction download post...ater that i could not generate videos until i turned my windows paging file size set to let windows decide.

jhon

try 121 length for 5

jhon

i think you might have already installed comfy ui and are trying to install again in seperate folder. this worked for me IF YOU ALREADY HAVE COMFYUI INSTALLED USING A PREVIOUS INSTALLER: Download the WAN2_2-ULTRA-MODELS-NODES_INSTALL.bat and place it in the ComfyUI_windows_portable\ComfyUI folder Run the bat file and select either option 1, 2 or 3 depending on your GPU vram and then determine if you want the Original Wan model (recommended) . Wait for the install to finish. i got cuda error after that but fixed following instructions on how to fix on wan 2.2 ultra video instruction download post...ater that i could not generate videos until i turned my windows paging file size set to let windows decide.

jhon

i think you might have already installed comfy ui and are trying to install again in seperate folder. this worked for me IF YOU ALREADY HAVE COMFYUI INSTALLED USING A PREVIOUS INSTALLER: Download the WAN2_2-ULTRA-MODELS-NODES_INSTALL.bat and place it in the ComfyUI_windows_portable\ComfyUI folder Run the bat file and select either option 1, 2 or 3 depending on your GPU vram and then determine if you want the Original Wan model (recommended) . Wait for the install to finish. i got cuda error after that but fixed following instructions on how to fix on wan 2.2 ultra video instruction download post...ater that i could not generate videos until i turned my windows paging file size set to let windows decide.

falkor

File "E:\AI\ComfyUI_windows_portable\python_embeded\Lib\site-packages\transformers\utils\versions.py", line 117, in require_version_core return require_version(requirement, hint) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "E:\AI\ComfyUI_windows_portable\python_embeded\Lib\site-packages\transformers\utils\versions.py", line 111, in require_version _compare_versions(op, got_ver, want_ver, requirement, pkg, hint) File "E:\AI\ComfyUI_windows_portable\python_embeded\Lib\site-packages\transformers\utils\versions.py", line 44, in _compare_versions raise ImportError( ImportError: huggingface-hub>=0.34.0,<1.0 is required for a normal functioning of this module, but found huggingface-hub==0.30.2. Try: `pip install transformers -U` or `pip install -e '.[dev]'` if you're working with git main Press any key to continue . . . Message after I executed the WAN2_2-ULTRA-COMFYUI-MANAGER_AUTO_INSTALL.bat When I run the run_nvidia_gpu.bat in my ComfyUI_windows_portable dir I get E:\AI\ComfyUI_windows_portable>.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build Total VRAM 24564 MB, total RAM 64652 MB pytorch version: 2.7.1+cu128 Set vram state to: NORMAL_VRAM Device: cuda:0 NVIDIA GeForce RTX 4090 : native Checkpoint files will always be loaded safely. Using pytorch attention Failed to execute startup-script: E:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\agilly1989_motorway\prestartup_script.py / huggingface-hub>=0.34.0,<1.0 is required for a normal functioning of this module, but found huggingface-hub==0.30.2. Try: `pip install transformers -U` or `pip install -e '.[dev]'` if you're working with git main [START] Security scan [DONE] Security scan ## ComfyUI-Manager: installing dependencies done. ** ComfyUI startup time: 2025-08-15 23:19:46.246 ** Platform: Windows ** Python version: 3.12.10 (tags/v3.12.10:0cc8128, Apr 8 2025, 12:21:36) [MSC v.1943 64 bit (AMD64)] ** Python executable: E:\AI\ComfyUI_windows_portable\python_embeded\python.exe ** ComfyUI Path: E:\AI\ComfyUI_windows_portable\ComfyUI ** ComfyUI Base Folder Path: E:\AI\ComfyUI_windows_portable\ComfyUI ** User directory: E:\AI\ComfyUI_windows_portable\ComfyUI\user ** ComfyUI-Manager config path: E:\AI\ComfyUI_windows_portable\ComfyUI\user\default\ComfyUI-Manager\config.ini ** Log path: E:\AI\ComfyUI_windows_portable\ComfyUI\user\comfyui.log Prestartup times for custom nodes: 0.0 seconds: E:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\rgthree-comfy 0.0 seconds: E:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\comfyui-easy-use 0.0 seconds: E:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\comfyui-marigold 1.5 seconds: E:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\comfyui-manager 3.4 seconds (PRESTARTUP FAILED): E:\AI\ComfyUI_windows_portable\ComfyUI\custom_nodes\agilly1989_motorway WARNING: Potential Error in code: Torch already imported, torch should never be imported before this point. Traceback (most recent call last): File "E:\AI\ComfyUI_windows_portable\ComfyUI\main.py", line 147, in import execution File "E:\AI\ComfyUI_windows_portable\ComfyUI\execution.py", line 16, in import nodes File "E:\AI\ComfyUI_windows_portable\ComfyUI\nodes.py", line 24, in import comfy.diffusers_load File "E:\AI\ComfyUI_windows_portable\ComfyUI\comfy\diffusers_load.py", line 3, in import comfy.sd File "E:\AI\ComfyUI_windows_portable\ComfyUI\comfy\sd.py", line 29, in from . import model_detection File "E:\AI\ComfyUI_windows_portable\ComfyUI\comfy\model_detection.py", line 2, in import comfy.supported_models File "E:\AI\ComfyUI_windows_portable\ComfyUI\comfy\supported_models.py", line 5, in from . import sd1_clip File "E:\AI\ComfyUI_windows_portable\ComfyUI\comfy\sd1_clip.py", line 3, in from transformers import CLIPTokenizer File "E:\AI\ComfyUI_windows_portable\python_embeded\Lib\site-packages\transformers\__init__.py", line 27, in from . import dependency_versions_check File "E:\AI\ComfyUI_windows_portable\python_embeded\Lib\site-packages\transformers\dependency_versions_check.py", line 57, in require_version_core(deps[pkg]) File "E:\AI\ComfyUI_windows_portable\python_embeded\Lib\site-packages\transformers\utils\versions.py", line 117, in require_version_core return require_version(requirement, hint) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "E:\AI\ComfyUI_windows_portable\python_embeded\Lib\site-packages\transformers\utils\versions.py", line 111, in require_version _compare_versions(op, got_ver, want_ver, requirement, pkg, hint) File "E:\AI\ComfyUI_windows_portable\python_embeded\Lib\site-packages\transformers\utils\versions.py", line 44, in _compare_versions raise ImportError( ImportError: huggingface-hub>=0.34.0,<1.0 is required for a normal functioning of this module, but found huggingface-hub==0.30.2. Try: `pip install transformers -U` or `pip install -e '.[dev]'` if you're working with git main E:\AI\ComfyUI_windows_portable>pause Press any key to continue . . .

Light is Consciousness (edited)

Comment edits

2026-09-08 20:00 Thanks for your work, Aitrepreneur! I'm new to all of this, so sorry if this is a noob question, but once I write my prompt to generate an image, and click Run, I get this Error message: 'Prompt execution failed Prompt has no outputs'. What am I doing wrong? Thanks for the help!
2025-08-16 23:14

Goblin

If we download another lora, how will we know which one to put it in?

Aitrepreneur

I have a new workflow available, you can try that one and compare

Aitrepreneur

dm me screenshots of your parameters before generation

Aitrepreneur

I don't understand the question, each lora has a name with a specific prompt, you select the lora in the node then use the trigger in the prompt

Goblin

well, the lora is attached to the high or low noise model, right?

Creative Khim

Why I am only getting 3 seconds video with 81 length? 720x1280

Creative Khim

Also, whats the settings need to change to make the output video not doing slow motion?

Brandon D'souza

I don't see the nodes that enable the workflow. How can I fix it?

Aitrepreneur

it's not "attached" but I guess you mean that for 2.2 you need a different lora for high noise and low noise models and once again, they will have different names.

Aitrepreneur

Because 81 frames at 24fps : 81/34= 3,375 s There are no settings that do that, it's a comon wan model training "issue" oftentimes a good prompt and not lora for the high noise model solve this issue

Aitrepreneur

use the update comfyui bat file inside the update folder to update comfyui correctly and use the update all to update all the nodes, then restart and redrag the workflow inside again

RongenNo1

same problem here too. i tryed to update comfyui like the overlord says, but it didn´t work. i get dozens of errors, something with python and some check_error. i really see how much effort and work he/she put into the workflows and installer. but the whole AI/videoAI creation process is extremely complex and hard to get running. i understand nothing what i´m doing here and for me as "user" or "customer" it is not usable if there is no masterdegree of computer sience on the wall. i hope some day our overlord will create some easy to use program like stabillity matrix.

Creative Khim

thanks for the response, also can you invite me on discord? the link on video description is already invalid

ПЕТЪР ЕВТИМОВ

WOW, all is amazing and works perfectly! With the mouse lady promt 1280/720 28 minutes of generation. All is perfect from the first time. 480/480 - 5 minutes generation. The Workflow is really good! Thanks. Do you have some fresh for Song Making with LIPS SINC? And something with Music Sinc for dance videos? It will be great!!!

Barnabás Telek

Did u manage to solve it? I have the same problem and tried everything

kakarot goku

When I hit enter after entering my VRAM it says not compatible with windows 64bit. I'm very new to this, and I have a 9060xt. Is this not compatible with AMD?

Leshen

I'm new here. How did you do that?

Jesse Huovinen

For me it worked perfectly in the beginning, but then I started having a bunch of errors, runpod throws me out, can't download the videos, missing nodes, when installing missing nodes, gives me this: VAELoader: - Value not in list: vae_name: 'wan_2.1_vae.safetensors' not in ['sdxl_vae.safetensors'] VAELoader: - Value not in list: vae_name: 'wan_2.1_vae.safetensors' not in ['sdxl_vae.safetensors'] CLIPLoaderGGUF: - Value not in list: clip_name: 'umt5-xxl-encoder-Q5_K_S.gguf' not in [] UnetLoaderGGUF: - Value not in list: unet_name: 'Wan2.2-I2V-A14B-HighNoise-Q8_0.gguf' not in [] UnetLoaderGGUF: - Value not in list: unet_name: 'Wan2.2-I2V-A14B-LowNoise-Q8_0.gguf' not in [] CLIPLoaderGGUF: - Value not in list: clip_name: 'umt5-xxl-encoder-Q5_K_S.gguf' not in [] UnetLoaderGGUF: - Value not in list: unet_name: 'Wan2.2-I2V-A14B-HighNoise-Q8_0.gguf' not in [] UnetLoaderGGUF: - Value not in list: unet_name: 'Wan2.2-I2V-A14B-LowNoise-Q8_0.gguf' not in [] VAELoader: - Value not in list: vae_name: 'wan_2.1_vae.safetensors' not in ['sdxl_vae.safetensors'] CLIPLoaderGGUF: - Value not in list: clip_name: 'umt5-xxl-encoder-Q5_K_S.gguf' not in [] UnetLoaderGGUF: - Value not in list: unet_name: 'Wan2.2-I2V-A14B-LowNoise-Q8_0.gguf' not in [] VAELoader: - Value not in list: vae_name: 'wan_2.1_vae.safetensors' not in ['sdxl_vae.safetensors'] UnetLoaderGGUF: - Value not in list: unet_name: 'Wan2.2-I2V-A14B-HighNoise-Q8_0.gguf' not in [] CLIPLoaderGGUF: - Value not in list: clip_name: 'umt5-xxl-encoder-Q5_K_S.gguf' not in [] UnetLoaderGGUF: - Value not in list: unet_name: 'Wan2.2-I2V-A14B-LowNoise-Q8_0.gguf' not in [] VAELoader: - Value not in list: vae_name: 'wan_2.1_vae.safetensors' not in ['sdxl_vae.safetensors'] UnetLoaderGGUF: - Value not in list: unet_name: 'Wan2.2-I2V-A14B-HighNoise-Q8_0.gguf' not in [] I have also DM'd you about this, thank you in advance brother

Don

On Runpod, will I have to download the models every single time I want to start it up? That's what it appears to be doing and I have storage hooked up to it. Thanks!

Jeffrey Dougan

I keep getting the error: "KSamplerAdvanced 'NoneType' object has no attribute 'to'. I'm having trouble figuring out this error with the Wan2.2 Text to Image generation. I have watched the updated video to mirror the correct workflow settings and inputs, but I can't seem to fix it. Any suggestions would be most welcome!

daniel baychev

did you just change the github link or something else as well.. I can't figure out why is this happening. I even tryed changing the link to other versions (newer)

Bimb-OH!

This, please. Isn't there an option to stop the pod, but keep the files and models installed?

Wuiiiwui

For us Linux users, do we just run the runpod.sh file? or is there no installer for the linux users (yet)?

satsFlowMaxi

It would be nice to have linux equivalent of the windows one click installs which prompted for the low vram model options

kirazo

It doesn't understand nsfw prompt. How can I active nsfw mode?

SK

Using an RTX3090 and image to video generation stops at VAE loading part (latest message in the web UI: "Clean VRAM used" ) and ComfyUI crashes. The first two parts of the workflow work fine, both low and high noise models are loaded and do their "things" for about 4 minutes each. I have Arch Linux (let Claude convert the Runpod script for easy installation). Latest messages in console: Requested to load WanVAE loaded completely 3313.197265625 242.02829551696777 True [1] 24863 killed python ./main.py --listen Having similar issue in text to image, too. The process gets killed when the UI shows "VAE Decode" and the console has: Requested to load WanVAE [1] 26429 killed python ./main.py --listen --use-sage-attention

Troy

When I try using text-to-image and text-to-video, I am getting this error: Prompt outputs failed validation: UnetLoaderGGUF: - Value not in list: unet_name: 'Wan2.2-T2V-A14B-LowNoise-Q8_0.gguf' not in ['Wan2.2-I2V-A14B-HighNoise-Q4_K_S.gguf', 'Wan2.2-I2V-A14B-LowNoise-Q4_K_S.gguf', 'Wan2.2-T2V-A14B-HighNoise-Q4_K_S.gguf', 'Wan2.2-T2V-A14B-LowNoise-Q4_K_S.gguf'] UnetLoaderGGUF: - Value not in list: unet_name: 'Wan2.2-T2V-A14B-HighNoise-Q8_0.gguf' not in ['Wan2.2-I2V-A14B-HighNoise-Q4_K_S.gguf', 'Wan2.2-I2V-A14B-LowNoise-Q4_K_S.gguf', 'Wan2.2-T2V-A14B-HighNoise-Q4_K_S.gguf', 'Wan2.2-T2V-A14B-LowNoise-Q4_K_S.gguf'] Also, when I try to upload an image (image-to-image or simple upscale), I get a "500 - Internal Server Error".

Francesco Gariboldi

i have the same problem with the oneclick instaler and i have an rtx 3080

Aitrepreneur

because you are not selecting the correct model in the model node

Troy

Thank you! I just changed the models (to what is available). But how do I solve the "500 - Internal Server Error"? This happens when I try to load an image for image-to-video.

Omnisen

Hi, I have a 4090 with 24 GB of VRAM and 96 GB of DDR5 5600 MHz RAM (G.SKILL). I know that Aitrepreneur mentioned that the Q5 version works better with GPUs that have 24 GB of VRAM. However, if the high- and low-noise models are run sequentially rather than simultaneously, then each model—which requires 15.4 GB of VRAM—should only use 15.4 GB of my 24 GB card. Only then I can use the Q8 Version for my card Am I wrong i need an answer please ?

Meshoo Channel

sorry i'm a complete beginner here, how do i open the workflow in the first place? in what app?

Kankatzu

any one knows if the RTX 3090 will be fine with the Q8 model, or should I go with the Q5 one?

Kankatzu

nevermind watched the video and he recommends not to go with q8 if you only have 24GB

Kankatzu

... comfyui? ya know... the thing you downloaded with the bat... and thats perfectly described in text and video alike and how to set it up...? To make it short... if you downloaded the auto install.bat you did most likely run it. Comfyui is now installed, go in the folder and somewhere there should be a run_nvidia_gpu.bat wich you will run. It should open the ui in your standard browser. There you can just drag and drop your workflow into. Or you could put the workflow file into ComfyUI/user/default/workflows wich would allow you to choose your workflow from the side bar. (useful if you use different workflows in the same comfyui install)

Untamed

i have used the q8 version with a 3090, but I'm not entirely sure if it wouldn't be faster using q5

games mail

Why do my videos from I2V always get cropped? I tried different size images and formats, including 1280x720 but the final video still gets cropped.

ilyas lachgar

Why did I get a CUDA compatibility error, I used RTX5090 ? CLIPTextEncode CUDA error: no kernel image is available for execution on the device CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1 Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

Otto Stephan

My output is always pure white. I had to manually enable the output image node, is that normal?

Lawrence007

Hey I keep getting download failed at around 240 megabytes.

Juan Gassa

I have this issue with a 5060ti 16gb :( . some ideas? cannot reshape array of size 23694782 into shape (5120,9504)

FractalD

I just got this error. I was hoping to use the 5090. Guess it can only be done on the 4090? Did you ever get it resolved or you had to switch?

Dragonbait

I'm looking to get a new GPU in the next few months, do these setups work with AMD or only Nvidia?

Senpai Ashua

can i install it on Mimicpc comfyUI?

Scott Beowulf Kaine

I have a question ive got a 5090 and Im trying to use the f16 models for say image to video and I keep getting a Unexpected architecture type in GGUF file: 'wan2.2', is there something special or node I need to use to make it compatible with the f16 models? I'm using the city96 and I even tried the Multigpu gguf loader and I get the same error

witteeend

I have been playing around with this for a bit on runpod and it is working, thank you. Question though. The majority of my output is kind of washed out, grainy.. Also, the speed seems to be too fast. any hints? :) thanks!

Holy Buckeye

is there anyone here that can assist me with the directions on what to do on the runpod setup so i can use a 5090?

Holy Buckeye

Ive been getting the same code, i even had someone give me directions, but seems something was missed because i keep getting it

Fredden

I get this on Runpod:CUDA error: no kernel image is available for execution on the device CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1 Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.

Leandro Lüssi

Did someone here manage to make it work? I also got this error using the rtx5090

Kamotz

Not using the 5090 for this workflow seems to be the solution. I only get this error with that GPU. I usually use the RTX 6000 Ada.

Fredden

thx for reply, will try :)

Bil Mor

You need a different CUDA for the 5090. I got it to work by running these commands from the terminal after all the other setup stuff... maybe they're not all needed, I dunno, but it worked. (I restarted ComfyUI after I ran them). source ./venv/activate pip install setuptools --upgrade pip install ninja cmake packaging pip uninstall -y torch torchvision torchaudio xformers || true rm -rf ~/.cache/torch_extensions pip install --force-reinstall torch==2.7.1+cu128 torchvision==0.22.1+cu128 --extra-index-url https://download.pytorch.org/whl/cu128 export TORCH_CUDA_ARCH_LIST="9.0"

Adam Adair

I think there is a bad node in the workflow. After the latest comfyui update, I keep getting failure to fetch disconnect at about 3/4 the way through every upscaler run. I suspect it's an incompatible custom node.

Sam Shennan

Who else sees this: I get wierd charcters. I'm au australian with english on my machine Which Wan 2.2 quantization? 1) Q4_K_S ΓÇô GPUs < 12 GB vRAM 2) Q5_K_S ΓÇô GPUs 12-24 GB 3) Q8_0 ΓÇô Best quality, GPUs 24 GB+ Enter 1, 2 or 3: 1 -------- Checking prerequisites -------- Git not found ΓÇô downloading silent installer... % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 9 100 9 0 0 96 0 --:--:-- --:--:-- --:--:-- 100 Then this error: Unsupported 16-Bit Application The program or feature cannot start or run due to incompatibity with 64-bit versions of Windows. Please contact the software vendor to ask if a 64-bit Windows compatible version is available. x

J

@satsFlowMaxi use claude.ai to convert the runpod installers to local installers on whatever iteration of linux you are running and it will do it flawlessly. The one click installers tend to get worse and worse as more time goes on and URLs change and updates break previous nodes etc but for now it still works very well, I did it yesterday

J

Did you ever find a solution? I have the same problem

Virtual Shards

I've been seeing those as well for a few different releases. It's something to do with unicode encoding. I've been ignoring it without any issues.

Scott Beowulf Kaine

I use an rtx 5090, I have zero problems with the entire workflow, other than how I get the video size to auto-change to the correct size of the image instead of having to manually change it. But that's not part of the original workflow, so it's not relevant, hence I'll ask the question here..

Scott Beowulf Kaine

How do I get the video size to auto-change to the correct size of the image instead of having to manually change it?

Dirk Diggler

The videos with a single character lora turn out decent, but when I add more than one the results are not great. I'm unclear on what strength I should be setting the lighting lora, and the additional single lora at for the best results. I've googled and researched for hours, and tested it and it seems like the high noise for lightning needs to be at 3 and the low noise lightning at 1.5. Then when I add the single character lora keeping both low and high at 1 produce good results. When adding a 2nd charcter lora it throws everything off and I cant seem to find the sweet spot in that situation. If you could clarify what the ideal strength for each lora in the power lora loader that would be very helpful.

jared huynh

Hi I have been trying to run this in runpod. However it has been unsuccessful. 10 missing nodes, which I cannot seem to fix. Any assist?

Chris

Hi, I am getting this issue :/ It seems it's not installing. I was hoping the Bat file would save me too much tech know.. -------- Checking prerequisites -------- 7-Zip not found ΓÇô downloading... % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 162 100 162 0 0 703 0 --:--:-- --:--:-- --:--:-- 704 The system cannot execute the specified program. 7-Zip install failed ΓÇô install it manually then rerun this script. Press any key to continue . . . EDIT: I installed 7-zip and it fixed this!! yey! Im trying to figure out how to install this all correctly now.

Boody KD

why is there a fkn typo in ur templates mount path...

SlimShady

This was working fine a few days ago, but now on Runpod it's causing some kind of problem with ComfyUI Manager, and after the initial install (chmod +x WAN2_2-ULTRA-AUTO_INSTALL-RUNPOD.sh ./WAN2_2-ULTRA-AUTO_INSTALL-RUNPOD.sh) there is no way to open up Comfy Manager to restart. If I try to drag the workflow in, it says it's missing all of the custom nodes (even though they've been downloaded) and trying to install the custom nodes through the error message does nothing, and no activity in logs. Super disapointed I subscribed just for this installation guide and it worked for 2 days.

SlimShady

what's the typo? maybe that's why I can't get this working all of a sudden

SlimShady

yes you need those

Foxy

My manager is gone as well. Have tried 3 different 1 click installs on runpod and it always disappears. I tried a manual install and it still didn't pop up.

Aitrepreneur

it's a browser issue, I have that as well, try another browser in private navigation, you'll see it'll work

Kristian Karakolev

When I try with 4090 it works perfectly, when I try with 5090 cuda error, because of the VRAM I think. In run pod.

spline

When run prompt I get error: Prompt execution failed TypeError: NetworkError when attempting to fetch resource.

Nick Sarkisov

I do exactly same things like in the youtube video but get this error: Some Nodes Are Missing When loading the graph, the following node types were not found. This may also happen if your installed version is lower and that node type can’t be found. Fast Groups Bypasser (rgthree) Label (rgthree) FastLaplacianSharpen FastFilmGrain

Mewael Debretsion

This needs to be a pinned comment seriously like wow thank you! Got the 5090 working on runpod cause of you! MVP

SuicuneSol

Hello, Nick. I found the issue. The developer of those node packages (VRgamedevgirl) made some updates. A module called "librosa" must now also be installed otherwise FastFilmGrain and FastLaplacianSharpen will be broken. Execute the command "pip install librosa" after installing Wan 2.2. If you're using Runpod, you would input the above command after the Auto Install for Runpod finishes. Run the command in the same place you do the other commands.

Michel daniel

Unable to install, it tells me there's an incompatibility with 64-bit versions of Windows. Does anyone have a solution?

Robert

this workflow stopped working completly on runpod. again this has costed me a few hours + 8~dollars rebooting/booting new pods. really really sucks. why the fck am i paying for this?!

Lagartija

on Pod i'm getting "The port is not up yet" on Port 3000 NOTE: This is not necessarily an error Check your container logs for the following: Ensure that the applications have completed syncing to /workspace. Ensure that there are no errors. Ensure that the logs say Container is READY! If the container is READY Check your application logs (refer to the README for more information)." however on the UI all ports are green - other ports work fine. Here's my log: Container Started, configuration in progress... NGINX: Starting Nginx service... * Starting nginx nginx ...done. SSH: Setting up SSH... * Starting OpenBSD Secure Shell server sshd ...done. SSH: Host keys: [REDACTED_SSH_KEY_BLOCKS] CRON: Starting Cron service... * Starting periodic command scheduler cron ...done. JUPYTER: Starting Jupyter Lab... JUPYTER: Jupyter Lab started CODE-SERVER: Starting Code Server... CODE-SERVER: Code Server started RUNPOD-UPLOADER: Starting RunPod Uploader... RUNPOD-UPLOADER: RunPod Uploader started PRE-START: Running pre-start script... TEMPLATE NAME: comfyui TEMPLATE VERSION: 2.3.5 VENV PATH: /workspace/ComfyUI/venv SYNC: Syncing to persistent storage started SYNC: Sync 1 of 1 SYNC: Syncing from /ComfyUI to /workspace/ComfyUI, please wait (this can take a few minutes)... SYNC: File system type: fuse SYNC: Using tar without compression for sync SYNC: Syncing COMPLETE! SYNC: Time taken: 0 minutes, 45 seconds VENV: Fixing venv... Fixing venv. Old Path: /ComfyUI/venv New Path: /workspace/ComfyUI/venv Starting ComfyUI ComfyUI started Log file: /workspace/logs/comfyui.log * Restarting OpenBSD Secure Shell server sshd ...done. FILEZILLA: FileZilla SFTP configuration file created at: /workspace/filezilla_sftp_config.xml RCLONE: Updating rclone... 2025/10/01 16:34:56 NOTICE: Successfully updated rclone from version v1.58.1 to version v1.71.1 PYTHON: Checking Python version... Python 3.11.9 ENV: Exporting environment variables... Container is READY! My Pods appreciate the help

John

I am having trouble installing since it says that it is incompatible with 64-bit Windows. Is there a way to fix this?

Kanjscat (edited)

Comment edits

2026-09-03 06:00 where is the WAN2_2-ULTRA_WORKFLOW.json? Because I can't find it anywhere.
2025-10-05 15:51

TheRangor

The post with the workflow is linked at the beginning of this post.

J

Thank you for all the hard work you put into this.

Troy Walton

It's not possible to get this to work on runpod anymore.

Phillip Gimmi

This works fine, the workflows liek the ultimate workflow is broken due to comfyui breaking changes wish we stuck to a specific verison in runpod without breaking chaanges and perhaps forking repos so that they cant change the experience

Son Goku

Love the work! Can we get this updated working again please?

Testtn

Would be great to have this updated, CUDA 12.8, sage attention, non broken nodes, and a fix for the unloading and loading models. I understandably you're going through some stuff. And I hope you're doing better. So when you're back and feeling up to it would love an update. Cheers

Testtn

Not VRAM, it's not CUDA 12.8

Phillip Gimmi

honestly feels a little misleading when you referencing a workflow that does not seem to work out of the box anymore on runpod, one would hope runpod is waxed instead of not usable

Raiyan A

This doesnt work at all!!! I feel like I just wasted my time and money with this unfortunately

mnkjiu

reinstalled with the bat file and got a new error with the DLL Traceback (most recent call last): File "D:\AI Tools\ComfyUI - WAN\ComfyUI\main.py", line 147, in import comfy.utils File "D:\AI Tools\ComfyUI - WAN\ComfyUI\comfy\utils.py", line 20, in import torch File "D:\AI Tools\ComfyUI - WAN\python_embeded\Lib\site-packages\torch\__init__.py", line 281, in _load_dll_libraries() File "D:\AI Tools\ComfyUI - WAN\python_embeded\Lib\site-packages\torch\__init__.py", line 264, in _load_dll_libraries raise err OSError: [WinError 1114] A dynamic link library (DLL) initialization routine failed. Error loading "D:\AI Tools\ComfyUI - WAN\python_embeded\Lib\site-packages\torch\lib\c10.dll" or one of its dependencies. D:\AI Tools\ComfyUI - WAN>pause Press any key to continue . . . this is the first time i saw this. I have updated comfui (with the bat file) and update the nodes manageras well, and use the node installer just in case. any help would be great

Mike J

Can I run these workflows on a Mac? Or do I need Windows?

Taiga

the workflows will probably work on a mac; the one-click installs will not. just drag the workflow onto comfyUI and see if it works.

estreel

I follow each step for RunPod, but it tells me I don't have these nodes. Fast Groups Bypasser (rgthree) Label (rgthree) FastLaplacianSharpen FastFilmGrain

J R

Having an issue on the 5090 via runpod - I've managed to get the manager and film grain node fixed using the advice further below in the comments and it'll open with the manager in an incognito window, but now hit the following error: CUDA error: no kernel image is available for execution on the device CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1 Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions. Advice welcome

J R

I've got the error log but it's a LOT longer!

J R

This was really helpful thanks

J R

Is it possible to move the GPU type on the pod I setup on 5090, or it better to just start again?

J R

Thanks - this worked

J R

Trying it now... IT WORKS!!!

Simon

Always when i follow your runpod installation the Manager is gone and i cant start the workflow. I did all step by step. Update all first, and then use your script. then the UI is broken and misses the manager.

ArslanRuslan

"1 CLICK INSTALLS" often have some afterhustle. last one turned into 3 hrs. And then another 3 hrs. I dunno why you calling them 1 click install.

Dan

It was a 1 click install when they posted these bat files. Since there have been some updates it seems to require some tweaking. I just did a fresh install today and it went mostly smooth because I already had the changes made to allow the python scripts to download properly. Sucks but that's just the nature of something that changes so quickly and has so many updates.

Dan

After my reinstall I had a few issues with this as it currently sits. This is for locally run workflows on a 5080. First I did a complete wipe so starting from scratch. My cuda version needed matched up using the tip they already posted in the main body. I could not get comfy UI to load until I ran the python dependencies update. Even though it showed errors during the update it made it so I could load comfy UI. I did have to go download the huggingface safetracer that's listed in the readme file. No problems running multiple workflows after that.

Ash

had installed a couple of times as well, but with the new ones, after i deleted and tried installing again it failed to work. Perhaps requesting for a new fresh install bat for WAN with the new addons would help.

Ng Luke

does rtx 5060ti 16GB work with this?

Caleb Williams

when installing from "WAN2_2-ULTRA-COMFYUI-MANAGER_AUTO_INSTALL.bat" I keep getting "download failed" at around 1200MB. what's going on??

troy X

What directory do you run your commands?

MAN MADE ALPHA

When loading up json file in runpod it says Im missing a bunch of nodes. I tried updating comfyui through manager but it didn't seem to work. Any help would be appreciated. ... When loading the graph, the following node types were not found: easy cleanGpuUsed CLIPLoaderGGUF UnetLoaderGGUF WanVideoNAG easy clearCacheAll FastLaplacianSharpen FastFilmGrain WanFirstLastFrameToVideo VHS_VideoCombine WanImageToVideo VHS_LoadVideo VHS_VideoInfo EmptyHunyuanLatentVideo Nodes that have failed to load will show as red on the graph.

floflo tutu

there is a problem on runpod, after i update comfyui, if i restart the comfyui, and i refresh the page it cannot load the page anymore, i have instead of comfyui: error: 403 forbidden.

Aitrepreneur

Just add a / at the end of the url then press enter