Home Creators Posts Import Register Favorites Logout

Downloads

Content

Tencent HunyuanImage 2.1 text-to-image app with a refiner and image refinement tab; the distilled model makes an image in about 5 seconds on an RTX 6000 PRO.

Enhanced Gradio app based on the official pipeline, with 1-click installers for Windows, RunPod and Massed Compute, a resumable model downloader and Flash Attention 2.8.3 for all GPUs.

Patreon exclusive posts index to find our scripts easily, Patreon scripts updates history to see which updates arrived to which scripts and amazing Patreon special generative scripts list that you can use in any of your task.

Join discord to get help, chat, discuss and also tell me your discord username to get your special rank : SECourses Discord

Please also Star, Watch and Fork our Stable Diffusion & Generative AI  GitHub repository and join our Reddit subreddit and follow me on LinkedIn (my real profile)

======

Latest Zip File : HunyuanImage_2_1_v5.zip

  • Full tutorial video published : https://www.youtube.com/watch?v=dNeA5mJ36hA

  • You can read more here : https://huggingface.co/tencent/HunyuanImage-2.1

  • Models will be downloaded inside ckpts sub-folder

  • Distilled model works with 8 steps and really faster but i find quality lower

    • It takes total 5 seconds per image with Distilled model on RTX 6000 PRO

  • I don't know the very best parameters yet but Default config should work fine

  • The model downloader I made is extremely robust and will download all models into accurate sub folders

    • If you get any error just run downloader again and it will skip accurately downloaded ones and fix issues and resume

  • Check all features of the app on the interface - pretty easy to use and convenient

  • All generated images will be saved inside outputs folder

  • I recommend using Massed Compute for now until VRAM requirement gets lower

  • Fully supports all GPUs with automatically installed Flash Attention 2.8.3 (compiled by me for both Windows and Linux with all GPUs support), Torch 2.8 and CUDA 12.9 (like A100, H100, B200, RTX 6000 PRO, etc.)

11 September 2025 V4

  • Full tutorial video published : https://www.youtube.com/watch?v=dNeA5mJ36hA

  • Some bugs fixed

  • Installers improved

  • Refiner is currently working perfect and I recommend to set 8 steps

  • We added Image Refinement tab that you can use to refine existing images

  • Now when you enable Use Refiner, it will save both non-refined version + refined version and return both

  • Just run Windows_Install_or_Update.bat or use start command in RunPod or Massed Compute (they are set to auto update)

Windows Requirements

How To Install

  • Download latest zip file from the attachments and extract files wherever you want to install

  • Double click windows_install .bat and let it install and download necessary models

  • Requires 80 GB GPUs right now

Massed Compute (Recommend Cloud) :

RunPod (Cloud):

Patreon exclusive posts index to find our scripts easily, Patreon scripts updates history to see which updates arrived to which scripts and amazing Patreon special generative scripts list that you can use in any of your task.

Join discord to get help, chat, discuss and also tell me your discord username to get your special rank : SECourses Discord

Please also Star, Watch and Fork our Stable Diffusion & Generative AI  GitHub repository and join our Reddit subreddit and follow me on LinkedIn (my real profile)

======

Latest Zip File : HunyuanImage_2_1_v4.zip

  • You can read more here : https://huggingface.co/tencent/HunyuanImage-2.1

  • Models will be downloaded inside ckpts sub-folder

  • Distilled model works with 8 steps and really faster but i find quality lower

    • It takes total 5 seconds per image with Distilled model on RTX 6000 PRO

  • I don't know the very best parameters yet but Default config should work fine

  • The model downloader I made is extremely robust and will download all models into accurate sub folders

    • If you get any error just run downloader again and it will skip accurately downloaded ones and fix issues and resume

  • Check all features of the app on the interface - pretty easy to use and convenient

  • All generated images will be saved inside outputs folder

  • I recommend using Massed Compute for now until VRAM requirement gets lower

  • Fully supports all GPUs with automatically installed Flash Attention 2.8.3 (compiled by me for both Windows and Linux with all GPUs support), Torch 2.8 and CUDA 12.9 (like A100, H100, B200, RTX 6000 PRO, etc.)

11 September 2025 V4

  • Full tutorial video published : https://www.youtube.com/watch?v=dNeA5mJ36hA

  • Some bugs fixed

  • Installers improved

  • Refiner is currently working perfect and I recommend to set 8 steps

  • We added Image Refinement tab that you can use to refine existing images

  • Now when you enable Use Refiner, it will save both non-refined version + refined version and return both

  • Just run Windows_Install_or_Update.bat or use start command in RunPod or Massed Compute (they are set to auto update)

Windows Requirements

How To Install

  • Download latest zip file from the attachments and extract files wherever you want to install

  • Double click windows_install .bat and let it install and download necessary models

  • Requires 80 GB GPUs right now

Massed Compute (Recommend Cloud) :

RunPod (Cloud):

Related posts

All posts in this topic: AI Image Generation and Editing Apps · New to SECourses? Start Here: Best and Most Popular SECourses Posts

Files

Comments

cool1

I doubt many people here will have a PC with an 80 GB or more GPU though so it's not something many people are likely to be able to use on their PC. It seems it's only likely to be usable for people paying extra to use cloud PCS . Not until they can lower it's requirements with maybe quantized models. And it says the current models are in a .ckpt folder. But .ckpt is an outdated/unsafe format. Hopefully it can be updated to use quantized models in the .safetensors format (safetensors has been available for years I think) so it can be used with GPUs that people are likely to to own (such as around 24 GB or less, 32 GB for those who have the best "consumer" GPU.). It's a shame if the refiner adds noise and affects quality but I assume it can be fixed, but the refiner probably adds complication to it (such as when using loras, which might affect them).

Stevie Tee

This looks great, thank you. It installed fine however I'm receiving the following errors because the C++ compiler cannot find the omp.h header file, which is required for OpenMP but I'm certain the correct individual components are installed with the correct Visual Studio version: FlashAttention v3 not found, falling back to v2. Loading last used config: Default ✓ Set last config to: Default 🚀 Starting HunyuanImage Gradio App... 🔧 Deferred model loading: Models will be loaded on first use with user-selected VRAM settings 🎯 Default model type: Regular 💻 Device: cuda 🌐 Share mode: Disabled ⚠️ Make sure you have the required model checkpoints in the 'ckpts' folder! * Running on local URL: http://127.0........ * To create a public link, set `share=True` in `launch()`. Loading pipeline for the first time with user-selected VRAM optimizations... Loading HunyuanImage pipeline (distilled=False)... DiT offloading: True Reprompt offloading: True Refiner offloading: True Model compilation: False 📊 VRAM detected: 12.0GB 2025-09-10 14:05:09.228 | INFO | hyimage.diffusion.pipelines.hunyuanimage_pipeline:_load_models:294 - Loading HunyuanImage models... 2025-09-10 14:05:15.861 | INFO | hyimage.diffusion.pipelines.hunyuanimage_pipeline:_load_vae:203 - ✓ VAE loaded 2025-09-10 14:09:41.004 | INFO | hyimage.diffusion.pipelines.hunyuanimage_pipeline:_load_dit:165 - ✓ DiT model loaded 2025-09-10 14:09:49.915 | INFO | hyimage.diffusion.pipelines.hunyuanimage_pipeline:_load_byt5:286 - ✓ byT5 glyph processor loaded `torch_dtype` is deprecated! Use `dtype` instead! Loading checkpoint shards: 100%|█████████████████████████████████████████████████████████| 5/5 [00:00<00:00, 19.50it/s] 2025-09-10 14:10:38.709 | INFO | hyimage.diffusion.pipelines.hunyuanimage_pipeline:_load_text_encoder:191 - ✓ HunyuanImage text encoder loaded ✓ Pipeline loaded successfully ✓ Pipeline loaded with user settings Settings updated without pipeline reload ============================================================ 🖼️ HunyuanImage Generation Task ------------------------------------------------------------ Prompt: A beautiful landscape with mountains Negative Prompt: blurry, low quality Guidance Scale: 3.5 CFG Mode: MIX_mode_0 Guidance Rescale: 0.0 Shift: 4 Seed: 269001 Use MeanFlow: False Use byT5: True Image Size: 1024 x 1024 Sampling Steps: 50 ============================================================ Denoising: 0%| | 0/50 [00:05

Stevie Tee

Apologies for posting the error here, I've just noticed the link to the Github issue page.

SECourses

your logs has no error. it is just working so far

John McCaffrey

How does one go about quantizing a model? is it something that will just eventually happen and someone will work on it? or is something that you can actually do yourself and if so how? I only have a RTX 5090 32GB

SECourses

yes will happen for sure. stay tuned i will publish tutorial for it hopefully

EvanGuo

Lora training?