Instructions to use Comfy-Org/Qwen-Image-2.1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusion Single File
How to use Comfy-Org/Qwen-Image-2.1 with Diffusion Single File:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
NVFP4 please
I have a RTX 5050 8GB laptop, a NVFP4 would really help for my low vram and also take advantage of blackwell architecture.
Try int8 first. Better quality, and speed is ok
I have a RTX 5050 8GB laptop, a NVFP4 would really help for my low vram and also take advantage of blackwell architecture.
I am also in desperate need of the NVFP4.
Int8 is worth a try before waiting for NVFP4. On my RTX 5060 Ti 16 GB (also Blackwell), qwen_image_2.1_int8_convrot does 1024x1024 at 25 steps in 17.2 s, versus 39.7 s for bf16. Peak VRAM was 14.3 GB, so on 8 GB ComfyUI will offload part of the model to system RAM. That still works, but it is slower, and your RAM becomes the limit. Use the qwen3vl_8b_int8_convrot text encoder too, and the more system RAM you have, the better. Laptop GPUs also run at lower power than desktop cards with the same name, so expect slower times than mine. More on that here: https://vramready.com/guides/laptop-gpus/