Text encoder and VAE model
Compared to the original Qwen3‑VL‑8B model, has your text encoder been fine‑tuned?
And is it necessary to use your fine‑tuned text encoder instead of the original Qwen3‑VL‑8B model?
Is the VAE model also fine‑tuned, rather than directly using the Wan 2.1 VAE?
Yes, our text encoder has been fine-tuned, and it is necessary to use our fine-tuned text encoder. However, the VAE from Wan 2.1 has not been fine-tuned and can be used directly as it is.
@makisekurisu-jp will you be hosting the ComfyUI weights for the JoyImage series of models? If so, to avoid any potential misunderstanding, I will delete our repository here once you have done so.
@makisekurisu-jp will you be hosting the ComfyUI weights for the JoyImage series of models? If so, to avoid any potential misunderstanding, I will delete our repository here once you have done so.
Sorry, I’m not a member of Comfy Org. You may consult @kijai for assistance.
ohh i have misunderstood
@huangfeice https://huggingface.co/Comfy-Org/JoyAI-Image-Edit/tree/main
is there a difference to the files here?
@huangfeice https://huggingface.co/Comfy-Org/JoyAI-Image-Edit/tree/main
is there a difference to the files here?
The weight files are identical. As is customary, the repository for weights in Comfy format is maintained by Comfy-Org. However, given that our current repository has already gained some attention, I will not delete it, and I will include a link to Comfy-Org repo in the README.