Model Request: AniSora-V3 & Question on Model Adaptation
Hello Kijai,
First of all, thank you for your amazing work on the ComfyUI-WanVideoWrapper.
I'm writing to kindly request support for the new AniSora-V3 model. It looks very promising and would be a great addition.
Additionally, I would love to learn how to adapt other models myself. Could you please provide some guidance on:
- The recommended process for quantizing new models for the wrapper?
- How to adapt other WanVideo-based models to make them compatible?
Any help or documentation you could point me to would be greatly appreciated.
Thanks for your time and consideration!
I already added AniSorav3 here: https://huggingface.co/Kijai/WanVideo_comfy_fp8_scaled/tree/main/I2V/AniSora
I'm trying to keep that repo cleaner and only have the fp8_scaled models.
The method I'm using is a bit experimental so I don't have a proper solution for every case, new models often have new layers that need to be taken into account.
Oh, okay, thanks so much for adding AniSora-v3!
I understand the method is still experimental. On that note, would you consider including the specific quantization script you used for each model within its directory? It would be a great resource for those of us trying to learn the process, even if it's not a one-size-fits-all solution.
Oh, okay, thanks so much for adding AniSora-v3!
I understand the method is still experimental. On that note, would you consider including the specific quantization script you used for each model within its directory? It would be a great resource for those of us trying to learn the process, even if it's not a one-size-fits-all solution.
I can consider that, I'd just want to clean it up etc. as it's embarrassing code and unintuitive to use...
Oh, okay, thanks so much for adding AniSora-v3!
I understand the method is still experimental. On that note, would you consider including the specific quantization script you used for each model within its directory? It would be a great resource for those of us trying to learn the process, even if it's not a one-size-fits-all solution.
I can consider that, I'd just want to clean it up etc. as it's embarrassing code and unintuitive to use...
thank you for considering it! We're really looking forward to it. And thanks again for your amazing project and the models!
Any chance you'd add the FP16 version? I have 48GB vram so it should fit. Thanks.