Spaces:

---

LPX55

/

HunYuan-Keyframe2VID-Control-Lora

like 9

Runtime error

Hello Huggingface wizards! AFAIK, this is the only KF-based IMG2VID space on HF Spaces for the HunYuan Video model.

This research and ControlNet model was implemented by our non-profit organization AI Without Borders as a part of an ongoing bigger project "OpenSight" which aims to democratize and open source SOTA deepfake detection models.

However, we thought it was a pretty cool space and decided to open it to the public last week. The problem is ... Hunyuan be fat, and while we commit to seek more ways of better inferencing and optimizations, it's extremely hard to do when it just bare loads into the 40gb provided by Zero spaces.

At the moment, while the space technically "works" you'll be lucky to get a dozen frames at a very low inf. step level. Most of the times it will raise error mid-inference because of VRAM.

If a grant is ... granted, this space will be moved out of my personal account into the AIWB org space, which will be upgrading to enterprise soon. Also highly interested in exploring deploying similar spaces for Wan2.1 , which will probably be done in the same space.

Huggingface gods, what do you say? Yay or nay?

(Could be seen as "company project" ... though we have never made a single dime 🥹)