EMOTION TRANSFER, eyelines match , High res character, speed motion transfer , object coherence, multiple camea in VACE? with upscaler?

#1
by t2das - opened

So for 2.2 VACE When can we expect a estimated release?
As Vace dont play good with multiple controlnets. request would be to do develop properly so it can highly be useable. Most production demand a 1080p or 1440p.

VACE have potential to take things to next level AI .

Multiple reference for larger smoother workflow is a long time request. atleast in steps to accomplish lot more with AI.
For just 1 image in a story of a scene which need to be super detailed high quality 1440p, 

here are the workflow i am thinking that if you can explore and discover solutions.

T & I workflow to VIdeo.

  1. Take character ref [1,2](king character sheet & fighter character sheet), put them in an scene ref [3] (arena)
  2. they are located at a certain position in the space ( inpaint inside arena). Blend it well.
  3. both have a specific pose reference [4,5] ( king is in attacking position, fighter is blocking position)
  4. while they both holding few things according to story ref [6,7] ( king have a specific sword he won , Fighter have specific shield)

  5. While they are giving a facial expression of ref [vid 8, vid 9], (king is angry, fighter is inner turmoil to not fight king himself]
  6. They are given Specific Movement reference full body ( Video ref 10)
  7. they are given specific Acting video , emotion facial reference. Dialogues & lip sync ( Video ref 11)
  8. They are given Video Ref Camera movement in the scene ( Vid ref 12) or text to Camera movement. 

to capture whole scene at once in multiple angle.
  9. Reprocess final for face for clarity on going distance from camera, and high speed motion, blurry face and motion. Match Eyes lines when moving, looking and other. No dead expression.

references: Acting expression, movement Pre-possesor, Cloth details, Position details, 

instead of 2d based , make 3d based pre-processors which can object tracks multiple perfectly and inject data.
blending: Lights when all are migrated to scene, scene with physics of water, air, VACE black and white for specific object animation tagging would be great but also multiple targeted inject will speed things up . 

All need to blend perfectly. . this is help all the other Ai vid and fimmakers folks a lot. it will be less wastage and more Eco friendly & faster precise output ,time saved using AI.

so what is this vace and the fp8 version on kijai ? i have tried in comfyui but just get a blurry mess - can someone get a comfyui workflow going?

so what is this vace and the fp8 version on kijai ? i have tried in comfyui but just get a blurry mess - can someone get a comfyui workflow going?

whered you find the gguf link? i only saw the full scale models at 34gbs a piece

Alibaba-PAI org

Sorry, we’re not the official VACE team. This model is our Control model’s implementation of the VACE architecture, but it’s different from the official VACE. Also, we’ve noted your request and, where appropriate, we’ll incorporate some of it into the next version.

Please let them know. I guess there in no official VACE 2.2 release yet so no hugginface page of 2.2 version. Is there any specific email id for sending feature requests? I will forward that team the same requests.. thank you.

Sorry, we’re not the official VACE team. This model is our Control model’s implementation of the VACE architecture, but it’s different from the official VACE. Also, we’ve noted your request and, where appropriate, we’ll incorporate some of it into the next version.

  • Does 'FUN VACE 2.2' allow inpainting/outpainting using a gray (rgb 127,127,127) mask the same way as with VACE 2.1?

  • Can there be multiple areas masked on a frame, or does the mask have to be a contiguous single area?

Thanks!

Sign up or log in to comment