Prompt following problem?

#1
by Sikaworld1990 - opened

Refeence imgaes from https://huggingface.co/alibaba-pai/Wan2.2-VACE-Fun-A14B
Prompt as u can see in the wf: a girl holding a camera in her hand is dancing.

QuantStack org

hmm this seems like either its a bad gen or the implementation in comfyui is not perfect (or the model itself has issues)

hmm this seems like either its a bad gen or the implementation in comfyui is not perfect (or the model itself has issues)

Couldnt get a satisfying result even with different seed and modfied wf.

QuantStack org

Are you using a video as input?

Are you using a video as input?

No 2 pictures, start and end frame. WF should be included in the sample vids

QuantStack org

Hmm, might be that fun vace doesnt like that? Or the implementation simply doesnt work that well for it.

QuantStack org

i

wsbagnsv1 changed discussion status to closed

Hmm, might be that fun vace doesnt like that? Or the implementation simply doesnt work that well for it.

I took the reference images from the original repo https://huggingface.co/alibaba-pai/Wan2.2-VACE-Fun-A14B

But you're using the wrong workflow to achieve your goal. You're using a first-frame-last-frame approach, which isn't the most effective method for what you're trying to accomplish.

wsbagnsv1 changed discussion status to open
QuantStack org

Sry, didnt want to close this lol 😅

But you're using the wrong workflow to achieve your goal. You're using a first-frame-last-frame approach, which isn't the most effective method for what you're trying to accomplish.

Which workflow do U recommend instead?

"Classic start/endframe works very well!

start:
424292759-6c301578-56ae-45c7-8d1c-9ac5f727bf53.png

end:
424292838-97de3844-e974-4be9-9157-0785c564574d.png

QuantStack org

You should use the Wan2.2-Fun-A14B-InP, not the VACE

You should use the Wan2.2-Fun-A14B-InP, not the VACE

Thx

"Classic start/endframe works very well!

QuantStack org

Wan2.2-Fun-A14B-InP
Use a image as start and end to create a video.

Wan2.2-Fun-A14B-Control and Wan2.2-VACE-Fun-A14B
Use reference (images) and video (control) to create a video controlled by the control input.

Wan2.2-Fun-A14B-Control-Camera
Use reference (image) and the camera movement do you choose to create a video controlled by the movement you select.

YarvixPA changed discussion status to closed

Wan2.2-Fun-A14B-InP
Use a image as start and end to create a video.

Wan2.2-Fun-A14B-Control and Wan2.2-VACE-Fun-A14B
Use reference (images) and video (control) to create a video controlled by the control input.

Wan2.2-Fun-A14B-Control-Camera
Use reference (image) and the camera movement do you choose to create a video controlled by the movement you select.

"Classic start/endframe works very well!

start:
424292759-6c301578-56ae-45c7-8d1c-9ac5f727bf53.png

end:
424292838-97de3844-e974-4be9-9157-0785c564574d.png

This is generated with Vace Fun using start and endframe! Will try the other models as comparison thx!

Sign up or log in to comment