kabachuha / sd-webui-text2video

Auto1111 extension implementing text2video diffusion models (like ModelScope or VideoCrafter) using only Auto1111 webui dependencies
Other
1.28k stars 108 forks source link

[Bug]: more frames = more noise #213

Closed andzejsp closed 1 year ago

andzejsp commented 1 year ago

Is there an existing issue for this?

Are you using the latest version of the extension?

What happened?

RTX3090

Started 256x256 @24 15fps

rendered 2s video close to prompt (didnt really matter much wich model was used, mostly zeroscope)

Used the same seed and went with 125 frames, result was noise.

Tried even increasing fps but with the same result.

How do you make longer videos?

How do you get something close to prompt? Text2img prompts are not working right, the video dont follow the prompt.

Is it the model problem?

I cant get anything close to what was shown in the showcase in the description of the repo.

Steps to reproduce the problem

Increase frames get noise

What should have happened?

No response

WebUI and Deforum extension Commit IDs

webui commit id - txt2vid commit id -

Torch version

latest i guess

What GPU were you using for launching?

RTX3090

On which platform are you launching the webui backend with the extension?

Local PC setup (Windows)

Settings

Default except for the frame counts

Console logs

None issue

Additional information

No response

github-actions[bot] commented 1 year ago

This issue has been closed due to incorrect formatting. Please address the following mistakes and reopen the issue: