[Bug]: a huge animation image glitch

[Mistermango23]

Is there an existing issue for this?

• [X] I have searched the existing issues and checked the recent builds/commits of both this extension and the webui

Have you read FAQ on README?

• [X] I have updated WebUI and this extension to the latest version

What happened?

I wanted to generate an animatediff gif but suddenly a huge glitch image comes out.

Steps to reproduce the problem

1. Go to ....
2. Press ....
3. ... image glitch

What should have happened?

No wonder it became an animation glitch. There is a new update for stable diffusion webui and sdxl, and that's why a glitch came out. and of course the models can't load, I assume.

Commit where the problem happens

webui: 1.6.0 extension: animatediff

What browsers do you use to access the UI ?

No response

Command Line Arguments

No

Console logs

in the browser console, it doesn't say it for me. I only use colab on google and there are no errors in the colab with code, unless the models are not set correctly, and of course this can't read or recognize the model because the model only shows glitches in the animated image.

Additional information

[continue-revolution]

I need your prompts and a screenshot of your screen which includes your configuration.

However, most likely it's a problem of the motion module which I can do little to.

[Mistermango23]

But wait, first I have to prepare something from colab and then I'll give you the prompt and a clip screenshot when I've done all that.

[Mistermango23]

@continue-revolution Here is the prompt and the clip screenshot from the video. Prompt: uploaded on e621, (by Whooo-Ya, by Sskomu (artist), by Kakure Eria), solo female (((rabbit Judy Hopps) grey body, wet fur, clear green eyes)), (wear police uniform, jacket, pants, belt:1.3), (small breasts), ((detailed fluffy fur)), (looking at viewer, three-quarter portrait, three-quarter view), BREAK, ((standing at city street on sunset raining day)), (rain, fog, mist, cyberpunk, neon light), (detailed background, depth of field, half body shadow, sunlight, ambient light on the body), (masterpiece, best quality, ultra realistic, 4k, 2k, (high detail:1.3), (3d (artwork):1.2), blender (software), (soft focus:1.2), ray tracing, (unreal engine:1.3), absurd res)

Negative Prompt: low res, lowres, blurry, bad anatomy, letterbox, deformity, mutilated, malformed, amputee, watermark, signature, unusual anatomy, username, sketch, monochrome

checkpoints models: https://huggingface.co/Mistermango24/Yiffymix_V3.4 or https://civitai.com/models/3671/yiffymix

Vae models: https://huggingface.co/Mistermango24/YiffymixVAE

clip video: https://www.youtube.com/watch?v=Aere7uIG5qk

[continue-revolution]

ok. you may wait some time before I can test and respond. I will use the original repo to see whether something is wrong.

• If the original repo generate gliches, then I can do little about this
• Otherwise, I will have to look into the detail of A1111, which may take another amount of time.

[Mistermango23]

@continue-revolution Now I know why the models can't load. It's because of "A1111" research and that's why the models can't load. I tested it on comfyui and I'll show you the results

[Mistermango23]

Here is a result from comparison:

A1111:

ComfyPlus:

Comfy:

That's strange, it looks somehow the same. Is it because of the prompts or the models?

[Mistermango23]

@continue-revolution I'm pretty confused. For me on the comfyui everything looks fine, but when it comes to SD webui an image from animation is generated then something looks different. No idea! Like I said, I'm pretty confused about generation. I don't know! lol

[zappityzap]

What happens if you use a prompt without BREAK in it?

[zopi4k]

"cat" : 1/75 tokken

"cat BREAK dog" : 76/150 tokkens

"cat AND dog" : 1/75 tokken

i think it's the problem ... DONT PASS 75 tokkens ! ^.^)v

[Mozoloa]

I also get the deepfried problem, without using any prompt, also the animation is non existent

https://github.com/continue-revolution/sd-webui-animatediff/assets/20817233/61060c31-b491-45b1-801f-5e35cdb21b7b a beautiful woman 3/75

Okay so actually this may have to do with the length, everytime I go 8 frames, it goes crazy, but 16 is fine

https://github.com/continue-revolution/sd-webui-animatediff/assets/20817233/52e4a591-077f-4d16-a727-c13f74ff1149 16

https://github.com/continue-revolution/sd-webui-animatediff/assets/20817233/a7f9efd2-3708-4b83-ad2d-1fd03125dac4 8

same prompt & seed "a cat" with SD1.5

[zopi4k]

Ho !!! ... do you have try retart the consol for web-ui ? when a message error is ritten ... it's impossible to generate other anim for my part, I always restart for get a good image without glitchs.

[mrmeseeks23]

I am having same issue. I had two great renders. Then everything began to come out completely broken (complete fractured noise appearance). Then when I tried to deactivate I began to receive a CUDA error.

[rkfg]

YiffyMix seems to not play well with animatediff and also various loras (try any anime lora, it comes out deep fried with very low details even with low lora weight). Perhaps because it went too far from the both base 1.5 and NAI models, or maybe its text encoder changed too much. I only have this issue with this model.

[UltraBlackLinux]

https://github.com/continue-revolution/sd-webui-animatediff/assets/20817233/a7f9efd2-3708-4b83-ad2d-1fd03125dac4 8

hmm. all my generated images look like the last frame of that gif. Maybe the noise like in #286 and #178 is related?

[hyperlogin21]

im having the same issue. hm..prompts are under 75 tokens