search

Kahsolt / stable-diffusion-webui-prompt-travel

Travel between prompts in the latent space to make pseudo-animation, extension script for AUTOMATIC1111/stable-diffusion-webui.
The Unlicense
252 stars 22 forks source link

Traceback (most recent call last): #34

Closed mazz3d closed 1 year ago

mazz3d commented 1 year ago

Hello! Is anyone else experiencing errors when using the script with SDXL 0.9? image I've been having a great time creating videos using Prompts Travel in the 2.1 models.

Kahsolt commented 1 year ago

I'm not familiar with SDXL 0.9, even don't know if ptravel will work with the 2.1 models.... :lol

But I've search the class name DictWithShape in the sd-webui repo (I'm on A1111's stream) and found nothing. It seems to need extra code support to run this SDXL 0.9?

If I'm able to get these stuff in your env, maybe I could help to support this. Point me away please?

mazz3d commented 1 year ago

Yes, it works wonderfully with the 2.1 models. The transition from one prompt to the next is incredibly smooth. 💯 :) sorry what do you mean of (If I'm able to get these stuff in your env, maybe I could help to support this. Point me away please?) I would love to help if I could as well. And perhaps it would be wise to wait for the official XL1.0 release on the 18th?

Kahsolt commented 1 year ago

Nice!! Feel relieved that this extension could outputs something meaningful :) And sorry for the late reply, there's some time lap due to the timezone.

I mean to support this SDXL ?.?, I need to reproduce the error you've just got in my develop env. So please tell me what the checkpoint file you actually use, where can I get it. And what the necessary extensions you activate in order to run this SDXL (if there's any)

mazz3d commented 1 year ago

Nice!! Feel relieved that this extension could outputs something meaningful :) And sorry for the late reply, there's some time lap due to the timezone.

I mean to support this SDXL ?.?, I need to reproduce the error you've just got in my develop env. So please tell me what the checkpoint file you actually use, where can I get it. And what the necessary extensions you activate in order to run this SDXL (if there's any)

Hi, I apologize for the delayed response. I just noticed your reply now :D I was referring to SDXL 0.9, the latest update of SD. Here are some screenshots that might be helpful. image image image I hope this information proves helpful.

Kahsolt commented 1 year ago

Yes, enough, thx. I'll check it next week as soon as possible. Currently busy with a coding competition :)

Kahsolt commented 1 year ago

Latest news, I've just downloaded the SDXL 1.0 and starting to code :) Two days later should be available.

Kahsolt commented 1 year ago

Now ready for SDXL v1.0 model, please try v3.1~

I've run few examples to find that the generated frames are not that consecutive like those v1.4 models. This might be due to the text encoder (CLIP) in SDXL v1.0 has a much higher dimension 2048, contrasted with 768 in legacy models, so more information could be compressed into the model, but subtle change of input params could lead to a jump in the outputs :(

Please have a try and report the performance~