-
This breaks everything, so I feel like I must have something wrong on my end, since I don't see anyone else discussing it after browsing all the pull requests, bug reports, and discussions. It works fine if I go back to a previous version of the plugin. No matter how I set strength_schedule every generated frame is completely new with no relation or weight given to the previous frame, like you would expect if it were set to 0. Here's how to reproduce my issue: What I expect (and get on commit '22b8d3d'): A picture of a cow in frame 0, the identical picture in frame 1, a very visually similar weird cow-kitten hybrid in frame 2, and something just a shade more kitten-like and less cow-like in each subsequent frame, probably reaching full-kittenhood around frame 20-30ish depending on the model. What I get: A picture of a cow in frame 0, a completely different unrelated picture of a cow in frame 1, a total kitten not at all cow-like in any way in frame 2, and a completely different kitten in every subsequent frame. I've been having this issue for days and through various updates to both this plugin and to automatic1111. I've tried every setting I can think might be even partially related and even wiped out all my settings and tried it as a fresh install. There are no errors reported anywhere in either case and it does generate all frames and make a video, but on recent versions it's just everey frame is utterly unrelated to the previous. Am I the only one? I'm not doing anyhting weird (I am on linux though) -- just a stock automatic1111 install. Did something change that I've been doing wrong all along perhaps? |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments 1 reply
-
Hi @ScribbleJ, The issue you report sounds like an issue we had quite a few times in the past, and it was related to non-default settings in auto1111. Here's one example that was uploaded by a member of the Deforum discord server: 20221208160723.mp4Original message. I am a bit confused as to what happens at your pc as you explicitly wrote that you tried a fresh install with the default values. Could you please upload your For the protocol, when I tested using your prompts and default params just with str_sch of 0.95 I got this: 20230105210340.mp4If you want to try my config.json, it's down here, just make sure to remove the .txt extension: It might also be worth to note that I'm testing on windows. |
Beta Was this translation helpful? Give feedback.
-
@hithereai I want to thank you so much for taking time to look into this in such depth. Seriously -- do you have a donation page up anywhere? I'm glad you did, and it helped me a lot, but I realize now that I'm an absolute idiot. After reading your response I completely deleted my config and my python venv and started over, and it worked as you say -- then I realized that I always turn on the option to do the number of iterations specified rather than to do the percentage that matches the strength. Yes -- exactly like it says not to do in the README. In the commit I mentioned and prior to that, it works fine for me that way. In subsequent versions, it does not. It makes a drastic difference to the quality of the output, especially when you're trying to use strength values that are so far from center. I'd have to set the iterations to 300-400 just to get a reasonable number at a strength of 0.9 and the UI doesn't even allow it that high. If you (or anyone) know how I can patch this so I can continue to use strength and iterations independently, please let me know. Meanwhile, I'll start digging though the diffs myself to see if it's something I can solve. And until then I guess I'll keep rolling on the old version. Edit: Nevermind, I now see there's an option to schedule the iterations! That should probably do fine. Thanks again for checking into it. I'm so sorry that you spent the time on a moron. |
Beta Was this translation helpful? Give feedback.
@hithereai I want to thank you so much for taking time to look into this in such depth. Seriously -- do you have a donation page up anywhere?
I'm glad you did, and it helped me a lot, but I realize now that I'm an absolute idiot. After reading your response I completely deleted my config and my python venv and started over, and it worked as you say -- then I realized that I always turn on the option to do the number of iterations specified rather than to do the percentage that matches the strength. Yes -- exactly like it says not to do in the README. In the commit I mentioned and prior to that, it works fine for me that way. In subsequent versions, it does not.
It makes a drastic differen…