Here we go. There was quite some changes to the code for this one.
Most of the changes are self explanatory, some bug fix or more options to select.
The main change is that this version does not contain models. You need to download them from here:
Models:
https://mega.nz/folder/2x1CBRJL#MzMfyijznMs-2t-VqBXYPQ
Driver Mirror:...
I was rushing to complete the next build of SD today, but ran out of time. Almost sure will be ready tomorrow.
Before writing more about it, just want to share that the new model for Rife-App is almost ready (check video)
This model definitly is the best one yet, and is still not finished, so it should be a little better than the example video above.
Here the original footage before being interpolated by Rife-App.
2022-12-07 01:54:48 +0000 UTC
View Post
Got Dreambooth to work with the new 2.0 model.
Right now there is quite a few types of models for SD.
Regular, 2.0, inpainting, Upscale, Depth2Img.
I'm redoing some of the code so is easier to handle all those types + future models.
I'm also taking a look on my TODO list to add some of suggestions in there. There was big updates after big updates (dreambooth, 2.0 models) so its been some time I was able to add some stuff from the TODO list.
Edit: Also, I prob...
Not really a regular build, this is just a hack so everyone can test the 2.0
It probably broke some options, but at least it render Txt2Img with 2.0
It can work with Xformer OR Attention Slice Disabled (So it wil use more Vram if not using Xformer)
There is only 2 models in this Hack:
Stable Diffusion 2 Base:
Stable Diffusion 2:
Gave it a quick read about it. There seen to be some neat new addictions, like upscaler and depth2img, with should be able to generate images with an alpha channel? Well, the link can explain it better than me.
Now lets make this work with the GUI 😋
SD 2.0 was trained with a new CLIP model and a dataset without celebrities and artists names. Those modifications probably will make 2.0 have different prompts that give good results.
V2 prompts different and will take ... 2022-11-24 11:05:14 +0000 UTC View Post
Current tutorial for the Stable Diffusion GUI:
Will keep updating the files to try to improve them in the future.
Prompt Examples:
https://docs.google.com/document/d/1f2vAJnwaJw4KisnHM_r8i0e7nWuGK1Wvc81o1U87NrY/edit?usp=sharing
Create Inpainting/Outpainting:
Hi all.
Ok, first of all, dreambooth training use a lot of crazy new libraries to train it under 12vram. So if the code runs in this first version, we already are very lucky lol
In my last post I did say that the GUI would be separated download from the models. I was not able to make this change in this version, since Dreambooth was a lot of work, but I'm already working to make this change on the next build.
There is also a new option.
Clip Skip: This pre...
The code for Dreambooth is working since Saturday, but I been fine tunning some paratemeters to make it easier to train a model without breaking it.
Here are some tests of a model trained with 25 images from the night before Christmas.
Also had some success training my face into a model.
Now I just need to see if the code is working with a huge Dataset and then can release this next build and create a tutorial on how to use it.
Hi all, managed to make dreambooth successfully train new models with my code.
In the meantime, been helping the libraries to fix this a small thing or two.
Also implemented Clip Skip, that is what NovelAI uses. It kinda of ignore the finer details in a image for a better overall quality. For real-li...
Dreambooth still not ready to use on this version, but there is other improvements:
More Schedulers:
There is 6 schedulers to select now. Euler Discrete is pretty popular, you can get good results with 30 steps.
Prompt Weighting:
You can now change the weight of the prompts:
(flaming) skull will add more weight to flaming
((flaming)) skull will add even more weight to flaming
...
2022-11-14 17:01:29 +0000 UTC View PostFor those newer on this Patreon, this is my app to interpolate frames. Input a anime or video and make more frames for it.
This update took quite a long time to be build. I swear I try to make the next one a loooot faster than this one. There was a lot of learning to make this build, mostly on the Pytorch area.
I can only hope that users of the older versions can find the model to be improved.
Tomorrow I will release a github with some codes that I made to train this model...
Some integration with the GUI is already working, it take all this options in consideration before training:
So you can select the scheduler, model, vae before start training.
The hardest part now will be to add those crazy new DLL to the build and work on other computers, cudos for those that optimized Dreambooth, there is some crazy optimizations on it and some of those require this we...
First, sorry for the delay to answer, like always I'm always rushing from one thing to another. Yesterday was a holiday here and went to visit some relatives, so it was harder to answer everyone.
â–ˆ The main issue with this last build seen to be "It won't render" and it show the message "Select a render mode."
This is a bug, but let me explain why.
There is two types of main models now. A regular model for Prompt2Img and Img2Img and another model that is...
Here we go again.
There is a few updates from the last build.
Now all models stay in a single folder. Read the txt inside the models foder for more information.
This build is a little bigger. It contain the new inpainting model (to make things easier)
Link: https://drive.google.com/file/d/1X53ybwjT8ii9DRZQG5H6h7QrRfhfxCjH/view?usp=share...
Hi all, will try to update SD today, there is a bunch of fixes from the last build, plus some extra features, like the new Inpainting, a button to generate a .PT file from any model (for faster load), faster upscalers and a option to use just FaceFix.
Messed with Xformer a little more, it might work better for some systems.
Rife-App:
I'm very close to get my desired results from the new model. Once is ready (it will probably take 4 or 5 more tries). I will re...
Hi all. Managed to get the new inpainting model working.
Also improved some of the upscaler code.
Upscale should be pretty fast now. Facefix should still be a little slow, it use some CPU code that will take some time to rewrite in the GPU, but still faster than the current build.
Also adding the option to only use Facefix.
Some new checkbox, like only export the upscale version.
Right now I wish to know what model in the Gui? 1.4 or 1.5?
1.5 usually is better, but i...
Thank you all for all the tests on the inner build.
It was a surprise that the code is already working on win 11.
I will try to add support to some more cards, there was a few reports that should be working, but is giving error currently.
There is also a error if Xformer fail, I'm already fixing it.
There is a lot of complains about the upscaler and I agree, the code is pretty bad now. Will focus on improving the upscaler, add only the face-fix option and add a option to...
Hi all, this is a new build for the SD GUI, but not really.
I really need to test if the xformer library is working on other machines.
Xformers need to be compiled to run in your system, but with some changes it can kinda of run in multiple systems. That is what I want to test.
This build will be full of small problems, since I'm still working on it.
But if you like to test it, feel free to send me any feedback, so is easier to me to fix it.
Close of completing the next build for SD. With SD releasing a new Vae Model, I had to remade the model loading options.
Now there is the option to load VAE and hypernetwork separately from the main model.
Everything is more organized now, all the custom files inside a single folder. With a TXT explaining each folder.
K_Euler_Ancestral is working for prom...
2022-10-24 17:51:19 +0000 UTC View PostJust made the scheduler Euler ancestral to work thanks to the help of smarter people than me, it should be working on the next update. Should not be long before the other K-Diffusers to work, but Euler ancestral is a community favorite.
SD Model 1.5 is released:
https://huggingface.co/runwayml/stable-diffusion-v1-5
Still didn't had the time to test it, or to check if the GUI is loadi...
Hi all, ok, this was a crazy ride between the last update and this one.
I'm trying to undestand all the SD code so its easier to implement stuff in the future, that involve a lot of tests, most that result in failed results, but it help me learn.
That create a mix of tests and easier stuff that I would like to implement, so right now I have quite a few ideas in the oven, that may appear soon in the GUI.
At the same time I did try to add some user suggestion and stuff that was ...
2022-10-19 02:10:56 +0000 UTC View PostChangelog summary:
Sorry everyone, didn't had much time to answer everyone in the last two days. Tomorrow morning will sit and answer everyone.
Been doing some crazy experimentations on SD and already managed to get some stuff working.
Outpaiting (as show in the images) is working a little better now, it sometimes manage to make a outpainting that make sense, it also apply to inpainting.
Got negative prompt working. I think?
Fixed the bug if the prompt is too big w...
Here we go, there was a lot of little stuff I had to fix yesterday:
There is probably still some bugs, but I did try to test most of options/tools on this build.
Link:
https://drive.google.com/file/d/1kLacShaNSBwRxvzQibFbEQa8TBAnl2wl/view?usp=sharing
Mirror:
Outpaiting have some differences in the workflow that made it a little trickier to implement.
For example, you render the outpaint.
You don't like the result: You can't use the new image as a source by pressing render, since you want to use the mask from the previews version.
You like the result: You have to use the new image instead of the older one as source.
There is also some changes that need to occur under the hood, for now promp...
2022-10-06 19:21:27 +0000 UTC View Post
The outpainting interface is pretty much ready, it read the correct pixels, concatenate with the original image and merge it back.
Just need now to run the AI instead of inverting it's colors.
Some side notes:
From time to time someone ask me why the inpainting not work that well. SD don't have a inpainting as good as Dalle2 and/or the technique that everyone use for inpainting is not that great.
There is two solutions for this.
The...
2022-10-04 20:19:47 +0000 UTC View Post
The harder part of outpainting is the interactive GUI and now it is almost working.
Just need a little more work and can release this new version.
2022-10-03 18:34:33 +0000 UTC View PostHello all, hope everyone is fine.
First of all, a quick note. If you are sending me feedback or a bug report, please send me a private message. The problem is that Patreon is really bad to show me messages inside posts. I try to answer everyone but is really easy to miss new comments inside a post.
There been quite a few bug reports and suggestions for SD, so in the next update expect:
Hi all, got home kinda of late yesterday, but managed to test and deploy this new version.
The new computer really is a blessing to do builds. The old one it was a hour of work, in this one the build get ready in less than 15 minutes.
Since I got home a little late, I wish it could had been a little more tested, but it seen that everything was working. i'm just unable to test the GPU selection since the pc only have one card.