Stable Diff:
Hi all, during the weekend I got my projects to work on the new SSD, now the computer can run 50 steps a lot faster than the time it took to run 10 on the old computer lol
Needless to say that this helped me me to fix the bugs from the last update a lot faster.
I could also test using the model with float32, so it was easy to fix this as well.
...
2022-09-26 19:35:59 +0000 UTC View PostGot a new SSD, a 3080 and two memory ram of 16gb.
The computer just had an upgrade from 4vram to 12vram and 8gb ram to 32gb.
Did not had the time to finish installation of all my applications on the new SSD, but I can already tell it will help a lot developing my apps.
There is the speed increase in testing the software and options that I could not even test it, like 32float and ...
2022-09-23 18:13:35 +0000 UTC View Posthttps://drive.google.com/file/d/1zY7BTTlhjVJWUW9Q647EjQ057hA8MXzu/view?usp=sharing
Wii try to upload the file to a mirror soon, there been some users with download problems on the last release.
GPU selection: 2022-09-22 13:26:24 +0000 UTC View Post
Besides bug fixes and fixing the memory using Upscaler + Save Vram, there is a few new options.
Scheduler: To change the scheduler used by the model, there is only a few options for now, its been hard finding the source for more schedulers, any link with more schedulers are welcome.
GPU: For those with multiple GPU's, you can select the one you want to use.
Model: It will hand...
2022-09-21 18:23:01 +0000 UTC View PostHi all. First of all. After messing with my computer in the weekend it seen to be working fine, in fact I spent most of the day working on SD with no problems.
There was already a some fixes, like memory usage for Upscale and some GUI smaller bugs.
There been some reports that Upscale is slower than 0.41
That is correct. I spent some time optimizing the upscaler for 0.41, b...
Made this tutorial really quick to help you guys to download new models.
The tutorial still need work, will try to slowly try to improve it.
Downloading models from HuggingFace the first time can be kinda hard, but is the faster way to get access to updates to a model and getting a new model once is released.
I will try to add easier ways to download models in the future, but ...
2022-09-16 17:08:08 +0000 UTC View PostI just made a render with Waifu Diffusion.
https://huggingface.co/hakurei/waifu-diffusion/tree/fp16
You just need to throw the model in the "models" folder and after that select the model on the GUI.
I made a mirror for Waifu so is easier:
2022-09-16 02:28:16 +0000 UTC
View Post
Hi all, my cleaning made the computer alive again, horray.
Even so, the computer is still kinda of funny, slowing down at random.
I'm afraid that it may not turn on tomorrow, so I gave it my best to complete the build today, because of that i'm ashamed to say that I was not able to test all the options as well as I would like it, the computer seen to be sweating to load the model ea...
2022-09-16 01:47:25 +0000 UTC View PostHello all. Here is an update about the next release of the GUI for Stable Diffusion.
There was a good chance that the build would be complete yesterday, but my computer felt like it was a good time to freeze and broke down.
I took some time messing with it and there is a pretty good chance that the Ram slots are dirty, so today I got some stuff to clean it at the end of the day. If ...
2022-09-15 19:21:34 +0000 UTC View Post
Tried to put a bunch of quality of life improvements that was suggested here.
Keep in mind that most the options will improve with time, for now i'm trying to find the quickest way to implement improvements, so it don't slow down development that much.
I just need to pretty much finish implementing gfpgan and this update should be ready.
Changing model will not be simple in this ...
2022-09-13 18:47:21 +0000 UTC View PostTextual Inversion is like that, you take 5~20 images of something and train a variable that can be used on Stable Diffusion. This image explain it well.
It seen that it require 12Vram to train it (did not test myself) and a few hours of training.
There still a LOT of stuff that I...
2022-09-09 20:10:22 +0000 UTC View Post
Most stuff don't show on the GUI, like bug fixes and improvements, but it make little sense to print-screen the code lol
Just a side note, usually I don't have much time to answer comments on the Weekend, but Monday I have plenty of time to answer everything.
2022-09-09 19:48:05 +0000 UTC View PostRight now I'm adding the full precision models and Waifu Model, it's already pretty big files, and probably will be even bigger in the future. So what should be the better option for future builds?
2022-09-08 19:26:52 +0000 UTC View PostI'm gonna try to add all this on the next update, but if something is to problematic it may be delayed to 0.6
A really quick build. It is exactly like 0.40, but with some memory optimizations that should allow to use 512X512 with 4vram. (Turn on Save VRam Option)
Didn't test on how much this save memory, but hope it will allow 512x512 at least on all 4vram cards.
2022-09-06 02:49:07 +0000 UTC View PostHi all. First of all, news about Rife-app.
I been testing models left and right for about two months now, throwing stuff on the wall to see what sticks. Yesterday I managed to train a model that I think is eggcellent, there still a little training to be done, then need to train the Live Action version of it, but I'm confident is the best model so far, so it should be ready in a week or tw...
2022-09-05 12:51:05 +0000 UTC View PostHi all. First of all, I can see there been a lot of messages for me on the weekend, I don't have time to read and answer right now, but tomorrow will read/answer all messages.
Some changes for 0.4:
Today or tomorrow will add it to the Stable Diffusion so you can upscale the results.
After that update, will add the option to use Stable Diffusion on .mp4 video files.
In-paint...
2022-08-30 22:54:19 +0000 UTC View Post
Still need to work on it tomorrow, but is 90% ready
2022-08-30 01:53:04 +0000 UTC View PostHi all. Stable Diffusion was released 7 days ago and everything been so crazy, each new day there is new options and tools build on top of it.
I would like to share some of my thoughts on the project and my build.
First, the model 1.4 seen a little heavier than the leaked one. With my 0.2 I did managed to improve memory a little, but 448X448 was my current limit for 4Vram.
I can ...
For this update I focused more on two of the most requested changes.
img2img: Select an image as a starting point to generate Images
Working Samples: Generate more than 1 image per render, this also use more Vram.
Other than that I did a little optimization and hide some unused options.
There still looooots of stuff to do, so expect more...
2022-08-27 19:03:50 +0000 UTC View Post
Just need to set it up, gonna help me develop SD GUI and the other projects.
Btw, SD GUI 0.2 is already in the making, having the option to use a input image and some other smaller options.
It took me 5h after all the code to be ready to make the .exe to work T.T
I really could not test it that much, since it took a long time to make the .exe work and I need to sleep, so please send me any feedback/bug
Ok, so this is pretty basic, I just wanted to have a basic project that work so everyone can enjoy the SD project.
Resolution need to be multiple of 64 (64, 128, 1...
2022-08-24 03:06:46 +0000 UTC View Post
I'm managing to make 512X256 images using 4Vram. More than that is break.
This is my first local master piece. "A horse"
If my compiler scripts work (can be a mess sometimes) it should be ready today.
As a bonus, a print from the coloring model I'm currently training (still a long way to go)
2022-08-23 13:52:12 +0000 UTC View PostDownloaded all the SD stuff and managed to get the code running locally.
To bad i'm still on my old card and it get out of memory along the way lol
But no problem, I'm can already start working on the GUI like that and just test it once my new card get in here.
The model load ...
2022-08-22 22:35:15 +0000 UTC View Post
If someone require a back/white -> Color dataset, send me a message.
I can't share this dataset publicly because it's chainsaw-man manga, but if someone want to train a model, just send me a message.
The dataset have around 500 pages of manually aligned 760X1200 pages.
I also removed all the useless pages from the dataset.
Just made a purchase for a 3080 and some extra parts to upgrade my little computer, just need to wait some days to arrive it.
This will really help me with my work flow and improve releases.
Coloring model:
Ok, for the coloring model, I knew that GAN was really hard to train, but I gave it a shot anyway. It was hard but in the end it generated colored images. The quality was fine for a showcase, but nothing that could really be used in real life.
I did learn a lot, some stuff that I'm using on Rife-App now.
So for now I put Gan aside and started with...
2022-08-12 17:15:30 +0000 UTC View Post