Some Patrons may remember that some time ago I posted a few pictures of a sketch 2 drawing.
The project was sleeping for some time, but I been studying since then to improve Rife-App and learned a bunch of stuff during this time.
The new Rife-App (in training) use some stuff that I managed to use on this AI of mine as well.
So I h...
Here we go, fixed some stuff from the closed build.
Two new models. Diffusion V1 and V2.
They are both pretty good. V2 may be a little better, but who can tell with art?
Guidance Scale: Only used on Diffusion models, it affect how "Creative" the final output is. If you render images bigger than 512X512 you may ...
2022-02-15 01:40:48 +0000 UTC View Post
Clip-App is already working on the command prompt, for those who want to create some script around it.
The interface is also saving all the configuration, so you can close and open it without losing info.
Fixed some bugs that my fellow Patrons reported on the closed Build. The new build should be working very soon. Just want to try t...
2022-02-11 19:10:54 +0000 UTC View PostAfter a slow January, I'm already working to release some stuff in a few days.
Clip-App: Got the new algorithm working, just fixing/improving some stuff on the GUI now. Bad news, my computer barely work with it, so I need someone willing to test the new build with a decent enough card? 8vram should work with a small resolu...
2022-02-01 18:36:57 +0000 UTC View Post
I been fighting a little with the implementation of the new algorithm for Clip-App.
The main reason is that it require JAX and it don't really like windows that much:
There is some initial native Windows support, but since it is still somewhat immature, th... 2021-12-28 13:54:06 +0000 UTC View Post
I been experimenting with some new scripts for Clip-App. Plus I'm also doing a side project using it with a buddy. Here are some results from it.
I got a server for this month so I could do some better renders with Clip, since my computer can't really handle it. Since it's December and everyone want to chill, I will pretty much take ...
2021-12-08 18:11:38 +0000 UTC View PostHere we go, the new update.
It have new models for animation and RL.
It have an alpha channel option for PNG seq. and gif output
A few minor fixes, like a filename with a dot in its name.
A lot of the memory improvements are kinda of experimental, so let me know if you find any error and I will fix asap.
2021-11-30 01:54:01 +0000 UTC View Post
Look at those models. Horray!
While I'm constantly working on the quality of the models, I think they are good enough for a release.
Gif is also improved.
Alpha option is back, for PNG sequence, and gif also get the alpha if both are turned on.
Anime Sharp give more detail on the edges, but the motion flow can look stiff,...
2021-11-29 17:31:18 +0000 UTC View Post
It seen than finally I am understanding the secrets to train a RL model, it also give me more freedom to the the results I prefer. This one is still training, but it's looking pretty nice.
The model for animation and RL should be ready til the end of the week.
2021-11-24 11:52:22 +0000 UTC View PostIt work on your browser, its the paintbrush of the future, paint some stuff or write some text and generate a landscape from it.
2021-11-23 18:21:35 +0000 UTC
View Post
Been working really hard to make Rife-App a somewhat stable application.
Fixed alpha export, plus now gif work with alpha (if turned on) and it should create crisp results overall.
The new model is looking really good. It is still in training for animation and still need to train for RL, but I have a good feeling it will work nicely ...
2021-11-22 11:38:42 +0000 UTC View PostOriginal
Mask
Result
The Ui is also ...
2021-11-11 18:34:52 +0000 UTC View Post
A great video from Patron Owen Davies using the new Rife animation model and Topaz for upscale.
2021-11-11 13:00:31 +0000 UTC View PostHere is the new version. Horray!
There is 3 new Animation models and 1 new RL model.
RL have a bunch of stuff I need to worry that don't happen on animation, so it's been a really pain in the ass to make a good model, but it's fun, so I don't mind that much.
I also fixed a a lot of bugs to try to improve memory and remove flickering color from the final o...
2021-11-08 14:31:02 +0000 UTC View PostReal Life mode is interpreting better the shapes of new frames than older models, but Real Life is tricky, sometimes you want some blur to create a credible new frame, so perf...
2021-11-04 12:20:26 +0000 UTC View PostWith the current development in the AI area, there is some new interesting libraries. So i'm starting to explore some stuff to make the projects work on AMD cards. Horray! Still to soon to know how well it will work.
2021-11-01 13:37:01 +0000 UTC View PostI been testing and mixing a lot of ideas for models, putting it to the test and training, training and training it.
So far I have 3 models for animations.
Starting a GUI so you can mask photos and stuff on the APP using Lama:
https://github.com/saic-mdal/lama
It should be a small/medium size project using the codes I already have ready from the other projects.
In the future I will try ...
2021-10-25 17:44:46 +0000 UTC View PostI have a few ideas in mind, it would not be a tool that could create "Professional animations" or anything like that, it would be more of silly animations just to play a little and have fun.
It would work something like this:
It's been a little more than two months since I began training my own model for Rife-App, since then it's easily been a model training at least half of each week.
So let's say, 12h of training per day, for 60 days. That give more or less 720h of training.
The hardest part of training a model is that I need around 1:30h to see if a te...
2021-10-20 12:14:12 +0000 UTC View PostOk, I kinda of rushed this version, because on 2.50 there is a annoying bug. If you are using batch size != 1 it may crash at the end of the render.
1920X1080
Animation mode, around 4X frames
06:48 to render, but my CPU was the bottleneck, CUDA was around 20%
Not the best quality, but great for casual interpolations, can easily run in real-time on a decent computer.
2021-10-15 22:44:35 +0000 UTC View Post
Look how the subtitle is almost perfect, the last model messed it put a lot.
The last model is also slower than Rife Models, this new one is as fast or faster than Rife.
I'm also improving the YUV and 10bit color code to see if it will fix the weird colors here and there. YUV is already fixed, still need to test 10bit.
... 2021-10-13 12:24:55 +0000 UTC View PostHere we go, after so many updates about it.
There is the new two models, hope they are good. For now they are a little slower from the default models, since I already had a hard time making they somewhat decent to use. In the future I can try to optimize them more.
Some personal issues made me delay a little the next update. I'm gonna try to release it today or tomorrow.
Also I'm playing around with making a AI sketch2grayscale. My model algorithm isreally stupid, I'm surprised they work at all. lol
My methodology is pretty much this.
2021-10-08 19:04:23 +0000 UTC
View Post
Animation model is complete. If my RL training goes well today, it should be finished tomorrow. The app changes also are pretty much complete. If everything goes well, I may be able to release the update tomorrow.
2021-10-05 22:55:48 +0000 UTC View PostThe last few examples was using only the first part of the model, those are using all the model.
It's far from perfect, but I'm sure it can be improved with time. Some stuff like the subtitle letters will be hard to fix, since the AI sometimes try to priority the background animation.
For the next update. Also a bunch of bug fixes.
2021-09-30 20:35:39 +0000 UTC View Post