chambersu
patreon
Design and Texture tool, AI body/face/hair/cloth/armor/mech/sword generation
🕑 Added 2024-03-13 12:43:41 +0000 UTCHere are some new LoRA I trained for my new workflow, AI-powered designs and textures tools:
AI Design LoRA.zip
Also training data here, about 30 percent of them rendered from artwork I created and posted before, the rest are bought and collected from the internet:
TrainDataPack.zip
Every LoRA was trained from a relatively tiny collection including 20~ 50 pictures, for this early and test version but it still helps a lot now I think.
Training setting for kohya_ss for every LoRA I create, only use different data, image size, and epoch, the rest are the same (highly recommend trying your own better data and setting and hope you can share any better results!~):
ToonHairTurn_v1.json
Here are the raw picture samples with the whole generated info that can be read (! take care of the body turn sample, they are naked bodies and not suitable to read in public~):
RawPictureSample.zip
They should work fine and get similar results above with the whole settings (prompt, sample param, other LoRA, and open pose or soft edge control net).
The rest LoRA and checkpoints I used are from Civitai: The Home of Open-Source Generative AI, you can search and download there, or replace/try anything else you want~
Add dynamic prompt and wildcard from Civital to help get unlimited different results, using Gemini and Chat GPT to generate prompt wildcards for testing work well too~
I am not very good at training LoRA now, just try to make every step/feature workable this time and keep settings as simple/default as possible,
will learn further about reasonable data, tags, and train settings later,
then train better by only using my own artwork render images to ensure it will be open to use, and follow my styles/tastes.
I consider how I can improve my artwork all the time,
keep practicing, creating, and collecting assets and tools,
it helps them become better in most cases but seems slow if always follow the traditional ways,
I do want better but also quicker, make artist/my work relaxing and smooth, which is the reason I love creating artwork, not about scarifier yourself for great artwork, but enjoying yourself and then generating great work naturally, grow together.
As a 3d model artist, I am crazy about vertex editing and procedural modeling/shading,
while I cannot design pretty things easily and feel annoyed with sculpture and painting,
there are a lot of reasons but it does not important here,
the point is AI can help to make design/texturing happier and better,
some people doubt about morality and law problem about AIGC,
but I have ideas to find or create open assets to train open AI tools,
LoRAcost about 20~50 images to train, which won‘t cost a lot to problem us,
the problem is open checkpoints cost more than thousands of images,
the only way I hold now is to use another legal AI image generator to prove them within a low budget,
and I know Adobe Firefly is trained by Adobe Stock, a royalty-free image library, but their license is complex and I am not a lawyer who can read and ensure in a short time,
and seems AI training is more open and free in Japan...
whatever, I will try to train open checkpoints if possible~
but even for the current version, I think there is no special kind of training data or prompts or something else to fake a special style or artist I know,
mostly depends on how you guide the AI.
finally, I got these first features-enough version AI tools for my new workflow,
I will be back to model work to create characters to ensure/prove they work fine,
to make every tool work well with my new character library/standard,(They will be more valuable than any other assets in the long term in the future~)
post them open and free to share and get feedback,
and figure out what should be fixed and improved during creation,
also plan some real-time shading improvements.
great~ talk is cheap, back to work, and will show the results when ready like the old same~
Feel free to contact me by leaving a comment or mailing to chambersu1996@gmail.com, if any questions or anything here interests you~
Also Discord Channel here~
https://discord.gg/SFJFcctfgD
Comments
Chamber Su
Yep, use it in the right way to benefit more people, also, Let's chase the champion with this magic ~ ◝(⑅•ᴗ•⑅)◜
Mbdot
AI FTW! As long as AI is just being used as a reference or guide, there's no need to worry about all the legal issues. Every artist started from zero and they all learnt from other artists, used their work as reference and then founded their own art style. AI will help a lot with the design stage. They can generate so many different kinds of designs a thousand times faster than our brains can think of.
