Hi, this is my first post here.
My name is david and I'm a professional concept/3D artist who currently is doing AI workflows (currently in Substance Designer and Photoshop mostly). I started using AI in my works around 6 months ago when discovered Disco Diffusion and later MidJourney till I stall with Stable Diffusion. Since then I have been daily finetunning and training models for me and other people (mostly AI instagram artists) and recently I opened a patreon (with 2 other fellas) so people could support us in the model finetunning/making and experiments/workflows. If you think what we are doing is good, feel free to support us. Thanks!
Also if you want to check my works or just my daily basis feel free to check my IG at @davidsayszawarudoagain
--
This is the result of the mini tutorial I have put on our patreon for free.
https://www.patreon.com/posts/workflow-how-to-80636187
EDIT: Settings since probably someone want to test them or just know how was trained:
200 repeats per image, 30-40 images, epoch 6-8, lrate/urate default in this case, ClipSkip 2, max vector token 1, FP16 in both areas, constant without warm ups and trained over a custom model in DIM/NET 8x8 . Nothing to hide here. Input is more important than this "magical settings".
A bit different of what people use to do when training but I think it's lot of fun. Feel free to test a bit this char and tell me if works or not. As a quick difference this wasn't trained on SD 1.5 or NAI/Anything/Abyss, rather than that I used a custom finetunned model that did a decent job.
Lora weight works even at higher than 1 but recommended is 0.7-1. 0-5-0.6 should give you some char design style. Keyword is "reinbolt" and if somehow you can't see his green armor with lightings put "green" or "armor", it should appear.
By the way, as everything from my part, feel free to sell, transform, mix or whatever you want. Just have fun and throw some feedback if possible.