Mostly working on generative AI and tuning it to 100% product reproduction. It's harder than you might think especially with more complex subjects like cars. Tr

Mostly working on generative AI and tuning it to 100% product reproduction. It's harder than you might think especially with more complex subjects like cars. Training custom controlnets is a pain just because of the sheer size of the datasets needed. Subject LoRas/DoRas are pretty much solved though. Here is an example[0] in 6k resolution from two weeks ago done with just text conditioning. The controlnet elevates the precision even more up to 99%, the last 1% is easier fixed in post. Which isn't optimal but a good way to go for now. [0] https://42lux.de/amg-gt-63 https://42lux.de/amg-gt-63