Usage for Krea 2 RAW:
Trigger word is not needed, but if results are bad, add 2000s to the start.
Image resolution should be 1MP for best results, but 2MP works at higher weights.
I personally use 15 steps, ddim, and ddim_uniform
Updates:
I am planning on training a v4.0 where I'll focus on improving image noise and finer details like that.
v3.0:
Weights have been fixed and look good at 1.0. If the look is still missing, try adding 2000s to the prompt and/or raise the weight to 1.5 (Usually happens with modern subjects, like a cybertruck)
v2.0:
No longer a struggle to get scenes void of people.
Better skin textures.
Lower LoRA weights change camera look, so flash photography, image noise, chromatic aberration, etc.
Higher LoRA weights change people's clothes, hairstyle, and environment to a more 2000s adjacent look.
Usage for Flux.2 Dev:
I personally use 40 steps, Flux guidance at 4-5, Euler, Flux2Scheduler.
1024x1024/1024x768/1080x810 are the resolutions I've been using.
Updates:
There will be no more updates for the Flux.2 Dev version.
Description
v2.0:
No longer needs 2000s as a trigger word, but can help boost the look.
No longer a struggles to get scenes void of people.
Better skin textures.
Lower LoRA weights change camera look, so flash photography, image noise, chromatic aberration, etc.
Higher LoRA weights change people's clothes, hairstyle, and environment to a more 2000s adjacent look.