.jpeg)
Start off your prompt with "Anime screenshot from Mushoku Tensei," and refrain from adding too much detail about line-art, drawing technique outside of this beginning text as above to stick to a Mushoku Tensei style artwork.
For this LoRA, found that I get best conformity and strike detail using res_2s sampling, ddim_uniform scheduler, 9 steps and CFG of 1.0.
I am shocked that with just 40 hand-picked screenshots from the anime, LoRA training was able to grasp the subtleties of the artwork techniques used in the show and not overfit to one particular design or another.
I generated the prompts using Grok with some project set-up including parsing it a Z-Image prompt guide .md file which you can find here. Putting guardrails like this on Grok and telling it strictly about only using "anime screenshot from mushoku tensei" to describe the art style gives you striking results without fiddling around yourself. Seriously cool!
Hope you like this LoRA as much as I do