Trained mainly on the artist’s later works, which were more semi-realistic (compared to 'dinoartforame' period).

Here are some tips. Certain prompts and tags (like 'catgirl', 'kimono', probably anything Japanese or anime-related) tend to trigger the built-in anime style of Flux, which is too simplified and does not provide a good basis for semi-realistic anime images. It's also bad with anatomy. So, with this LoRA you probably don't want to trigger native Flux-anime style, and to avoid that, you can try adding tags like 'realism, realistic, highly detailed, intricate' to your prompt. You may have to increase the weight of the LoRA after this for it to be effective. 'Loose brushwork' tag helps to add a painterly effect.