Looking for other anons that are using local-AI for their gamedev/art projects. In cartooning or anime styles, is the best option still just something like Illustrious? I feel like I'm constantly in a trade-off between quality and prompt adherence. Models like ILL have great quality and customization support with loras, but I fucking hate how character-focused they are. Even with negative prompts, I too often struggle to get a single object to render without a superfluous butt, tiddy, or hand.
Is there a model of similar quality that is more tuned for environmental details and objects over characters? Loras haven't been cutting it.
>pic not related, just something I put together for fun

>>106198985
I usually gen in a series of batches that I square exponentially. If the change didn't obviously degrade quality in 2 batches, then I do 4 batches, then 8 and so on. Once its at >200 images per interaction, I assume its fine enough and feel okay leaving it run over night. The only advice I can give is try and find any ways you can shave a few seconds per image generation. It saves so much fucking time over a large run.