If you enjoy this lora, please follow, support, and appreciate M-da S-tarou for their amazing works. Pixiv
This is a Flux Dev lora trained using images inspired by the artistic style of M-da S-tarou. This lora effective at applying the style in most circumstances, even without explicitly prompting for in the style of m-da_s-tarou, but will do better with women due to the nature of the style tending to focus on them. You should be able to use short or long natural language prompts, as captioning was done with natural language, not tags.
This lora should be fairly compatible with other Flux loras, assuming they were trained well.
Hands are an issue if unprompted in most loras for Flux, I advise prompting specific desired positions and adjusting the strength of the lora if issues are faced that can't be easily resolved. Also, latent upscaling seems to improve hands, and images overall, significantly, I highly recommend using latent upscaling.
Only a few images with nudity were included (so did not learn it strongly), and no images with sexually explicit content were included.
Description
v2.0_Experimental, a new version using a new optimizer that is very promising, still need to refine settings for it, but figured I share this sooner rather than later. Dynamically resized using sv_fro with 0.98 threshold. More info to come. Please provide feedback!
FAQ
Comments (3)
Hey Havoc! Thank you for LoRA, I downloaded it and tested it to compare with my LoRA and share my observations. Your LoRA doesn't need trigger words, as well as mine, which I trained. I tried to make the style LoRA work with trigger words, but all my attempts were in vain (at least for now). So you can try to use your LoRA without trigger words, and to help LoRA work in more scenarios, you can try my prompt "A highly detailed anime artwork." or you can try just "Anime artwork." (but I haven't tested this prompt). Also, LoRA with some of my prompts, especially when the character is in full shot, breaks limbs and fingers. Also, I noticed problems with colors, saturation and contrast? And sometimes details are lost? I don't know the reason for this behavior yet, but if I find it, I'll add one more comment here. I only had one LoRA without all these defects...
Thanks for the feedback, let me know if you figure out anything more! I'll incorporate your feedback into my next version!
Great work !
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.