"Please read this before using, thank you."
Generate simple anime style portraits for Flux Dev.
Q: "Can this do things other than portrait?"
A: "Perhaps. But since this is Flux and I use quite a high DIM size, it may not yield desirable result. But feel free to try it!"
You may have seen other Anitrait projects based on Flux on Tensor (TE) or other site. What's the difference between this one and those? This one uses a different dataset, and a bit different training process than Anitrait TE or Anitrait FX. Also different trigger word. It is still based on images generated via Animagine XL 3.1, but I included some LoRas to alter the style a bit differently. The result leans more towards 2D, but on some occasion like close-up portraits it gives you 2.5D image, so I named it 2.25D. Can this generate male character? Yes, to some extent. It's mostly waifu oriented but I did put male images in the data.
How to Use:
Use natural language prompting. I never test it with tag prompting (not that it won't work, but you can also try it). Try putting the trigger before a subject (example in showcase image).
Use 0.8-1 weight. You can try lowering it to 0.6 if you want to use with other LoRas.
Use Euler-Simple, Euler-beta, or Euler normal for the sampler-scheduler combo.
Most of the time, it doesn't require uscale, but you can do it for higher fidelity. Set the denoise to 0.2 if you want to upscale.
If you want to lock it into full 2D, integrate "anime" into your prompt.
If you have questions or want to roast me if the model is bad, feel free to contact me at Tensor's Discord server.
Thank you for using this LoRa for your image generation.