Download
1 variant available
The FLUX.1 [dev] Model is licensed by Black Forest Labs. Inc. under the FLUX.1 [dev] Non-Commercial License. Copyright Black Forest Labs. Inc.
IN NO EVENT SHALL BLACK FOREST LABS, INC. BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.
Kyoto Animation Style — New LoRA for Flux.2 klein 9B, trained in the modern Kyoto Animation style.
This model captures the clean, warm, and instantly recognizable aesthetic of Kyoto Animation’s recent works. The training dataset consists of screenshots from:
50% Violet Evergarden
25% A Silent Voice (Koe no Katachi)
25% Sound! Euphonium (Hibike! Euphonium)
Style behavior:
The more “cute” and gentle your prompt is (e.g. “cute girl”, “warm smile”, “peaceful everyday scene”), the closer the result gets to the soft, bright style of Koe no Katachi and Hibike! Euphonium.
The more dramatic, emotional, or melancholic your prompt becomes (e.g. “dramatic scene”, “melancholic atmosphere”, “serious expression”), the closer it leans toward the beautiful, painterly style of Violet Evergarden.
The model excels at generating female characters — from young adult girls to mature women — as well as male characters of all ages, from teenagers to elderly.
Key features:
Prompting in natural English only (no tags or tokens)
Trigger word: Anime
Primarily designed for text-to-image, but also works great in img2img for style transfer
Trained on ~270 reference photos + 180 screenshots + 90 synthetic examples (to improve anatomy)
Unfortunately, occasional finger anatomy issues can still appear. I'm currently working on a universal hand-fixer LoRA and will share the link soon after release.
The model was not trained on NSFW content, but it can be combined with specialized NSFW LoRA if needed.
Sometimes the model struggles with very small bust sizes. In those cases, this LoRA works excellently.
Generation recommendations:
I usually generate at 6–12 steps with Euler sampler. A strong recommendation for you to try euler_ancestral or res_multistep
Feel free to try other samplers—the results may vary
High-quality prompt = high-quality result
The model handles complex backgrounds very well
For 16:9 aspect ratio or scenes with multiple characters, I strongly recommend using a latent resolution of 1440×900 or higher to preserve fine details
For common anatomy errors of the Klein 9b model, use this anatomy fixer.