anime-style-flux-lora

Anime Style LoRA for FLUX.1-dev

A style LoRA trained on anime/illustration artwork for FLUX.1-dev.

Details

• Base model: FLUX.1-dev
• Training steps: 14,000
• Rank/Alpha: 42/42
• Learning rate: 5e-5
• Optimizer: AdamW 8-bit
• Caption dropout: 0.35
• EMA: enabled (decay 0.99)
• No trigger word -- this is a style LoRA. Just describe what you want.

Sample Generations (step 14,000)

<img src="samples/1774034638157__000014000_0.jpg" width="400">

"a girl with blue hair sitting on a rooftop at sunset, legs dangling over the edge"

<img src="samples/1774034682198__000014000_1.jpg" width="400">

"close-up, a boy with glasses reading a book in a library"

<img src="samples/1774034726321__000014000_2.jpg" width="400">

"wide shot, a cat sleeping on a windowsill, rain outside"

<img src="samples/1774034770386__000014000_3.jpg" width="400">

"portrait, a woman with flowers in her hair, looking to the side"

<img src="samples/1774034814417__000014000_4.jpg" width="400">

"a knight standing in a field of flowers, sword at their side"

<img src="samples/1774034858452__000014000_5.jpg" width="400">

"two friends walking down a narrow street at night, neon signs"

Usage

Load as a LoRA in ComfyUI, A1111, or diffusers with FLUX.1-dev as the base model. No trigger word needed -- the style is applied automatically.

Recommended settings:

• Guidance scale: 3.5
• Steps: 35
• LoRA strength: 0.7-1.0