Spikouuu/persona-klein-lora-v1

Persona Klein v1 — FLUX.2 Klein 9B Style LoRA

A style LoRA trained on a custom dataset blending Persona characters into Bebop-style environments. Trained with Ostris AI Toolkit on FLUX.2-klein-base-9B.

Trigger word

Use p3rsona at the start of prompts, e.g.

Composition tags used during training:

• — for character-focused prompts
• — for pure scenes/environments

Training details

• Base model: FLUX.2-klein-base-9B
• Trainer: Ostris AI Toolkit (sd_trainer)
• Rank: linear=128 / linear_alpha=64 / conv=64 / conv_alpha=32
• Steps: 3000
• Learning rate: 9.5e-5
• Optimizer: adamw8bit, weight_decay=1e-5
• Dataset: 71 images (33 character-in-scene composites + 38 Bebop environments)

Recommended inference

• LoRA scale: 0.8–1.0 for natural scenes, 1.0–1.2 for stylized portraits
• Inference steps: 50
• Guidance scale: 4.0
• Use with: diffusers Flux2KleinPipeline, or the Fal AI Klein 9B LoRA endpoint