songyiren/prehuman-flux-kontext-lora

Prehuman - FLUX.1-Kontext-dev LoRA

LoRA fine-tune of black-forest-labs/FLUX.1-Kontext-dev, trained with DiffSynth-Studio.

Two checkpoints from the same training run are included:

Training configuration

• —Base model: black-forest-labs/FLUX.1-Kontext-dev (DiT weights only; LoRA applied to pipe.dit.)
• —Task: sft, image editing (kontext_images extra input)
• —LoRA rank: 64
• —LoRA target modules: a_to_qkv, b_to_qkv, ff_a.0, ff_a.2, ff_b.0, ff_b.2, a_to_out, b_to_out, proj_out, norm.linear, norm1_a.linear, norm1_b.linear, to_qkv_mlp
• —Learning rate: 1e-4
• —Epochs: 5
• —Gradient checkpointing: enabled

License

This LoRA is a derivative of FLUX.1-Kontext-dev and is distributed under the same [FLUX.1 [dev] Non-Commercial License](https://huggingface.co/black-forest-labs/FLUX.1-Kontext-dev/blob/main/LICENSE.md). Non-commercial use only; see the base model's license for full terms.