Limbicnation/pixel-art-lora

Pixel Art Sprite LoRA (FLUX.2-klein-4B)

A LoRA fine-tune of FLUX.2-klein-4B for generating pixel art game sprites.

Usage

from diffusers import Flux2KleinPipeline
import torch

pipe = Flux2KleinPipeline.from_pretrained(
    "black-forest-labs/FLUX.2-klein-4B",
    torch_dtype=torch.bfloat16,
)
pipe.load_lora_weights("Limbicnation/pixel-art-lora")
pipe.to("cuda")

image = pipe(
    "pixel art sprite, a brave knight with sword and shield, game asset",
    num_inference_steps=4,
    guidance_scale=1.0,
).images[0]
image.save("knight_sprite.png")

Trigger word: pixel art sprite

Training Details

Parameter	Value
Base model	black-forest-labs/FLUX.2-klein-4B
LoRA rank	64
LoRA alpha	128
rsLoRA	enabled
Dataset	1010 images, 512px
Training steps	1000
Batch size	1 (grad_accum 4)
Learning rate	1e-4 (cosine w/ restarts)
Optimizer	AdamW 8-bit
Precision	bfloat16
Loss range	1.62-1.93 (stable)
Hardware	NVIDIA RTX 4090

Inference Parameters

• Steps: 4 (FLUX.2-klein is optimized for few-step inference)
• Guidance scale: 1.0
• Resolution: 512x512 recommended (trained resolution)

License

Apache 2.0 (same as base model)