README.md

---
license: other
license_name: flux-1-dev-non-commercial-license
license_link: https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md
base_model: black-forest-labs/FLUX.2-klein-base-9B
tags:
  - lora
  - flux
  - flux-lora
  - sprite-sheet
  - pixel-art
  - game-dev
  - animation
library_name: diffusers
---

# FLUX.2-klein Sprite Sheet LoRA

A LoRA trained on FLUX.2-klein-base-9B for generating game sprite sheets in the style of LPC (Liberated Pixel Cup) assets.

## Model Details

- **Base Model:** [black-forest-labs/FLUX.2-klein-base-9B](https://huggingface.co/black-forest-labs/FLUX.2-klein-base-9B)
- **Training Data:** ~18,700 LPC-style sprite sheet images
- **Training Steps:** 3,000
- **LoRA Rank:** 32
- **LoRA Alpha:** 32
- **Trigger Word:** `gmsspritesheet`

## Available Checkpoints

This repository contains multiple checkpoints from different training stages:

- `gmsspritesheet1.safetensors` - Final model (step 3000) ⭐ **Recommended**
- `gmsspritesheet1_000002750.safetensors` - Step 2750
- `gmsspritesheet1_000002500.safetensors` - Step 2500
- `gmsspritesheet1_000002250.safetensors` - Step 2250
- `gmsspritesheet1_000002000.safetensors` - Step 2000

## Usage

### Using with Diffusers

```python
import torch
from diffusers import DiffusionPipeline

# Load base model
pipe = DiffusionPipeline.from_pretrained(
    "black-forest-labs/FLUX.2-klein-base-9B",
    torch_dtype=torch.bfloat16
)
pipe.to("cuda")

# Load LoRA
pipe.load_lora_weights("Mystic07/flux-lora-spritesheet", weight_name="gmsspritesheet1.safetensors")

# Generate
prompt = "gmsspritesheet, 4 direction sprite sheet, warrior character, walk cycle animation, pixel art"
image = pipe(
    prompt=prompt,
    num_inference_steps=30,
    guidance_scale=4.0,
    width=1024,
    height=1024
).images[0]

image.save("sprite_sheet.png")
```

### Using with ComfyUI

1. Download `gmsspritesheet1.safetensors`
2. Place it in `ComfyUI/models/loras/`
3. Add a "Load LoRA" node in your workflow
4. Use trigger word `gmsspritesheet` in your prompt

## Example Prompts

```
gmsspritesheet, 4 direction sprite sheet, warrior character, walk cycle animation, pixel art

gmsspritesheet, sprite sheet, mage character, spell casting animation, top-down view

gmsspritesheet, 4 direction sprite sheet, archer character, shooting bow, pixel art style

gmsspritesheet, character sprite sheet, knight in armor, idle and attack animations

gmsspritesheet, sprite sheet, rogue character, stealth movement, multiple poses
```

## Recommended Settings

- **Steps:** 20-40 (30 recommended)
- **Guidance Scale:** 3.5-5.0 (4.0 recommended)
- **Resolution:** 1024x1024 or 768x768
- **Sampler:** Any (Euler, DPM++ work well)

## Training Details

- **Dataset:** 18,706 LPC-style sprite sheet images
- **Learning Rate:** 1e-4
- **Optimizer:** AdamW
- **Batch Size:** 1
- **Resolution:** 1024x1024
- **Training Time:** ~3 hours on A100-40GB

## Use Cases

- Game development sprite assets
- Character animation references
- Pixel art style generation
- Top-down RPG character sprites
- Walk cycles and action animations

## Limitations

- Best results with pixel art and sprite sheet styles
- Optimized for character sprites (humanoid forms)
- May struggle with very complex multi-character scenes
- Works best with the trigger word `gmsspritesheet`

## License

This LoRA inherits the license from FLUX.2-klein-base-9B. Please see the [base model license](https://huggingface.co/black-forest-labs/FLUX.2-klein-base-9B) for usage terms.

## Citation

If you use this LoRA in your work, please credit:
```
Mystic07/flux-lora-spritesheet
Trained on FLUX.2-klein-base-9B by Black Forest Labs
```

## Sample Images

*Coming soon - example generations from the model*

---

**Created by:** [Mystic07](https://huggingface.co/Mystic07)  
**Training Framework:** [ai-toolkit](https://github.com/ostris/ai-toolkit)