import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("fill-in-base-model", dtype=torch.bfloat16, device_map="cuda")
pipe.load_lora_weights("jayark/sammy-style-lora")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]Sammy Style LoRA
Style LoRA trained on Qwen-Image-2512 to capture the Sammy children's book art style. Generates new characters, environments, and backgrounds in the style — not limited to any single character.
Usage
- Trigger word:
sammy style - Base model: Qwen-Image-2512
- Architecture: LoRA (rank 16, alpha 4)
Checkpoints
| File | Training Step | Notes |
|---|---|---|
sammy-style-400.safetensors |
400 | Earlier checkpoint |
sammy-style-600.safetensors |
600 | Recommended — best quality/fidelity balance |
Training Details
- ~50-70 training images (character images, backgrounds, synthetic fills)
- Natural language captions (not booru tags) — required by Qwen2.5-VL text encoder
- Captions describe content only; style is the learned residual
- 1500 total training steps; step 400-600 is the sweet spot before overfitting
- Downloads last month
- 6