import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("Tongyi-MAI/Z-Image-Turbo", dtype=torch.bfloat16, device_map="cuda")
pipe.load_lora_weights("halilugur/tshirt_lora_v1")
prompt = "cute panda resting on blue clouds, crescent moon and yellow stars background, flat vector illustration, tshirt"
image = pipe(prompt).images[0]
| Sample 1 | Sample 2 | Sample 3 |
|---|---|---|
 Panda Clouds |
 Vintage Car |
 Retro Mountain |
T-Shirt LoRA v1 (Z-Image-Turbo)
This LoRA was trained on Z-Image-Turbo using the ai-toolkit by Ostris. It is optimized for creating t-shirt designs, specifically focusing on vintage and graphic styles.
Model Description
This LoRA is optimized for creating T-shirt graphics and vector-style illustrations. It works exceptionally well with clean backgrounds and vibrant colors, as seen in the panda example above.
Model Details
- Trigger Word:
tshirt - Base Model:
Tongyi-MAI/Z-Image-Turbo - Training Resolution: 1024x1024
- LoRA Rank: 32 (Linear) / 16 (Conv)
- Training Steps: 3000
Usage Instructions
Trigger Word
Always include tshirt in your prompt to activate the LoRA effect.
Recommended Inference Settings
Since this is based on a Turbo model:
- Steps: 8 - 12 steps (as per your training config)
- Guidance Scale (CFG): 1.0 - 1.5
- Sampler: FlowMatch
Prompting Tip
For best results, describe the graphic first, then the style, and end with the trigger word.
Example: A minimalist owl illustration, geometric style, white background, tshirt
Training Configuration
This model was trained with the following key parameters:
- Optimizer: AdamW 8-bit
- Learning Rate: 0.0001
- Network Dim: 32
- Alpha: 32
- Dtype: bf16
Files
tshirt_lora_v1.safetensors: The main LoRA weight file.
- Downloads last month
- 4
Model tree for halilugur/tshirt_lora_v1
Base model
Tongyi-MAI/Z-Image-Turbo