m2-craftly-lu

M2-Craftly-Lumina (by Mushfiqur)

A private, rebranded build of [Lumina-Image-2.0](https://huggingface.co/Alpha-VLLM/Lumina-Image-2.0)
by Alpha-VLLM — a 2 billion parameter flow-based diffusion transformer for
text-to-image generation.

**This is a rebrand, not a new architecture.** Weights and architecture are
unchanged from the upstream model; only class names (pipeline + inner
transformer block), packaging, and this model card are customized for
personal/private use.

- Original model: Alpha-VLLM/Lumina-Image-2.0 (Apache 2.0 license)
- Rebrand: M2-Craftly-Lumina by Mushfiqur
- Visibility: Private (shared only with invited collaborators)

## Usage

```python
import torch
from diffusers import Lumina2Pipeline as M2CraftlyLuminaPipeline  # same class, renamed on load

pipe = M2CraftlyLuminaPipeline.from_pretrained("mushfiqur-good-boyyy/m2-craftly-lu", torch_dtype=torch.bfloat16)
pipe.enable_model_cpu_offload()

image = pipe(
    "a cat holding a sign that says hello",
    height=1024,
    width=1024,
    guidance_scale=4.0,
    num_inference_steps=40,
).images[0]
image.save("output.png")
```

Note: this is a 2B-parameter model. A GPU with meaningful VRAM (or CPU
offload, which will be slow) is recommended — a plain 8GB-RAM, CPU-only
machine will struggle.