butterfly-ai-labs/Lumina_1.0

Lumina 1.0 (by ButterFly AI Labs) 🦋

Lumina 1.0 is a next-generation, lightweight text-to-image model developed by ButterFly AI Labs. Fine-tuned on top of the classic Stable Diffusion 1.5 architecture, it is meticulously optimized to deliver high-quality, modern aesthetics and sharp details while drastically reducing computational overhead.

True to our mission of creating accessible AI tools, Lumina 1.0 demonstrates that high-performance media generation doesn't require massive budgets or enterprise-level clusters.

📊 Model Specifications & Technical Details

• Base Model: Stable Diffusion 1.5 (U-Net architecture)
• Total Parameters: ~1.2 Billion (highly efficient, ideal for consumer hardware and edge devices)
• Training Cost: Just $7.50 USD 💰 (Achieved via advanced compute optimization, low-rank adaptation techniques, and a highly curated dataset)
• Target Resolution: 512x512 pixels (native), scaling exceptionally well with modern upscalers.
• Primary Focus: Fast rendering, lower VRAM consumption, and enhanced prompt responsiveness.

🎨 Lumina 1.0 Generation Examples

🦋 Example 1	🌆 Example 2	🦄 Example 3
Prompt: Red car	Prompt: Lion in the desert	Prompt: Beautiful house in Jungles

⚙️ Recommended Generation Settings

To get the best possible results out of Lumina 1.0, we highly recommend using the following parameters:

• Resolution: 512x512 pixels (Native)
• Sampling Steps: 30 steps
• CFG Scale (Prompt Guidance): 7.5
• Sampler: Euler a, DPM++ 2M Karras, or UniPC

⚠️ Important Note on Quality: Please keep in mind that Lumina 1.0 is built on an older model architecture (Stable Diffusion 1.5). While it is extremely lightweight, fast, and cost-efficient, the raw generation quality, fine text rendering, and complex anatomies may be lower or less detailed compared to modern, multi-billion-parameter heavyweights like SD3 or Flux. For the best outcomes, we recommend using modern upscalers (Hi-Res Fix) or face-restoration tools.

🚀 How to Use (with Diffusers)

You can easily run Lumina 1.0 using the diffusers library:

from diffusers import StableDiffusionPipeline
import torch

pipe = StableDiffusionPipeline.from_pretrained("ButterFly-AI-Labs/your-model-name", torch_dtype=torch.float16)
pipe = pipe.to("cuda")

prompt = "A cinematic, highly detailed photograph of a butterfly wings glowing in the dark, neon cyberpunk style"
image = pipe(prompt).images[0]
image.save("lumina_output.png")