Complete end-to-end workflow for generating 3D models from text prompts using FLUX.1 and TRELLIS models.

• Step 1: Text to Image: Generate high-quality images from text prompts using FLUX.1 [dev]
• Step 2: Background Removal: Automatically remove backgrounds using rembg
• Step 3: 3D Generation: Convert images to 3D models using Microsoft's TRELLIS
• Interactive Workflow: Each step requires user confirmation before proceeding
• Multiple Output Formats: GLB files and Gaussian Splatting (.ply) files
• Live Preview: Watch images generate in real-time during text-to-image generation

1. Enter your text prompt in the input field
2. Adjust settings (optional):
   • Image dimensions (width/height)
   • Guidance scale and inference steps
   • Enable live preview for real-time generation
   • Choose between fast or quality VAE
3. Click "Generate Image"
4. Review the generated image
5. Click "Confirm & Remove Background" to proceed

1. Review the image with background
2. Click "Remove Background" to process
3. Review the image with transparent background (RGBA)
4. Click "Confirm & Generate 3D" to proceed

1. Adjust 3D generation settings (optional):
   • Sparse Structure Generation parameters (Stage 1)
   • Structured Latent Generation parameters (Stage 2)
   • Mesh simplification and texture size
2. Click "Generate 3D Model"
3. Wait for generation to complete (typically 2-5 minutes)
4. Preview the 3D model in the interactive viewer
5. Download the GLB file or extract Gaussian splatting data

This app requires Hugging Face authentication for FLUX.1 model access:

1. Go to FLUX.1-dev model page and accept the license
2. Create a Hugging Face token at https://huggingface.co/settings/tokens
3. In your Hugging Face Space settings, add a secret named HF_TOKEN with your token value

The code will automatically use the token from the environment variable to authenticate when loading the model.

This app requires the trellis module to be present in the root directory. The trellis module should be copied from the reference implementation or obtained from the TRELLIS repository.

• Be descriptive and specific in your prompts
• Use quality VAE for final output if you have time
• Enable live preview to see progress

• Images with clear subjects work best
• The tool automatically handles background removal

• Use clear, well-lit images with good contrast
• Images with transparent backgrounds (alpha channel) work best
• Ensure the main object is clearly visible and centered
• Higher guidance strength and more sampling steps may improve quality but take longer

• FLUX.1 [dev]: black-forest-labs/FLUX.1-dev
  • 12B parameter rectified flow transformer
  • Non-commercial license
• TRELLIS: microsoft/TRELLIS-image-large
  • Scalable and versatile 3D generation model
• rembg: Automatic background removal

• ZeroGPU (T4): GPU resources are allocated on-demand
• Processing time varies based on GPU availability and image complexity

• GLB: Universal 3D format compatible with most 3D software, game engines, and web viewers
• Gaussian Splatting (.ply): Advanced point-based representation for high-quality rendering (~50MB files)

• FLUX.1 Blog
• TRELLIS Project Page
• TRELLIS Paper

• Processing requires GPU resources - you may need to wait if all GPUs are in use
• Gaussian splatting files can be large (~50MB) and may take time to download
• Each step in the workflow requires user confirmation before proceeding
• The app uses ZeroGPU, so GPU resources are allocated on-demand

• FLUX.1 [dev]: Non-commercial license - see LICENSE
• TRELLIS: MIT License
• This Space: Apache 2.0

Built with Gradio and powered by Hugging Face Spaces