DocumentationSAM 3 Image-to-3D
SAM 3 Image-to-3D (SAM 3D) is a cutting-edge AI technology that automatically converts a single 2D image into a high-quality 3D model with full textures (GLB format). This tool significantly lowers the barrier to 3D modeling, allowing designers, developers, and creators to generate usable 3D assets without complex modeling software.
Features
- One-Click Generation: Generate a 3D model from an uploaded image in approximately 30 seconds.
- Automatic Segmentation: Integrated with SAM 3's powerful segmentation capabilities, allowing you to precisely select specific objects from an image for 3D conversion.
- High-Quality Textures: AI generates both the mesh geometry and automatically maps optimized surface textures.
- Universal Format Export: Generates standard GLB files that can be directly imported into Blender, Unity, Unreal Engine, or web-based 3D engines.
- Intelligent Interaction: Support for object identification via clicking, text prompts, or AI Agents.
How to Use
1. Upload Image
Click the upload area or drag and drop a file to get started.
- Image Requirements: We recommend using JPG, PNG, or WEBP images with simple backgrounds, even lighting, and clear subjects (max 10MB).
- Best Results: Independent objects (e.g., shoes, toys, furniture, sculptures) yield the best identification and generation results.
2. Analyze and Segment Objects
Before generating a 3D model, you need to tell the AI specific targets you want to convert:
- Upload and Analyze: After uploading, click "Analyze Image." The AI will automatically identify objects.
- Select Target: You can directly click on the segmented color blocks in the image or enter keywords (e.g.,
person,cat) in the sidebar to filter. - AI Agent: If you're unsure how to describe something, switch to "AI Agent" mode and use natural language (e.g.,
Identify the largest sofa in the center) to let the AI lock onto the target.
3. Generate 3D Model
- Confirm Generation: Once the target object is selected, click "Generate 3D Object."
- Wait for Processing: The process usually takes about 30 seconds. The system is performing complex mesh inference and texture baking in the cloud.
- Track Progress: You can wait on the page or leave. The generation task will automatically appear in your "History," and you can check it anytime once complete.
4. Preview and Download
- Real-Time Preview: Once generation is complete, the preview window on the right will display an interactive 3D model. You can rotate and zoom to inspect details.
- Download GLB: Click "Download GLB" to get your model file.
Credits
- Each successful 3D generation task consumes 5 credits.
- If a generation task fails due to technical issues, credits will be automatically refunded.
FAQ
Why does the back of the 3D model look occluded or estimated?
Current AI technology primarily relies on depth inference from a single image. For single-image-to-3D, the back is usually an "estimated" result based on frontal features. It works well for simple, regular objects, but extremely complex objects may require manual fine-tuning.
Is animation supported?
Currently, only static 3D models are supported. You can import your downloaded GLB file into professional software like Mixamo or Blender for rigging and animation.
Can any image be turned into 3D?
As long as the subject is clearly visible and complete (not severely occluded), most independent objects will yield good conversion results.