Nova3D
Nova3D open-source client that generates editable, part-aware 3D assets from text prompts or reference images via a Blender Python script pipeline.
What is Nova3D?
Nova3D is an open-source client for nova3d.xyz that generates 3D assets as executable construction procedures rather than a single, merged mesh. It produces Blender-native Python scripts and returns a structured GLB where parts are named and separately addressable.
The core purpose of this client is to connect to the hosted Nova3D pipeline (by default) so you can generate editable, part-aware 3D outputs from either text prompts or reference images.
Key Features
- Editable, part-aware 3D outputs: The pipeline outputs a structured GLB with named, separately addressable parts, designed for targeted edits (e.g., changing one part without re-generating everything).
- Script-native generation (prompt-to-code / image-to-code): Instead of diffusion-style “image-to-3D” that yields a single merged mesh, Nova3D generates Blender Python scripts that represent the construction procedure.
- Uses Blender’s scene graph as the native representation: The client aligns its output model with Blender’s scene graph, which supports a logical hierarchy and scene-level editing workflows.
- Model/LLM agnostic client workflow: The generation harness can switch between LLM backends via settings (e.g., Claude 3.5, GPT-4o, or Gemini 1.5 Pro), with validation/execution handled by the pipeline.
- Integrated web viewport editor: Includes a built-in Three.js editor with transform tools, snapping, and material editing.
- Reference-image guidance: Lets you attach a photo to guide the spatial logic of the generated script.
- Local in-browser caching: Models are cached in the browser so you can view your history even after remote URLs expire.
How to Use Nova3D
- Run the client locally: Clone the repository, install dependencies (
flutter pub get), and start the local web UI.- The quick start notes that port 5555 is required for OAuth redirect authorization.
- Sign in: Open the local URL, then sign in using Google or Email.
- Add an API key in settings: The client supports adding API keys for OpenAI, Anthropic, or Gemini.
- Provide input and generate: Enter a prompt (and optionally attach a reference image) and run the generation flow.
- Edit and review results: Use the integrated viewport editor for transforms/snapping/material editing, and rely on local caching to revisit previous generations.
To deploy a production build for static hosting, the repository includes a command to build the Flutter web app (flutter build web --release).
Use Cases
- Editable asset generation for workflows that require hierarchy: Generate 3D models where parts are named and parented so you can adjust individual components without regenerating the full result.
- Prompt-driven or photo-guided 3D ideation: Use text prompts or attach reference images to influence the spatial logic of the generated construction scripts.
- Material iteration with a PBR-oriented mapping workflow: Work with outputs designed to support material mapping rather than relying on baked vertex colors.
- Rapid review and adjustment in a browser: Use the integrated Three.js viewport editor (transform tools, snapping, and material editing) without leaving the client.
- Local browsing of previous generations: Leverage in-browser caching to keep viewing history even when remote URLs expire.
FAQ
-
Does Nova3D output a single merged mesh? No. The pipeline is described as producing a structured GLB with separately addressable parts, generated from Blender-native Python scripts.
-
Can I use different LLM providers? Yes. The client is described as model-agnostic, with settings allowing switching between backends such as Claude 3.5, GPT-4o, or Gemini 1.5 Pro.
-
What local setup is needed for sign-in? The quick start notes that port 5555 is required for OAuth redirect authorization, and it specifically warns that using
localhost:5555can cause Google Sign-In to fail due to OAuth origin policies. -
Where does the client send generation requests by default? By default, it communicates with the hosted nova3d.xyz API. A custom backend can be targeted by setting
API_BASE_URLvia--dart-define. -
Is there a way to influence generation with an image? Yes. The client supports reference images so you can attach a photo to guide the spatial logic of the generated script.
Alternatives
- Image-to-3D diffusion tools that output a single merged mesh: These tend to focus on producing one consolidated geometry result, which differs from Nova3D’s part-aware, editable outputs.
- OpenSCAD/CSG-style systems focused on solid modeling: Approaches in this category can prioritize manifold solids, but may trade off organic-shape flexibility and detailed hierarchy/material editability.
- Prompt-to-CAD or procedural CAD tools: If your primary goal is editable procedural construction, look for tools that support script-like or parametric workflows rather than purely generative mesh outputs.
- Standalone 3D modeling/editing tools paired with importable procedural results: For workflows where you already use Blender or other DCC tools, you can treat Nova3D as the generation step and then refine outputs in a modeling environment—though the key difference is Nova3D’s generation of Blender-native scripts and part-aware GLB structure.
Alternatives
promptoMANIA
promptoMANIA is an AI art prompt generator that helps users create detailed prompts for various text-to-image diffusion models.
改图鸭
The AI painting generator is an online AI-powered software that automatically creates artwork based on text descriptions provided by users.
Prompty Town
Prompty Town is a tiny internet city of links—buy a tile, attach your link, prompt it with text/content, and let others browse.
Creativly
Creativly is a browser-based creative tool that helps you create without writing prompts—generate creative outputs fast with a simple workflow.
DeepMotion
DeepMotion is an AI motion capture and body-tracking platform to generate 3D animations from video (and text) in your web browser, via Animate 3D API.
Edgee
Edgee is an edge-native AI gateway that compresses prompts before LLM providers, using one OpenAI-compatible API to route 200+ models.