blendi-remade/falcraft

3D structure generation in Minecraft, powered by fal.ai

What it solves

falcraft allows Minecraft players to generate complex 3D structures and objects directly inside their game world using AI, removing the need to manually build intricate models block-by-block or use external 3D modeling software.

How it works

The mod connects to fal.ai cloud APIs to transform text prompts into in-game structures through a multi-step pipeline:

  1. Image Generation: A 2D image is created from a text prompt using models like Z-Image Turbo or Nano Banana Pro.
  2. 3D Conversion: The image is converted into a 3D GLB model using pipelines such as SAM-3D or Hunyuan 3D.
  3. Voxelization: The 3D mesh is converted into Minecraft blocks via triangle-voxel intersection.
  4. Color Mapping: A block mapper uses CIE-LAB perceptual color space to match the model's colors to a palette of over 100 Minecraft blocks.
  5. Placement: A "ghost preview" allows the player to rotate and move the structure before it is animatedly built layer-by-layer.

Additionally, the "FalTV" feature uses the MiniMax H3 Max Director API via a local sidecar bridge to stream live-generated AI video onto in-game walls.

Who it’s for

Minecraft players using version 1.21.1 with the Fabric Loader who want to quickly add detailed AI-generated architecture, creatures, or objects to their worlds.

Highlights

  • Multiple Generation Modes: Includes a standard generate mode, a high-quality UV-textured "Craft" mode, and a "Stream" mode for real-time geometry previews.
  • FalTV: A live-generated AI television channel that can be steered via chat commands.
  • Interactive Placement: Full control over rotation (yaw, pitch, roll) and translation of generated structures before final placement.
  • Automated Voxelization: Seamlessly converts high-fidelity 3D AI models into block-based Minecraft geometry.

Related

  • Project
    NetManAIOps/ChatTS

    ChatTS is a multimodal LLM designed for native understanding and reasoning over multivariate time series, allowing users to perform conversational question-answering on numerical data.

  • Project
    tonyd2wild/DeepSeek-v4-Flash-Vision-Exp-DSpark-1M-NVFP4-KV-2x-DGX-Spark

    A deployment recipe for DeepSeek-V4-Flash-Vision-Exp on DGX Spark clusters, enabling native vision support, 1M token context, and high-throughput speculative decoding via vLLM.

  • Project
    KangLiao929/Puffin

    A series of unified multimodal models for 3D world modeling that enable camera-centric spatial intelligence and 3D world generation using native physics, geometry, and appearance states.

  • Project
    PhiloLabs/fable51-worlds

    A system that uses AI agent swarms to generate verifiable, walkable 3D worlds from text, images, or video, rendered as code-based Three.js applications.