Forge a 3D model from words or photos
Describe an object, or upload up to four photos of it from different angles. Forge reconstructs a textured mesh you can orbit, view in AR, and download as GLB — and more views mean a sharper model with no guessed-at back.
Nothing forged yet
Type a description and tap an example, or switch to From photos and drop in up to four views of one object. Generation reconstructs a textured mesh — usually 30–90 seconds.
Stylize
One-click geometric filters. Restyle this model as voxels, bricks, an open lattice, or low-poly — source color is carried over where the style allows.
Generation failed
Something went wrong. Try a simpler, single-subject prompt.
Generator not configured
Text-to-3D needs a Replicate API token on this deployment. Once
REPLICATE_API_TOKEN is set, Forge runs the full flux → TRELLIS
pipeline with no further changes.
Your creations
Powered by flux-schnell + Microsoft TRELLIS on Replicate, with multi-view conditioning when you supply more than one photo. Best results: one clearly described subject, or clean evenly-lit views from distinct angles. The same pipeline is available to agents over MCP at /api/mcp-3d.