Image-to-3D Generation
Convert a reference image into a textured 3D asset. The fal.ai TRELLIS.2 endpoint currently accepts an image URL as its required input and returns a GLB model.
Explore Microsoft's TRELLIS.2 image-to-3D technology and see example results. While TRELLIS.2 generation is being prepared for Trellis3D, new users can try one free generation with our basic TRELLIS 3D model.
New users get one free TRELLIS 3D basic generation after signing up.
رفع الصور
أو اسحب وأفلت الصورة هنا
حد أقصى 20 ميجابايت لكل ملف
معاينة النموذج ثلاثي الأبعاد
في انتظار إنشاء النموذج ثلاثي الأبعاد
See how the existing Trellis3D image-to-3D workflow presents generated assets. Important: the current Trellis 3D V2 preview is associated with the site's existing Tripo-based model configuration, so it should not be treated as a verified output from Microsoft's TRELLIS.2 model.

Existing Trellis3D V2 demo preview — not verified as an output from Microsoft's TRELLIS.2 API. Replace this asset with a verified TRELLIS.2 example when available.
TRELLIS.2 is an image-to-3D generative model released by Microsoft Research. The fal.ai endpoint exposes controls for output resolution, geometry guidance, texture generation, mesh decimation, remeshing, and UV processing.
Convert a reference image into a textured 3D asset. The fal.ai TRELLIS.2 endpoint currently accepts an image URL as its required input and returns a GLB model.
The fal.ai endpoint documents output resolutions of 512, 1024, and 1536, with 1024 as the default. Higher resolutions take longer and may produce more detailed results.
TRELLIS.2 uses an O-Voxel representation designed to model complex shapes, including open surfaces and non-manifold geometry. The representation is a model capability; individual outputs may still need inspection or cleanup.
The endpoint exposes texture-size options of 1024, 2048, and 4096, with 2048 as the default, together with texture-generation controls for balancing image fidelity and artifacts.
The API includes a target vertex count from 5,000 to 2 million (500,000 by default), remeshing controls, and UV-unwrapping settings. Remeshing is enabled by default. These are processing options, not a guarantee that every output is production-ready.
fal.ai also provides a separate TRELLIS.2 Retexture endpoint. It accepts an existing GLB, OBJ, PLY, or STL mesh together with a reference image, then returns a textured GLB. This is a separate 3D-to-3D workflow and is not integrated into Trellis3D yet.
These specifications describe the fal.ai API. They are not currently selectable in the Trellis3D generator on this page.
You can explore the TRELLIS.2 demo today and use the existing TRELLIS 3D generator while the dedicated TRELLIS.2 integration is being prepared.
Review the demo preview and model examples to understand the image-to-3D workflow and expected output.
Create an account and use your one free generation with the existing TRELLIS 3D basic model. This free generation is not a TRELLIS.2 API generation.
Inspect the generated asset in the existing 3D viewer and download the available output files.
A clear reference image gives image-to-3D models more useful visual information. These practical guidelines can help you prepare your input.
Choose a sharp image where the main object is fully visible and not cut off by the image edges.
A clean background can make the object's silhouette easier to distinguish from surrounding details.
Keep important parts visible where possible. Multiple overlapping objects, strong shadows, and reflections can make reconstruction harder.
A single image does not show every side of an object. Generated backsides and unseen surfaces are plausible reconstructions, not guaranteed copies of the real object.
A useful 3D workflow includes more than generation. Preview the result, inspect its surfaces, and choose an export format that fits your next step.
Use the existing Trellis3D viewer to rotate and inspect generated models before downloading.
The existing generator provides downloadable model files and format conversion options. Available formats depend on the generated result and export workflow.
Look for missing surfaces, unwanted internal geometry, texture issues, and unexpected transparency before using a model in another application.
Assets intended for games, animation, or 3D printing may need additional topology cleanup, polygon reduction, UV adjustments, or printability checks.
Create starting points for props, objects, and environment assets. Game-ready delivery may require additional optimization.
Explore turning product references into 3D assets for interactive product presentation and design review.
Use image-to-3D generation to accelerate early design exploration and create editable starting models.
Create a starting model from a visual reference, then check watertightness, wall thickness, and mesh integrity before printing.
TRELLIS.2 and Hunyuan3D are both image-to-3D model families. The best choice depends on the input, the type of object, the output requirements, and the workflow you plan to use.
Use identical reference images and comparable output settings when evaluating the two models.
Compare overall shape, fine details, backside reconstruction, surface continuity, and unwanted internal geometry.
Check material appearance, texture details, file size, and how each GLB behaves in your target 3D application.
Trellis3D has not published a controlled benchmark comparing these models. Avoid treating demo images from different inputs as a fair quality comparison.
TRELLIS.2 is an image-to-3D generative model released by Microsoft Research. It creates 3D assets from reference images and supports geometry and material generation.
Yes. TRELLIS.2 was released by Microsoft Research. Its official project and model resources are available through Microsoft's project page, GitHub, and Hugging Face.
Not yet. This page currently provides TRELLIS.2 demo information and examples. The dedicated fal.ai TRELLIS.2 generation integration will be added separately. You can use the existing basic TRELLIS 3D generator in the meantime.
Yes. New users can sign up and receive one free generation with the basic TRELLIS 3D model. This free generation is for the basic TRELLIS 3D model, not the TRELLIS.2 API.
The current fal.ai image-to-3D endpoint documents output resolutions of 512, 1024, and 1536. Availability in Trellis3D will depend on the future integration.
The fal.ai endpoint documents texture sizes of 1024, 2048, and 4096. Larger textures can increase detail and file size.
The fal.ai image-to-3D endpoint described here requires a single image_url input. Do not assume multi-view reconstruction is supported by this endpoint.
Yes. The fal.ai image-to-3D endpoint returns a generated GLB file. Trellis3D's dedicated integration is not enabled yet.
GLB is supported by Blender and many other 3D applications. Always inspect geometry, materials, and textures after importing, as some assets may need further editing.
No. Generated geometry may need retopology, polygon reduction, UV adjustments, mesh repair, or printability checks depending on the target workflow.
Trellis3D automatically refunds the credits for failed generation tasks. The task status and result are available through the existing account workflow.
Official resources include the Microsoft project page at https://microsoft.github.io/TRELLIS.2/, GitHub at https://github.com/microsoft/TRELLIS.2, and the Hugging Face model at https://huggingface.co/microsoft/TRELLIS.2-4B.
fal.ai provides a separate Retexture endpoint that accepts a mesh file (GLB, OBJ, PLY, or STL) and a reference image, then returns a GLB with generated textures. This feature is not yet integrated into Trellis3D.
Ready to Create a 3D Model?
Sign up to try one free generation with the basic TRELLIS 3D model, or explore the TRELLIS.2 demo above.
Try TRELLIS 3D Free