TRELLIS.2

TRELLIS.2 is a 4-billion-parameter 3D generative model that transforms 2D images into fully textured 3D assets with complex topology, sharp features, and physically based rendering materials. It uses a novel field-free sparse voxel representation and scalable structured latent space to enable efficient, high-resolution 3D generation from single image input.

Complete technical specification for integration
Ready-to-use code snippets for common workflows
API Options
Platform-level options for task execution and delivery.
taskType
stringrequiredvalue: 3dInferenceIdentifier for the type of task being performed
taskUUID
stringrequiredUUID v4UUID v4 identifier for tracking tasks and matching async responses. Must be unique per task.
outputType
stringdefault: URL3D output type.
Allowed values1 value
outputFormat
stringdefault: GLBSpecifies the file format of the generated output. The available values depend on the task type and the specific model's capabilities.
- `GLB`: Binary glTF format for 3D models, including geometry, textures, and PBR materials.
Allowed values1 value
webhookURL
stringuriSpecifies a webhook URL where JSON responses will be sent via HTTP POST when generation tasks complete. For batch requests with multiple results, each completed item triggers a separate webhook call as it becomes available.
Learn more1 resource
- WebhooksPLATFORM
- Webhooks
deliveryMethod
stringdefault: asyncDetermines how the API delivers task results.
Allowed values2 values
- Returns complete results directly in the API response.
- Returns an immediate acknowledgment with the task UUID. Poll for results using getResponse.
Learn more1 resource
- Task PollingPLATFORM
uploadEndpoint
stringuriSpecifies a URL where the generated content will be automatically uploaded using the HTTP PUT method. The raw binary data of the media file is sent directly as the request body. For secure uploads to cloud storage, use presigned URLs that include temporary authentication credentials.
Common use cases:
- Cloud storage: Upload directly to S3 buckets, Google Cloud Storage, or Azure Blob Storage using presigned URLs.
- CDN integration: Upload to content delivery networks for immediate distribution.
// S3 presigned URL for secure upload https://your-bucket.s3.amazonaws.com/generated/content.mp4?X-Amz-Signature=abc123&X-Amz-Expires=3600 // Google Cloud Storage presigned URL https://storage.googleapis.com/your-bucket/content.jpg?X-Goog-Signature=xyz789 // Custom storage endpoint https://storage.example.com/uploads/generated-image.jpgThe content data will be sent as the request body to the specified URL when generation is complete.
ttl
integermin: 60Time-to-live (TTL) in seconds for generated content. Only applies when
outputTypeisURL.
includeCost
booleanInclude task cost in the response.
numberResults
integermin: 1max: 4default: 1Number of results to generate. Each result uses a different seed, producing variations of the same parameters.
Inputs
Input resources for the task (images, audio, etc). These must be nested inside the inputs object.
inputs object.Core Parameters
Primary parameters that define the task output.
model
stringrequiredvalue: microsoft:trellis-2@4bIdentifier of the model to use for generation.
Learn more3 resources
Settings
Technical parameters to fine-tune the inference process. These must be nested inside the settings object.
settings object.settings»resolutionresolution
integerdefault: 1024Voxel grid resolution for 3D generation. Higher values capture finer geometric detail.
Allowed values3 values
settings»alphaModealphaMode
stringdefault: OPAQUEMaterial alpha mode for the output mesh.
Allowed values3 values
settings»decimationTargetdecimationTarget
integermin: 5000max: 1000000default: 500000Target polygon count for mesh simplification. Lower values produce lighter models at the cost of geometric detail.
settings»dracoCompressiondracoCompression
objectApply KHR_draco_mesh_compression to the output GLB. Reduces geometry size 5-10x with no triangle loss.
Properties5 properties
settings»dracoCompression»enabledenabled
booleanCompress the output GLB geometry with Draco.
settings»dracoCompression»levellevel
integermin: 0max: 10default: 7Encoding effort for Draco compression. Higher values reduce file size further but encode more slowly.
settings»dracoCompression»quantizationNormalquantizationNormal
integermin: 8max: 30default: 14Quantization bits for vertex normals. Higher values produce smoother shading at the cost of file size.
settings»dracoCompression»quantizationPositionquantizationPosition
integermin: 8max: 30default: 16Quantization bits for vertex positions. Higher values preserve more positional detail at the cost of file size.
settings»dracoCompression»quantizationTexCoordquantizationTexCoord
integermin: 8max: 30default: 14Quantization bits for texture coordinates. Higher values keep textures sharper at UV seams at the cost of file size.
settings»meshClustermeshCluster
objectUV chart clustering settings for the output mesh.
Properties4 properties
settings»meshCluster»globalIterationsglobalIterations
integermin: 0max: 64default: 1Global iterations for chart clustering.
settings»meshCluster»refineIterationsrefineIterations
integermin: 0max: 64default: 0Local refinement iterations for chart clustering.
settings»meshCluster»smoothStrengthsmoothStrength
integermin: 0max: 10default: 1Smoothing strength applied during chart clustering.
settings»meshCluster»thresholdConeHalfAngleRadthresholdConeHalfAngleRad
floatmin: 0max: 3.141592653589793default: 1.5707963267948966Cone half-angle threshold in radians used during UV chart clustering.
settings»remeshremesh
booleandefault: trueWhether to remesh the output for cleaner topology.
settings»remeshBandremeshBand
floatmin: 0max: 16default: 1Narrow-band width in voxels used for dual-contouring remeshing.
settings»remeshProjectremeshProject
floatmin: 0max: 1default: 0Projection factor for snapping remeshed vertices back to the original surface.
settings»shapeSlatshapeSlat
objectSampler for the Structured Latent stage that refines the 3D geometry.
Properties6 properties
settings»shapeSlat»stepssteps
integermin: 1max: 50default: 12Number of denoising steps. More steps improve quality but increase generation time.
settings»shapeSlat»guidanceIntervalEndguidanceIntervalEnd
floatmin: 0max: 1default: 1End of the sampling schedule fraction over which guidance is applied.
settings»shapeSlat»guidanceIntervalStartguidanceIntervalStart
floatmin: 0max: 1default: 0.6Start of the sampling schedule fraction over which guidance is applied.
settings»shapeSlat»guidanceRescaleguidanceRescale
floatmin: 0max: 1default: 0.5Dampens oversaturation from high guidance values. Lower values apply more correction.
settings»shapeSlat»guidanceStrengthguidanceStrength
floatmin: 1max: 10default: 7.5How closely the output follows the input image. Higher values enforce stricter adherence to the visible geometry.
settings»shapeSlat»rescaleTrescaleT
floatmin: 1max: 6default: 3Timestep rescaling factor for the sampling schedule. Controls how guidance strength is distributed across steps.
settings»sparseStructuresparseStructure
objectSampler for the initial sparse voxel grid that establishes the basic 3D geometry.
Properties6 properties
settings»sparseStructure»stepssteps
integermin: 1max: 50default: 12Number of denoising steps. More steps improve quality but increase generation time.
settings»sparseStructure»guidanceIntervalEndguidanceIntervalEnd
floatmin: 0max: 1default: 1End of the sampling schedule fraction over which guidance is applied.
settings»sparseStructure»guidanceIntervalStartguidanceIntervalStart
floatmin: 0max: 1default: 0.6Start of the sampling schedule fraction over which guidance is applied.
settings»sparseStructure»guidanceRescaleguidanceRescale
floatmin: 0max: 1default: 0.7Dampens oversaturation from high guidance values. Lower values apply more correction.
settings»sparseStructure»guidanceStrengthguidanceStrength
floatmin: 1max: 10default: 7.5How closely the output follows the input image. Higher values enforce stricter adherence to the visible geometry.
settings»sparseStructure»rescaleTrescaleT
floatmin: 1max: 6default: 5Timestep rescaling factor for the sampling schedule. Controls how guidance strength is distributed across steps.
settings»texSlattexSlat
objectSampler for the Structured Latent stage that generates surface appearance.
Properties6 properties
settings»texSlat»stepssteps
integermin: 1max: 50default: 12Number of denoising steps. More steps improve quality but increase generation time.
settings»texSlat»guidanceIntervalEndguidanceIntervalEnd
floatmin: 0max: 1default: 0.9End of the sampling schedule fraction over which guidance is applied.
settings»texSlat»guidanceIntervalStartguidanceIntervalStart
floatmin: 0max: 1default: 0.6Start of the sampling schedule fraction over which guidance is applied.
settings»texSlat»guidanceRescaleguidanceRescale
floatmin: 0max: 1default: 0Dampens oversaturation from high guidance values. Lower values apply more correction.
settings»texSlat»guidanceStrengthguidanceStrength
floatmin: 1max: 10default: 1How closely the output follows the input image. Higher values enforce stricter adherence to the visible geometry.
settings»texSlat»rescaleTrescaleT
floatmin: 1max: 6default: 3Timestep rescaling factor for the sampling schedule. Controls how guidance strength is distributed across steps.
settings»textureFormattextureFormat
stringdefault: WEBPTexture format for the output mesh.
Allowed values2 values
settings»textureSizetextureSize
integermin: 1024max: 4096step: 1024default: 2048Resolution of the output texture map in pixels. Higher values produce finer surface detail.