fishwowater / trellis-image

SOTA image-to-3D generator TRELLIS. Equipped with **All image-condition types** you need: (1) single-image (2) multi-images (3) different images for geometry and texture generation (4) mesh+images for detail variation(texture painting)

  • Public
  • 157 runs
  • L40S
  • GitHub
  • Weights
  • Paper
  • License

Input

*file[]

List of input images to generate 3D asset from

file[]

List of images to generate the texture of 3D asset from

file

Input mesh (Z-UP, ONLY required for the detail variation mode)

integer

Random seed for generation

Default: 0

boolean

Randomize seed

Default: true

boolean

Generate color video render

Default: true

boolean

Generate normal video render

Default: false

boolean

Generate 3D model file (GLB)

Default: false

boolean

Save Gaussian point cloud as PLY file

Default: false

boolean

Return the preprocessed images without background

Default: false

number
(minimum: 0, maximum: 10)

Stage 1: Sparse Structure Generation - Guidance Strength

Default: 7.5

integer
(minimum: 1, maximum: 50)

Stage 1: Sparse Structure Generation - Sampling Steps

Default: 12

number
(minimum: 0, maximum: 10)

Stage 2: Structured Latent Generation - Guidance Strength

Default: 3

integer
(minimum: 1, maximum: 50)

Stage 2: Structured Latent Generation - Sampling Steps

Default: 12

number
(minimum: 0.9, maximum: 0.98)

GLB Extraction - Mesh Simplification (only used if generate_model=True)

Default: 0.95

integer
(minimum: 512, maximum: 2048)

GLB Extraction - Texture Size (only used if generate_model=True)

Default: 1024

string

Mode for baking the texture (defaults to rgb, which will be darker)

Default: "rgb"

Output

model_file

Loading...

color_video

Generated in

This output was created using a different version of the model, fishwowater/trellis-image:8fcea9ca.

Run time and cost

This model runs on Nvidia L40S GPU hardware. We don't yet have enough runs of this model to provide performance information.

Readme

SOTA image-to-3D generator TRELLIS. Equipped with all image-condition types you need: (1) single-image (2) multi-images (3) different images for geometry and texture generation (4) mesh+images for detail variation(texture painting)

This deployment is based on a fork of a fork of Trellis: https://github.com/firtoz/TRELLIS