from teamwork.pipelines import TeamworkPipeline
from PIL import Image

pipe = TeamworkPipeline.from_checkpoint(
    'samsartor/teamwork-release',
    'decomposition_heterogeneous_sd3.safetensors',
).to('cuda')

generated = pipe({'image': Image.open('./demo/red_glass_sphere.png'})

Teamwork's safetensors files contain (roughly) the parameters of LoRAs to apply to a base model. However, in addition to adapting the base model for a new task, Teamwork enables the model to accept any number of input images and then collaboratively generate any number of outputs. The available models are:

	Description	Base Model	Resolution	Inputs	Outputs
decomposition_heterogeneous_sd3.safetensors	Intrinsic image decomposition trained on heterogeneous datasets (InteriorVerse, HyperSim, CGIntrinsics, Infinigen, ABC).	SD3	1024x1024	image	diffuse specular roughness normals depth albedo inverseshading diffuseshading residual
decomposition_interiorverse_sd3.safetensors	Intrinsic image decomposition specialized for InteriorVerse dataset.	SD3	640x480	image	diffuse specular roughness normals depth albedo inverseshading
inpainting_sd3.safetensors	Text-conditional inpainting of masked regions.	SD3	1024x1024	mask masked	image
neuralshading_sd3.safetensors	Synthesizes images from intrinsic components under unknown random lighting.	SD3	1024x1024	albedo diffuse specular roughness normals depth	image diffuseshading inverseshading residual
colocated_svbrdf_estimation_flux_256.safetensors	Estimates material properties from a single flash-lit photograph.	FLUX.1-dev	256x256	image halfway	diffuse specular roughness normals
colocated_svbrdf_estimation_flux_512.safetensors	Estimates material properties from a single flash-lit photograph.	FLUX.1-dev	512x512	image halfway	diffuse specular roughness normals
colocated_svbrdf_estimation_sd3_256.safetensors	Estimates material properties from a single flash-lit photograph.	SD3	256x256	image halfway	diffuse specular roughness normals
colocated_svbrdf_estimation_sd3_512.safetensors	Estimates material properties from a single flash-lit photograph.	SD3	512x512	image halfway	diffuse specular roughness normals
colocated_svbrdf_estimation_sdxl_256.safetensors	Estimates material properties from a single flash-lit photograph.	SDXL	256x256	image halfway	diffuse specular roughness normals

Citation

@conference{Sartor:2025:TCD,
    author    = {Sartor, Sam and Peers, Pieter},
    title     = {Teamwork: Collaborative Diffusion with Low-rank Coordination and Adaptation},
    month     = {December},
    year      = {2025},
    booktitle = {ACM SIGGRAPH Asia Conference Proceedings},
}