Marigold V2 released: DiT-based depth estimator trained on a single consumer GPU

AntonObukhov1 · x · 2026-09-09

Marigold V2 (to appear at SIGGRAPH Asia 2026) is out. V1 post-trained an image generator into a depth estimator on a single GPU; V2 upgrades to a diffusion transformer, producing very sharp edges and covering multiple modalities: depth, surface normals, and albedo. Paper, code, weights, and demo are all publicly available.

Related event: Marigold V2 Released: Single-GPU Fine-tuned DiT for Depth Estimation(5 posts)→

Original post →

More from Multimodal

Multimodal channel →