Marigold V2 Open-Sources Single-Step Dense Prediction from Image-Editing DiT

Marigold V2 adapts the Qwen-Image-Edit DiT into a single-step dense prediction model via 4-bit QLoRA fine-tuning, trainable on a single GPU in days. It sets new depth-estimation SOTA on multiple datasets and is open-sourced under Apache 2.0.

2026-09-14 ~ 2026-09-14 · 2 related posts

• Marigold V2 repurposes diffusion Transformers for dense prediction, trainable on a single GPU in days — RexDouglass · 2026-09-14
• Marigold V2 turns image-editing DiT into single-step depth model, beats SOTA by 16-26% — bdsqlsz · 2026-09-14