Marigold V2 launches at SIGGRAPH Asia 2026: sharp diffusion-transformer depth estimation
AntonObukhov1 · x · 2026-09-09
Anton Obukhov announced Marigold V2, accepted to SIGGRAPH Asia 2026.
- V1 recap: Marigold post-trains an image generator into a depth estimator on a single GPU — an accessible research recipe.
- V2 upgrade: the backbone moves to a diffusion transformer (DiT), producing very sharp edges with stronger versatility.
Paper, code, weights, and a demo are all available, and the author has shared a low-cost fine-tuning path (4-bit quantization + QLoRA on one consumer GPU).
Related event: Marigold V2: A Single-GPU DiT Depth Estimator Hits SIGGRAPH Asia 2026(11 posts)→
More from Multimodal
- ComfyUI Krea2 Prompt Styler Node Revived in a Fork With a Massive Style Library — Winter_unmuted · 2026-10-27
- Single HTML file generates retro poster designs in-browser with zero LLM calls — Afinetheorem · 2026-10-27
- A Redditor Used Copilot to Imagine and Archive Their Own '80s/'90s Winter Streetwear Concepts — Valuable-Sky2134 · 2026-10-27
- Transkun open-source piano transcription model nails MIDI, sustain events included — alexanderchen · 2026-10-27
- fal adds FLUX 3 Image Klein LoRA training: 10 reference images, 3 LoRAs per request — OdinLovis · 2026-10-27
- Guy turns his girlfriend's selfie into a playable AI game ad about popping pimples — CurieuxExplorer · 2026-10-27