SenseTime and HKUST release MTEdit: 500K multi-turn image editing trajectories

sensenova · hf · 2026-10-19

SenseTime Research, together with HKUST MMLab, Stanford, and Tsinghua AIR, released MTEdit, a large-scale multi-turn image editing dataset plus benchmark.

• The pipeline centers on packed trajectory canvas (PTC) generation: multiple editing states are generated jointly in one unified visual context, avoiding the error accumulation and visual-quality drift that plague conventional sequential rollouts.
• The dataset contains 500K multi-turn editing trajectories, up to 8 turns each with an average of 5.38 turns, covering both step-wise and context-aware editing scenarios — among the largest resources of its kind.
• The team also introduces MTEdit-Bench to evaluate how well models follow sequences of editing instructions.

Original post →

More from Multimodal

• Training Krea2 LoRA on 16GB VRAM without block swapping, down to 5-7s/it — Corgice · 2026-10-19
• MiniMax Launches Open Multimodal Video Model H3 and M3.1-Flash With 1M Context — testingcatalog · 2026-10-19
• Local Qwen3 27B generates a full video in ~2 hours, completely free — aziz4ai · 2026-10-19
• selflift-Fish: 20s 1080p MiniMax H3 video on a 16GB GPU, 2.2x faster — Hot_Store_5699 · 2026-10-19
• AI-generated series CTVA releases Chapter 3: Disco Fever — cranialtrauma · 2026-10-19
• CTVA AI series drops Chapter 4: The Tallahassee Carrousel — cranialtrauma · 2026-10-19

Multimodal channel →