ChronoEdit: Image and Video Editing with Temporal Reasoning
ChronoEdit by NV Labs is an ICLR 2026 image and video editing model built on Wan2.1. Reframes editing as video generation using distillation LoRAs for temporally consistent edits.
ChronoEdit
Wan2.1Image EditingVideo EditingTemporal ReasoningICLR 2026ICLR 2026 image and video editing model built on Wan2.1. Reframes image editing as a video generation task using input and edited images as start/end frames with distillation LoRAs for temporal consistency.
| Developer | NV Labs (NVIDIA) |
| Architecture | Wan2.1-based with temporal reasoning |
| Model Type | Image/video editing with distillation LoRA |
| Capabilities | Image editing, video editing, world simulation |
| Conference | ICLR 2026 |
Overview
ChronoEdit by NV Labs (NVIDIA) is an ICLR 2026 paper that reframes image editing as a video generation task. Using input and edited images as start and end frames, it leverages pretrained video models (Wan2.1) with temporal consistency to produce coherent edits. The model uses distillation LoRAs for efficient inference, enabling high-quality edits with temporal awareness.
The approach enables both image editing with temporal reasoning and video editing capabilities, making it suitable for applications requiring consistent modifications across frames.
ComfyUI Integration
ChronoEdit is compatible with ComfyUI workflows. The model is available through the Diffusers pipeline as nvidia/ChronoEdit-14B-Diffusers. For best results, the paintbrush LoRA can be combined with the 8-step distillation LoRA.
Guides and workflows related to this model series.