ChronoEdit - AI image editing framework jointly open-sourced by NVIDIA and the University of Toronto

What is ChronoEdit

ChronoEdit, an open source AI image editing framework jointly developed by NVIDIA and the University of Toronto, redefines the image editing task as a video generation task to ensure temporal and physical consistency of editing results. By distilling temporal a priori knowledge from a 14B-parameter pre-trained video generation model, ChronoEdit splits the inference process into two phases, video inference and contextual editing, to realize image editing driven by temporal inference. Complex editing tasks such as viewpoint transformation, pose rotation and physical interaction simulation are supported.

Features of ChronoEdit

• Temporal inference-driven image editing: redefines the image editing task as a video generation task, ensures temporal and physical consistency of the editing results through temporal reasoning, and is capable of handling complex editing tasks such as viewpoint transformations, pose rotations, and physical interaction simulations.
• Customized Timing Denoising Diffusion Transformer Architecture: A pre-trained video generation model based on a 14B parameter supports efficient inference and high quality editing results.
• Supports physically aware image editing and action-condition based world simulation: Physical interactions and motion changes can be simulated to make editing results more realistic and natural.
• Complete open source framework: Provides inference using Diffusers and LoRA fine-tuning using DiffSynth-Studio, supporting distributed inference and large-scale fine-tuning.
• Multiple model weights and training frameworksThe model weights are available in different scales such as ChronoEdit-14B and ChronoEdit-2B, as well as 8-step distillation LoRA model weights to satisfy the needs of different users.
• Easy-to-use command line operations: With simple command line commands, users can quickly realize high-quality image editing tasks without complex configuration and operation.

ChronoEdit's Core Benefits

• Innovative temporal reasoning mechanisms: By transforming image editing into a video generation task and utilizing temporal reasoning to ensure that the editing results are temporally and physically consistent, it solves the incoherence problem common in traditional image editing.
• Powerful pre-trained model base: A pre-trained video generation model based on a 14B parameter with powerful generative capabilities and rich a priori knowledge of timing, capable of handling complex editing tasks such as viewpoint transformation, pose rotation and physical interaction simulation.
• Efficient inference performance: Denoising through customized timing Diffusion Transformer With its architecture and optimized inference process, ChronoEdit is able to achieve efficient inference speed while maintaining high quality output.
• Support for physical perception and motion simulation: The ability to simulate physical interactions and motion changes makes editing results more realistic and natural for advanced image editing tasks that require physical consistency.
• Flexible fine-tuning capabilities: Providing the ability to fine-tune LoRA using DiffSynth-Studio, users can customize the training of the model to suit their needs for specific editing tasks and datasets.
• Complete open source framework: Providing complete training and inference code with support for distributed inference and large-scale fine-tuning, it provides researchers and developers with powerful tools to facilitate further research and development.
• easy-to-use: With simple command line operation, users can quickly realize high-quality image editing tasks without complex configuration and operation, lowering the threshold of use.
• Multiple model optionsThe model weights of different scales such as ChronoEdit-14B and ChronoEdit-2B, as well as 8-step distillation LoRA model weights are provided to meet the diversified needs of different users in terms of performance and resource consumption.

What is ChronoEdit's official website

• Project website:: https://research.nvidia.com/labs/toronto-ai/chronoedit/
• Github repository:: https://github.com/nv-tlabs/ChronoEdit
• HuggingFace Model Library:: https://huggingface.co/nvidia/ChronoEdit-14B-Diffusers
• arXiv Technical Paper:: https://arxiv.org/pdf/2510.04290

Who is ChronoEdit for?

• Professional image editorsChronoEdit helps photographers, graphic designers, etc., who need to perform high-quality image editing and have high demands on the physical consistency and realism of the editing results, to accomplish complex image editing tasks, such as perspective changes, pose adjustments, etc., in a more efficient manner.
• Video content creatorsChronoEdit provides video creators with a convenient tool to convert image editing tasks into video generation tasks, ensuring that the edited images are consistent in time sequence.
• Artificial intelligence researchers: ChronoEdit provides a complete open-source framework and a variety of model weights, which can be used by researchers for further research and development to explore more possibilities in the field of image editing and video generation, such as improving model architectures and optimizing inference algorithms.
• Machine Learning Engineer: Distributed inference and large-scale fine-tuning can be performed using ChronoEdit's training framework and codebase to adapt to specific application scenarios and datasets to develop an image editing solution that meets specific needs.
• Developers interested in image editing and AI technology: ChronoEdit's ease of use and powerful features make it a great tool for developers to learn and practice image editing techniques, get started quickly with simple command line operations, and explore the application of AI in image editing.