ChronoEdit - Temporal Image Editing (Patreon)

Video - https://youtu.be/ZdGpgs2qFWY

Overview

ChronoEdit - https://github.com/nv-tlabs/ChronoEdit - TL;DR:

"reframes image editing as a video generation task, using input and edited images as start/end frames to leverage pretrained video models with temporal consistency. A temporal reasoning stage introduces reasoning tokens to ensure physically plausible edits and visualize the editing trajectory"

Or, in other words, it's a modified version of WAN that creates a really short video such that the last frame is the edited image :)

Is it as good as Qwen Image Edit? Maybe. Sometimes. It is certainly quite good at temporal changes, such as making a person jump or pick up an object, but not so good at things like style changes.

Still, it's always good to have another tool in the toolbox!

NB. 26/11/2025 - Seems they are re-organising files. See https://github.com/nv-tlabs/ChronoEdit/issues/23 for more info & alternate links!

Example input:

Prompt: "Change to a side view"

Example image:

Prompt: "the woman is playing a guitar"

Update!

11/Nov: An upscale LoRA has also just been released. Check it out!

https://huggingface.co/nvidia/ChronoEdit-14B-Diffusers-Upscaler-Lora

Workflow

The workflow overview is shown in the screenshot below:

Two versions are attached - the regular and the GGUF (Unet) loader version. If you're super-low on VRAM, then there are loads of GGUF files available from https://huggingface.co/QuantStack/ChronoEdit-14B-GGUF, and there is a Q4 which comes in at under 10GB! If you've got plenty of VRAM, then the standard ComfyUI fp16 file is a pretty beefy 33GB in size but nvidia-smi showed just 20GB VRAM use on my 3090. Tricky to say what the actual minimum VRAM is as the block swap is done automatically.

Not many custom nodes, but I found that using the AI enhanced prompting certainly helped.

This does mean having ollama installed, but with the keep_alive set to 0 it won't use any extra VRAM.

There are boat loads of LLMs with vision capabilities at https://ollama.com/search, not such the ones with the vision tag. Gemma3 is a great option for lower VRAM cards with loads of smaller sizes available, for example.

You don't need to use the AI enhanced prompt, and there's a toggle to turn that group off if you'd prefer.

By default, it's set up for use with the distill LoRA, but if you're testing it without then do check the note for various settings to change.

Have fun!