ChronoEdit CLIP Vision (CRT AutoDL)
The vision encoder for ChronoEdit's image-to-video trick (auto-downloaded)
- CLIP_VISION
ChronoEdit is secretly an image-to-video model: you feed it a starting image, it rolls out a short clip, and the last frame is your edit. That means it needs to see the starting image, and seeing happens through a vision encoder. For Wan-based models that encoder is Wan's CLIP-Vision-H, and it is not the same thing as the CLIP-Vision that ships with SDXL or Flux. This node downloads the right one and hands you a CLIP_VISION output.
What it loads
Zero inputs, one output: CLIP_VISION. On first run it checks ComfyUI/models/clip_vision for clip_vision_h.safetensors and pulls it from Comfy-Org's Wan 2.1 repackaged set if it's missing. That's the same clip_vision_h file every Wan 2.1/2.2 I2V workflow uses, which is convenient: if you already run Wan I2V locally, the file is probably on disk and this node just finds it. No second download, no duplicate. The node validates the file after loading and throws a clear error if you've pointed something odd at it.
Where it sits in the workflow
ChronoEdit CLIP Vision → CLIP Vision Encode → ... conditioning
In a typical ChronoEdit flow the encoded reference image conditions the sampler, and the model generates the 2 or 26 frames from that starting point. You only need this node when you're doing the reference-image flavor of ChronoEdit - for pure text-to-image edits it's not required. The rest of the set is the usual gang: ChronoEdit Model (the fp8 14B), ChronoEdit CLIP - WAN for the text side, ChronoEdit VAE, and the Distill / Upscaler LoRA nodes.
Install
Just the pack. ComfyUI Manager → search CRT-Nodes, or:
cd ComfyUI/custom_nodes
git clone https://github.com/PGCRT/CRT-Nodes
pip install -r requirements.txt
Restart ComfyUI and the node shows up under CRT/AutoDL/ChronoEdit. The model file downloads on first use through the node itself. One practical tip for a first run: each ChronoEdit AutoDL node downloads its own file independently, and this one plus the CLIP, VAE, and 14B model will all flash progress bars on the same first run. It's a lot of waiting the first time and zero waiting after.
Inputs (0)
No inputs
Outputs (1)
| Name | Type | Description |
|---|---|---|
| CLIP_VISION | CLIP_VISION | — |