ChronoEdit CLIP - WAN (CRT AutoDL)

ChronoEdit is a Wan model in disguise, and this node is the proof. NVIDIA built ChronoEdit-14B as an image editor on top of Wan's I2V architecture, which means it doesn't use an SD-style CLIP. It needs Wan's own text encoder: the UMT5-XXL. If you drop a generic CLIP loader into your ChronoEdit workflow, nothing sensible comes out - the tokenizer and encoder are just different.

CRTAutoDLChronoEditCLIP exists so you never have to think about any of that. It has zero inputs. You drop it in, it makes sure the right file exists, and it hands you a CLIP output.

What it actually loads

On first run the node looks in ComfyUI/models/text_encoders for umt5_xxl_fp8_e4m3fn_scaled.safetensors, and if it's not there it downloads it from Comfy-Org's Wan 2.1 repackaged set on HuggingFace. That's the fp8-scaled UMT5-XXL, the encoder every Wan 2.1/2.2 workflow uses for text. Because ChronoEdit shares Wan's DNA, this is the encoder it wants - same tokenizer, same embedding space. This is also why the node's display name says "WAN" on it, which now makes sense.

The file is a few GB and fp8-scaled rather than fp16, so it's the reasonable memory tradeoff: text encoders are pure throughput, and running UMT5-XXL in fp8 barely costs quality while saving a chunk of VRAM on a model that's already eating ~14GB.

Wiring it up

Output CLIP goes straight into your standard CLIP Text Encode nodes for the positive and negative prompts, just like any other encoder. In a ChronoEdit flow that's:

ChronoEdit CLIP - WAN  →  CLIP Text Encode (pos/neg)  →  conditioning

If you're doing a reference-image edit (the 26-frame I2V flavor), you'll also want the pack's ChronoEdit CLIP Vision node on the image side, plus ChronoEdit Model, ChronoEdit VAE, and ChronoEdit Distill LoRA. Every one of those downloads its own file on first use - this node is the first of several "wait, another download?" moments in your first ChronoEdit run.

Install

Same pack as everything else here. ComfyUI Manager, search CRT-Nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/PGCRT/CRT-Nodes
pip install -r requirements.txt

Restart ComfyUI and the node appears under CRT/AutoDL/ChronoEdit. One honest note: the pack's requirements.txt is long and pulls in heavy stuff (opencv, transformers, audio tooling) for nodes you may never touch - the AutoDL CLIP loader itself just needs huggingface_hub, which ComfyUI already has. If the install is painful, you don't need all of it for this node, but you do need the pack registered cleanly.