OpenAI CLIP ViT-L/14

This node answers one question: what is in this image? You feed it an IMAGE, it talks to OpenAI's CLIP ViT-L/14 running on Runware's cloud, and it hands you back a text string. If you've ever wanted to reverse a prompt out of a picture - interrogate it, caption a dataset, or grab a seed description to feed back into a generator - this is the least setup you'll ever do to get one.

There's a nice bit of history under the hood. CLIP ViT-L/14 is the exact text encoder Stable Diffusion 1.5 used to read your prompts, the one with the famous 77-token limit. This node just points it the other way: instead of turning words into conditioning vectors for a diffusion model, it runs the full CLIP model against an image and produces language. Same little encoder, different job.

What you actually set

The schema is refreshingly short:

• image (required, IMAGE): the picture you're asking about. Wire any image out of your graph - a Load Image, a just-generated result, a frame grab.
• prompt (optional, STRING, multiline): the author's tooltip calls it "instructions or questions to guide the image analysis." Leave it empty and you get a plain caption; ask "what color is the car?" and you get a targeted answer. This is the difference between a captioner and a vision-LLM-lite.

The output is a single text (STRING) - that's it. Wire it into a prompt input, a Save Text node, or a Runware model's prompt socket to build a caption-to-image loop.

How it fits the pack

Like every node in Runware/ComfyUI-Runware, this one has no local weights. Your IMAGE tensor is encoded to a PNG data URI, sent over REST with the model's ID, and CLIP runs on Runware's GPUs. The cost per call shows up on the node's title bar. It's the same pack-wide pattern: install once, add an API key, and every node in the catalog works.

Install and the one gotcha

Install the pack once via ComfyUI Manager (search Runware) or:

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

Restart, then give it a key - in ComfyUI Settings → Runware API key, or set RUNWARE_API_KEY, or run runware auth login once. No API key is the single most common failure, and the node says so plainly: "No Runware API key. Set RUNWARE_API_KEY or run runware auth login."

A couple of honest caveats. This is a paid API, so every caption costs a fraction of a cent - fine for occasional interrogation, not the tool for bulk-captioning 10,000 images. And CLIP's captions read like CLIP captions: they describe content, not aesthetics, and they're nowhere near what a modern vision LLM gives you for detail. For a quick "what is this, so I can prompt it again," it's perfect. For rich alt-text, keep your Qwen. That's not a knock - it's knowing which tool you grabbed.