Quiver Image to SVG

Most image nodes output pixels. Quiver Image to SVG outputs actual SVG - a vector file you can open in Illustrator, scale to a billboard, and edit path-by-path. It takes a raster image, sends it to Quiver AI's vectorization API, and hands you back structured vector graphics. Partner node, ships with ComfyUI core, no install, cloud-hosted, billed per call against your Comfy account.

Why does vector output matter? Because an upscale makes pixels bigger; a vector makes the image resolution-independent. Logos, icons, decals, anything you want to re-use at any size without quality loss - that's the job this node exists for. In a ComfyUI graph it's the odd one out on purpose: the output type is SVG, not IMAGE, and it's meant to terminate at the built-in Save SVG node, which writes the .svg file with your workflow metadata embedded.

How it works

Your image is uploaded to Quiver's svgs/vectorizations endpoint through Comfy's proxy, and the returned SVG comes back as an SVG object you can save directly. The model dropdown is dynamic and offers Quiver's Arrow model family - arrow-1.1, arrow-1.1-max, and arrow-preview - with sampling knobs (temperature, top_p, and a target_size for the square resize the vectorizer works from) appearing once you pick one.

The inputs that matter

• image - the raster you're vectorizing. Feed it something clean: flat colors and sharp edges vectorize far better than photographic gradients.
• model - the Arrow variant. The -max model costs more; the base Arrow is the sensible default.
• auto_crop - off by default. Flip it on to automatically crop to the dominant subject before vectorizing - handy when your image has dead space around the logo.
• seed - the usual partner-node disclaimer: it controls whether the node re-runs, not the output.

Output is SVG, which you wire into Save SVG (the built-in consumer for this type).

Where people get burned

• Expecting a photo to vectorize well. Vectorization is for logos, icons, illustrations, typography. Throw a photorealistic portrait at it and you'll get back a weirdly simplified mess - that's the nature of the task, not a bug.
• The SVG output needs the right sink. A normal Save Image node won't accept it. Use Save SVG (core, category "image") to write the file.
• Auto-crop is binary. It crops to the dominant subject and it's on or off - no threshold. If it crops too aggressively, turn it off and crop manually first.
• Cost is per model. The badge changes with the Arrow variant you pick; -max and -preview are pricier than the base model.