Nodes/ComfyUI/Quiver Text to SVG
ComfyUI Node Runs on cloud

Quiver Text to SVG

Type a prompt, get editable vector graphics

By Comfy-Org·Created 4 years ago·Updated about 12 hours ago· 130,663
Quiver Text to SVG
  • reference_images
  • SVG
prompt
model
seed0
instructions

Quiver Text to SVG is the prompt-driven half of Quiver's pair: instead of vectorizing an existing image, you describe one and get back an actual SVG - real vector paths, not pixels pretending to be vectors. It's a ComfyUI core partner node, cloud-hosted through Comfy's proxy, billed per call against your Comfy account, and it pairs with the built-in Save SVG node, which is where the SVG output is meant to land.

This is the node to reach for when you want an icon, a logo mark, a sticker, or a clean flat illustration at a size that can't blur. Because the output is structured SVG rather than a raster, you can take it into an editor and keep working - recolor it, resize it to infinity, adjust individual paths. The model is Quiver's Arrow family (arrow-1.1, arrow-1.1-max, arrow-preview), a text-to-vector generator that's much better at flat, graphic, single-subject requests than at "photorealistic scene" ones.

How it works

Your prompt goes to Quiver's svgs/generations endpoint through the proxy, and the response comes back as an SVG object for Save SVG. The model dropdown is dynamic - pick an Arrow variant and sampling options (temperature, top_p, presence penalty) appear under it. Text generation that produces paths instead of characters, is the simplest honest description.

The inputs that matter

  • prompt - the text description of the desired SVG. Be concrete: "flat vector icon of a coffee cup, minimal line art, single color." Vague prompts give you mush, same as everywhere.
  • model - the Arrow variant. Base arrow-1.1 is the value pick; -max and -preview cost more.
  • instructions - optional extra style or formatting guidance, separate from the prompt. Use it for constraints like "no gradients" or "stroke-only." It's an advanced field but it's the one that rescues mediocre results.
  • reference_images - up to 4 reference images to steer the generation. Attach a rough sketch or a color palette you want echoed.
  • seed - the standard partner-node caveat: gates re-running, not the result.

Output is SVG, which goes into Save SVG to hit disk.

Where people get burned

  • Wrong input genre. This is a vector generator; it does flat graphics and simple subjects. Ask for photorealism and you'll get a simplified cartoon of it, because that's what SVG is.
  • The save node. SVG output doesn't fit a Save Image node - that's the single most common confusion with this family. Save SVG is the matching sink and it's also core.
  • Prompts without instructions. The instructions field exists because the prompt alone often isn't enough. If the output keeps having the wrong format, that's the knob.
  • Reference discipline. Up to 4 reference images - batch more than that and you'll error.
  • Cost per model. The badge tracks the Arrow variant; the fancier models are roughly a third pricier, so default to the base until you need the upgrade.
Categorypartner/image/Quiver

Inputs (5)

NameTypeDefaultDescription
promptSTRINGText description of the desired SVG output.
modelCOMBOModel to use for SVG generation.
seedINT00–2147483647Seed to determine if node should re-run; actual results are nondeterministic regardless of seed.
instructionsoptSTRINGAdditional style or formatting guidance.
reference_imagesoptCOMFY_AUTOGROW_V3Up to 4 reference images to guide the generation.

Outputs (1)

NameTypeDescription
SVGSVG