Recraft V3 Text to Vector
Real SVG
- recraft_controls
- SVG
Diffusion models output pixels. That's not a limitation you can fix locally - no matter what checkpoint or LoRA you throw at it, a raster generator gives you a raster. Recraft's one genuinely unique trick is that it can produce actual vector graphics: real SVG output, resolution-independent, editable in Illustrator. This node is that trick, exposed to ComfyUI.
The story checks out historically. Recraft's V3 model climbed to the top of the leaderboard under the "red_panda" alias, and the thing that separated it wasn't photorealism - it was that it was the only image generator producing vector output natively, aimed at designers rather than hobbyists. This node is the design-tool end of that product, and there is no free local equivalent for what it does.
Inputs that matter
- prompt - what you want as a vector
- substyle - a dropdown of the vector-illustration family:
line_art,bold_stroke,thin,editorial,segmented_colors,vivid_shapes, and more, plus "None" - size - canvas size from Recraft's list,
1024x1024default - n - 1 to 6
- seed - re-run control only, nondeterministic results
- optional negative_prompt, optional recraft_controls
The output is SVG - not an IMAGE. That trips people up, because the rest of your graph is built around image tensors.
Wiring the SVG output
The SVG type has its own core nodes. The straightforward one is SaveSVGNode (Save SVG), which writes your vector to disk under output/svg/ - that's the node you want for actually keeping the file. If you need to preview or further rasterize, there are SVG conversion utilities in core. Don't try to plug the SVG wire into an IMAGE input; the types are distinct and the graph will complain.
How it works and what it costs
Request goes to /proxy/recraft/image_generation with model recraftv3 and the vector-illustration style baked in; the SVG comes back as a download and is packaged into the core SVG type. Billed per image from your Comfy account credits. Ships with ComfyUI core; needs a Comfy account, no local model files.
Common issues
- It's a vector look, not a vector guarantee. The output is a real, editable SVG - but complex subjects with gradients and heavy shading vectorize poorly. Flat, clean, icon-like prompts give the cleanest paths. Ask for a busy photoreal scene and you'll get a stylized simplification, because that's what vectorizing means.
- Check the file in an editor. The SVG will open in Illustrator, Inkscape, Figma, or a browser. If a client needs production paths, budget time to clean up what comes out.
- Different from vectorize. This node generates a vector from a prompt. If you have an existing raster image you want converted to SVG, that's the separate
RecraftVectorizeImageNode.
If your pipeline touches logos, icons, or any design deliverable that needs to survive scaling, this is the node that justifies the whole partner-node system. There's simply nothing else in core that produces editable vector output from a text prompt.
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | Prompt for the image generation. | |
| substyle | COMBO | 25 options: None, bold_stroke, chemistry, colored_stencil, contour_pop_art, cosmics, +19 | |
| size | COMBO | 1024x1024 | The size of the generated image. |
| n | INT | 11–6 | The number of images to generate. |
| seed | INT | 00–18446744073709550000 | Seed to determine if node should re-run; actual results are nondeterministic regardless of seed. |
| negative_promptopt | STRING | An optional text description of undesired elements on an image. | |
| recraft_controlsopt | RECRAFT_CONTROLS | Optional additional controls over the generation via the Recraft Controls node. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| SVG | SVG | — |