Recraft V4 Text to Vector

Every image node in this family outputs pixels. Recraft V4 Text to Vector is the exception that will make you double-take: it outputs an SVG. Real vector graphics, generated directly from a text prompt by Recraft's V4 vector models - not a rasterized image wearing a vector costume. For logos, icons, stickers, and anything that needs to scale to a billboard without artifacts, that's a genuinely different capability than any diffusion model gives you. Recraft is basically the only major lab that does native vector output, and this is the ComfyUI door to it.

Partner-node mechanics as usual: your prompt goes to Comfy's proxy at api.comfy.org, Recraft's servers run the model, credits come off your Comfy account, nothing runs locally. But the output plumbing is different, and it matters. The node returns an SVG type, not IMAGE. That means you can't just wire it into Save Image and call it done - you need whatever the vector-saving node is in your workflow, or you save the SVG data out. Plan the tail of your graph before you click, or you'll have a perfectly generated SVG with nowhere to go.

Inputs: prompt (up to 10,000 characters - plenty), negative_prompt (and here's the honest quirk: the tooltip in the source says flat-out it's ignored - Recraft's V4 vector models don't support negative prompting, the input is kept for interface consistency), model (a DynamicCombo with six options - recraftv4_1_vector, recraftv4_1_utility_vector, recraftv4_1_pro_vector, recraftv4_1_utility_pro_vector, plus the older recraftv4 and recraftv4_pro), n (1–6, how many results), and seed (which the tooltip admits is just a rerun trigger - results are nondeterministic regardless). The model combo also carries a nested size selector; note the "pro" tiers default to 2048×2048 while the standard ones default to 1024×1024.

There's also an optional recraft_controls input - the same RECRAFT_CONTROLS type used across the Recraft family, letting you feed in control signals from the Recraft Controls node for finer generation.

Which model? The "utility" variants are Recraft's workhorse vector tier - the ones you'd pick for clean, production-usable graphics - while the plain vector models skew more illustrative. The pro tiers cost a lot more per call and buy you larger output and more polish; start standard.

The grounded judgment: this is a niche node, but the niche is real. If you produce logos, game assets, or any vector work, "prompt → editable SVG" is a workflow nothing else in ComfyUI can do, and it's worth the credits for a handful of concepts. Just remember the two gotchas: the negative prompt does nothing, and the output is not pixels. Wire a proper SVG sink in, and this becomes one of the most unusual tools in your graph.