Logos And Vectors

Make true scalable vector art (logos, icons, app glyphs, scalable illustrations) or convert a raster image into clean editable SVG. Use when the user says "make a logo", "I need an SVG", "vectorize this PNG", "turn this into a vector", "a flat icon set", "a scalable mark", or wants artwork that stays crisp at any size and opens in a design tool. The output is real SVG, not a flat PNG that looks vector-ish. For a raster poster, ad, or packaging with exact copy on it, use text-in-image instead.

Produce genuine SVG vector art: paths, fills, and color regions that scale to any size and open editable in Illustrator, Figma, or the browser. Two jobs live here: generate a vector from a prompt, or vectorize an existing raster image into clean SVG. This is distinct from a normal image model, which only ever returns a flat raster (PNG/JPG) that cannot be rescaled or edited as paths.

Inputs to collect

• The job: generate a new vector from a description, or convert an image you already have.
• For generation: what the mark/icon is, the style (flat, line, geometric, badge), and any exact color set to lock.
• For conversion: the source PNG/JPG (a clean, high-contrast, flat-color image vectorizes far better than a busy photo).
• For a brand mark with words in it: the exact text string. Vector models do not reliably set type, so pair with text-in-image for the legible wordmark (see Technique).
• Optional: background transparent vs a fill color.

Models

All three carry the op:vectorize capability and use taskType: vectorize. Confirm each is live and inspect its schema via the runware-models + runware-run skills before calling. Never hardcode a stale choice.

• Generate a vector: Recraft V4 Pro Vector (recraft:v4-pro@vector), io:text-to-image + op:vectorize. Takes a positivePrompt and emits SVG directly. Best pick for new logos, icons, and scalable illustration with controllable line quality and color regions.
• Raster to SVG: Recraft Vectorize (recraft:1@1), io:image-to-image + op:vectorize. Converts a PNG/JPG to clean SVG with sharp edges and minimal artifacts. Default for converting existing flat artwork, logos, and icons.
• Raster to SVG alternative: Picsart Image Vectorizer (picsart:1@1), edge-aware tracing that separates flat color regions into compact, resolution-independent SVG. Try it as a second opinion when Recraft Vectorize over- or under-simplifies a source.

Workflow

1. Resolve the model schema (runware-run) and confirm the fields below. All three use taskType: vectorize, not imageInference.
2. Generate: call recraft:v4-pro@vector with positivePrompt. Optionally lock color via providerSettings.recraft.colors / backgroundColor, and set width/height from the model's fixed 2K size set (both required together if either is sent).
3. Convert: upload the source under inputs.image (required) and call recraft:1@1 or picsart:1@1.
4. Run synchronously. Vectorize is a fast image task with no async polling. outputFormat is SVG (the only value, and the default).
5. Read the result: the SVG comes back as imageURL (plus imageUUID). Open it to confirm it is real vector paths, not an SVG wrapper around an embedded bitmap.

Technique

• A vector is not a "vector-looking raster." A regular image model can draw a flat, logo-style PNG, but it is still pixels: blurry when scaled, with no editable paths. These models output actual SVG geometry. If the user needs to resize, recolor, or hand off to a designer, you must use a vectorize model, not a styled raster.
• Generate flat, then it vectorizes clean. Prompt for simple, bold, flat-color shapes: "flat geometric logo", "minimal line icon", "two-color emblem". High contrast and few colors trace into tight paths. Photographic, gradient-heavy, or noisy descriptions produce messy, bloated SVGs.
• Lock the palette. On Recraft V4 Pro Vector, constrain color with providerSettings.recraft.colors (and backgroundColor) instead of naming hex codes in prose. It holds the brand palette far more reliably. This is the Recraft lever called out in runware-prompting.
• Words belong to a typography model, not the vectorizer. Vector models render shapes well and lettering poorly. For a wordmark or a logo with a tagline, generate the legible text with the text-in-image skill and compose it with the vector mark, rather than trusting the vectorizer to spell.
• Clean source in, clean SVG out. When converting, a flat high-contrast image (a screenshot of a flat logo, a scanned line drawing) vectorizes beautifully. A detailed photo produces thousands of stray paths. If a conversion is muddy, simplify or flatten the raster first, or generate fresh with Recraft V4 Pro Vector instead.
• Two tracers, two opinions. Recraft Vectorize and Picsart Image Vectorizer simplify differently. If one loses a detail or over-quantizes the colors, run the other on the same source and compare.

Parameters that matter

• taskType: vectorize for all three (not imageInference). Pass it explicitly.
• inputs.image: required source for the raster-to-SVG models (recraft:1@1, picsart:1@1).
• positivePrompt: required for recraft:v4-pro@vector (1 to 10000 chars).
• providerSettings.recraft.colors / backgroundColor: palette and background control on Recraft V4 Pro Vector. Prefer this over hex-in-prose.
• width / height: optional on Recraft V4 Pro Vector, but only from the model's fixed 2K aspect-ratio set, and both must be sent together.
• outputFormat: SVG only (and the default). The result arrives as imageURL.
• Confirm every field against the live schema (runware-run) before calling. Never guess names.

Quality bar

• The output is a real .svg of editable paths, not a raster embedded in an SVG tag. Open and scale it to verify it stays crisp.
• Color regions are clean and the palette matches what was requested (no muddy gradients where flat fills were asked for).
• A converted SVG keeps the source's recognizable shape with a sane path count, not thousands of stray slivers from a noisy input.
• Any text in the mark is correct and legible. If it is misspelled or fuzzy, redo the lettering via text-in-image and compose, do not ship the vectorizer's guess.

Related skills

runware-run, runware-models, runware-prompting; text-in-image (exact legible wordmarks to pair with a vector mark), product-photography (raster brand/product imagery, the non-vector sibling).