Vector or nothing: why SVG is the only identity deliverable that matters

Every brand identity will eventually be asked to live somewhere unexpected. A 16px favicon. A 30-meter trade show banner. An embroidered patch. A laser-engraved metal plate. A vector animation. An e-paper newsletter signature.

A raster logo (PNG, JPG, WebP) has one job: look correct at the resolution it was exported at. Take it anywhere else and you're either losing fidelity (favicon) or losing your shirt re-creating it from scratch (banner).

A vector logo (SVG, PDF, EPS) is the original mathematical definition of the mark. It scales infinitely, edits losslessly, and exports to any raster size on demand. There is no trade-off.

So why are so many AI logo tools still shipping raster?

The AI raster problem

Most image-generation models (Stable Diffusion, Flux, Midjourney) produce pixels. They're trained on raster datasets and they output raster. The "logo" they hand you is a 1024×1024 PNG that looks fine in chat but breaks the moment you try to use it as a real brand asset.

Three workarounds, in order of how often they fail:

1. Manual trace. Take the PNG into Illustrator and rebuild it by hand. The result is technically vector but loses the AI's specific curves. Cost: hours per logo.
2. Auto-trace. Run the PNG through a tool like Vector Magic. Output is "vector" but the bezier curves are noisy, anchor points are doubled, and color counts are wrong. Looks fine until you zoom in.
3. Re-generate and pray. Ask the model again with "vector style". You get another raster that looks like a vector.

The native-vector path

The only clean solution is to use a model that outputs vector natively. We integrate Recraft V4 (vector mode) and Ideogram V3 specifically for this — they produce true SVG with editable paths, not rasterized facsimiles.

When you generate a logomark in our chat, the response includes an "Open in Studio" button. That handoff opens the actual SVG in the V2 editor where you can:

• Edit anchor points directly with the pen tool
• Run boolean operations (union, subtract, intersect)
• Snap to a construction grid
• Export at any size — 16px favicon, 4K splash screen, 4-meter print — all from the same source

That's the workflow. Vector in, vector edited, raster out only when needed.

"But the AI image is so good"

Yeah — for inspiration, mood, exploration. Photoreal renders, marketing visuals, illustration-style banners — those should be raster, because they are raster by nature. There's no "infinitely scalable" version of a photograph.

But a logo is not a photograph. It's a logical construct made of shapes, lines, and curves. Pretending otherwise is a category error that costs you the moment you need to ship anywhere outside the original canvas size.

What to ask for

If you're using any AI tool to generate brand identity, ask:

• Does this produce native SVG or a rasterized PNG-that-looks-vector?
• Can I edit the anchor points after generation?
• Does it preserve the same paths if I export at different sizes?
• Is the color count clean (5 colors, not 47)?

If the answer to any of these is "no" or "kinda", you'll spend more time fixing the output than you saved generating it.

The exports we ship

When the swarm finalises a brand, the Export agent produces:

• SVG primary mark — the source of truth.
• PNG fallbacks — 16, 32, 64, 128, 256, 512, 1024 — for systems that can't render SVG.
• PDF — for print production.
• Favicon set — ICO, multiple PNG sizes, Apple touch icon.
• Social formats — square, landscape, story.

All of these are derived from the SVG. If you ever need a size we didn't ship, you can re-export from the original mark in the Studio. The SVG is the brand. Everything else is just a rendering of it.