About Sigilry
Why this exists, who it is for, and what it refuses to do. The workbench is one click from here.
Why this exists
Three options, and what each one actually hands you.
A designer
The best logo, for money and days — and the files that come back are usually not ones you can keep editing.
An image model
Fast and cheap. What comes out is a picture of a logo: it cannot be recoloured, and it cannot be scaled for print.
A tracer
An SVG that follows your pixels, drawn path by path. That is the right answer when the file is already what you want — and it is one of the three free tools here.
The same picture, three ways:
A picturepawta.png 
A tracepawta-stencil.svg 
A redrawpawta-rethought.svg
Two products share the word logo generator
One is asked to make a picture of a logo, the other to make a logo you can still work in. The two goals genuinely conflict, and a tool that tries to do both does neither. This is the second one, which is the harder half and the one almost nobody builds.
What is different
Three claims, each one checkable on the file the landing page shows you.
The file arrives finished
One editable SVG is the whole deliverable. Nothing is attached to it: no report, no score, no watermark, and no second artifact to keep in step with it.
It is a redrawing, not a copy
The logo you chose is re-thought in paths. What you get is a sibling of the picture you picked rather than a copy of it — the largest risk this product carries, and the reason you are asked to choose twice.
One operation, and an honest one
The outermost parts of a delivered logo can be moved. Nothing can redraw, recolour, add or delete a shape: a logo that was drawn wrong is fixed by generating again, not by the editor.
Who this is for
One segment, and one explicit refusal.
A solo founder, or a very small team
Someone shipping a product, a site or an app this week who needs a usable logo and is going to put it in code. Not a designer, and not going to open Illustrator to fix one.
Not an agency or a brand team
They have designers already, and what they need is a different product: palettes, review workflows, asset management.
That this is the right customer is a hypothesis. It has not been tested against real users, and it is written down so the work can be pointed at a specific person rather than at a market.
What we refuse to build
Each of these is plausible, and each one would take the pressure off generation quality.
A drawing editor: A delivered logo’s parts can be moved; they cannot be redrawn, recoloured, added or deleted. A move corrects placement, and it cannot rescue a shape that was drawn wrong — an editor that could would take the pressure off the step that makes the logo.
Brand palettes and design systems: A different customer, with designers and review workflows that are not what is being built here.
A gallery of what the product made, or a remix community: Distribution should be earned rather than manufactured, and a wall of the product’s own output is not evidence about the one you would get.
Our own model: The claim is the quality and the editability of what is delivered, not model training.
A score on the file: What is delivered is the file, and no number is attached to it. A measurement this runtime cannot take is worse than no measurement, and an unverifiable badge on every file is a disclaimer rather than evidence.
Running it is the shortest way to read this page
A brief in, six candidates out, and one editable file at the end of it.
Open the workbench