Superdesign alternatives: remix or specify

Superdesign began as an open source design agent that ran inside your editor, billed as the first of its kind, working in Cursor, Windsurf, Claude Code and plain VS Code. That extension is still on GitHub and is now described there as no longer actively maintained. The product moved to the web.

What runs at superdesign.dev today is a canvas. You describe a screen, pick a model, and get a design back. Around the canvas sit tools for recreating a screenshot, importing from a site, generating images and video, and applying a design system. Under it sits a public gallery labelled Discover design prompts, sorted into Style, Animation, UI Component, Landing Page and Presentations, where individual entries carry remix and view counts in the thousands.

The verdict up front: if you want a canvas and a starting point, the alternatives below are all canvases with starting points. If you want the decisions written as values you can paste anywhere, that is a different product, and it is Thanor.

What the gallery actually hands you#

A remix. That is the unit, and it is worth being precise about what a remix is.

You open an entry, you like it, you press the button, and now you have a copy you can change. The parts you change are yours. The parts you do not change are the author's, and they are most of the page.

In practice that means:

• You swap the colours, because colours are the easiest thing to see.
• You change the copy, because the copy is obviously not yours.
• You leave the type scale, because you would have to know what it was to change it.
• You leave the motion timing, because it is not visible as a number anywhere.
• You leave the section order, because it already reads fine.

Three of those five are the substance. A visitor cannot tell you what a type ratio is, and will absolutely tell you that one page feels expensive and another does not, and the difference is largely in the three you left alone.

That is not a flaw in remixing. It is what remixing is for: fast, cheap, close enough. The trouble starts when the same starting point has been remixed several thousand times, which the counters on the gallery report openly. An entry sitting at several thousand remixes and a couple of hundred thousand views is a good entry, and it is also a look that a fair number of people in your market have already seen twice this month.

Every gallery in this category has that property. It is the price of a public starting point, and it is not a reason to avoid one. It is a reason to be clear about which parts of the result you actually authored.

The alternatives, side by side#

Tool names on the left. Read the right column downward.

Six products, one column that never empties.

This is the shape of the whole category, and it is deliberate rather than careless. A tool with a strong house style is a tool nobody can point at a client brand, so the sane product decision is to execute any look faithfully and hold no opinion about which look belongs here. Every row above does exactly that, and does it well.

The consequence is that somebody upstream has to hold the opinion. When nobody does, the tool supplies a median: a centred hero, three feature cards, a soft fade upward on scroll, a weightless sans at one of four sizes.

That median is why so many generated pages rhyme, and it does not care which of the six produced them.

Why a named style is not art direction#

The gallery names its entries by look. Glassmorphism Style. Red Noir Style. Kinetic Orange Style. That is a better naming convention than most of the market manages, because it at least points at an aesthetic rather than at an industry.

It is still a name rather than a specification. A name is a pointer, and two people pointing at the same word land in different places.

Take glassmorphism. Everyone agrees on the idea: translucent panels, blur behind, a thin light border. Nobody agrees on:

1. The blur radius. 12px and 40px are different products.
2. The panel opacity, and whether it changes with depth.
3. The border, and whether it is one value or a gradient along an edge.
4. What is behind the glass, which is the actual decision, and which the style name says nothing about.

A specification answers all four with values. A style name answers none of them, and the remix answers them by inheritance: whatever the original author chose, still in place, in every copy that did not open that panel.

The same applies to motion. Animation is a category on that gallery, and motion is the place where a number decides everything: cubic-bezier(0.22, 1, 0.36, 1) at 620ms reads as considered, and the identical move at 200ms reads as a website from 2014.

What a design system feature assumes#

Superdesign's canvas includes an option to use a design system, and so does most of the market now. It is worth reading that feature the same way everywhere.

Lovable's version is documented precisely: it reads tokens and components from a source you provide, then runs adherence enforcement during ongoing work, catching raw colours, customised components and other deviations. Framer's version reviews a site before publishing and flags contrast problems, typos, missing alt text, SEO gaps and inconsistent styles.

Deviations. Inconsistent. Both comparative. Both measure the page against a standard that already exists somewhere, and neither writes the standard.

• Bring a palette, a scale and a motion vocabulary, and enforcement is worth every penny, for years.
• Bring none of them and the feature passes a page that is perfectly consistent and entirely anonymous.

Consistency is not direction. A template is consistent, which is exactly why four studios in the same market can ship the same one without noticing.

So the useful question is never which canvas. It is what you hand the canvas, and whether that thing is a name, a remix, or a set of values.

There is a structural reason none of these companies sells the values. Across roughly 110,000 indexed URLs belonging to Framer, Webflow, Lovable, Replit, Cursor and Windsurf combined, the string easing appears in three. cubic-bezier appears in zero. A tool cannot publish the specification, because a specification exact enough to paste works in every competing product and turns the choice of canvas into the small decision.

Which alternative to choose#

Three honest answers, depending on what you were missing.

1. You wanted components. Take the free registries. They are good, they cost nothing, and a component and a decision solve different halves of the job.

1. You wanted a whole page generated. Any of v0, Lovable or Bolt will do it, and the choice between them is about the meter and the runtime rather than the output. Superdesign's own canvas covers this too, with a model picker and a code export.

1. You wanted the design decided. This is the one no canvas answers, and it is why the remix counters climb into the thousands: people keep arriving for direction and leaving with an artefact. The artefact is useful. It is simply not the same object as a decision, and the difference only becomes obvious on the third project, when all three have the same rhythm.

The verdict: use whichever canvas you like, and change what you bring to it. A written specification carrying a palette in hex, a type scale at a stated ratio resolving to a stated size, motion as a named curve at a real duration, and a section order with a reason for each position, turns a remix into a build. It is also portable, so it works in Superdesign, in v0, in Lovable, and in whatever appears next year.

The test is simple enough to run this afternoon. Generate the same page twice on the same canvas, once from a style name and once from a written set of values, and put the two results beside each other. Nothing about the software changed between them.

Browse the Thanor library and read one full specification before your next remix.