How to search website design by image, and find every other way to show it

Sometimes you know what you have, not what you want. A section that works but looks like everyone’s. A frame in Figma that’s nearly right. Searching for more of the same won’t help. What helps is seeing the other ways it could go.

regards takes a picture and a question about it: what looks like this, what does the same job better, and what’s the opposite. What comes back is real sections from real sites, to study and borrow from. It won’t design the section for you; that part stays yours, in Figma, in code or on paper. Here is how to ask, in the app and from your coding agent, and six real searches with what came back.

How do you search website design by image?

In search, click the lens at the end of the search bar, Visual search, and pick a picture. Or paste one into the bar, or drop one anywhere on the page. A screenshot of one section works best: crop to the part you mean.

Then say what you want from it, or nothing at all. The words steer the search, and the answer comes back in a group named for what you asked:

Instant and Deep both take a picture. Deep reads the picture and your words together and looks harder, and a Deep search with a picture uses one of your Deep searches, as any Deep search does. Every signed-in account can search with a picture, Free included, at 10 searches a day. Your pictures are kept for 30 days, and no one else’s searches can use them.

What else could a row of cards be?

To find out, we searched with a section of our own: the row of style cards on our landing page. A heading, a line, two arrows, and cards you step through.

With no words, it finds what looks like it:

Motion

Nearly the same section: a heading and a line on black, then a row of captioned previews that runs off both edges.

LogoArchive Shop

The ruled version: magazine covers in hairline cells on black, and the next shelf’s title with a View all link under them.

Red Antler

The same carousel in daylight: a result as each card’s headline, and a red progress line and round arrows under the row.

Ask what else it could be, and it finds the same job done other ways:

Motion

The cards in three dimensions: a helix of photographs, with tabs that turn it into a barrel, a ring or a sphere.

IDEA

No carousel at all: every category at once, a six-column grid of square photographs, each named under it.

App Stacks

A directory instead: categories down the left as coloured pills, and the cards in a grid beside them.

Name the control you’re thinking of, and it answers for that. Here, the tabs:

Raycast

A paragraph as the tabs: each short sentence is one, the active one lit and the rest dimmed, the app window beside it changing.

Rivian

A segmented pill over a carousel: Technology, Performance and Design, each a video card with its neighbours cut off at the edges.

Red Antler

Filter tabs over a mosaic: one tab lit red, then uneven photo and video tiles, each with a name in its corner.

And the opposite, for when you’ve been staring at one idea too long:

Specht Studio

Everything ours isn’t: posters and covers packed edge to edge in loud colour, with no captions, arrows or rules.

DJ Meisner

Warm and loose: drawings, prints and etched metal on a peach ground, each with a small caption like a museum label.

Adam Ho

One project, full bleed: a yellow and magenta identity, a lanyard and a type specimen, side by side in one strip.

What are all the ways to show the same information?

Most sections start as the obvious shape. Pricing is the clearest case: three cards, the middle one marked. So we made one up, for a coffee roaster called Tidewell that doesn’t exist, and asked the same question.

Baked Design

Rows, not cards: each plan one line with its price on the right, and the one most people pick tinted, not raised.

Visitors

No tiers at all: one card and a slider for how much you need. Bags a month would work the same way.

Daybridge

Two plans, then a table that answers what’s in each, line by line.

Norma

A table instead of tiers: one black column for yours against the other ways to get the same thing, the price its last row.

Four answers to the same brief, none of them three cards. Then take the one you like and search again with it. Here, Baked’s rows, “like this but dark”:

Polar

Plans side by side in one framed panel on near-black, the type as quiet as Baked’s, with the fee under each price.

Plasticity

Square columns with small monospace headers, and the top plan in lime from its header to its footnote.

That one came back looser. Pricing set as rows is rare in the library, and dark rows rarer, so the search kept the dark and the quiet type and let go of the rows. When what you want is rare, a picture finds its neighbours, not its twins. And it reads one section best: from a full-page screenshot it takes the top, so crop first.

How do you search by image from Claude Code, Cursor or Codex?

Over MCP. Your agent asks the regards server for a link, sends the picture to it with curl (a PNG, JPEG or WebP of up to 4 MB, within 15 minutes), and searches with it, words and all, the way you would in the app. So it works in agents that can run a command: Claude Code, Cursor, Codex and the like. Your agent does the looking, in the middle of a task; the choosing and the designing stay with you. Give it a screenshot and ask for references:

Upload this screenshot to regards and find every other way sites show this kind of list. Show me the four most different, with a line on what each does well.

Here’s our pricing section. Search regards with it for better versions of the same thing, then for its opposite. Show me three of each with a line on what each does differently; I’ll take it from there in Figma.

Add the server by its address, as the agents page shows, or through our listing in xAI’s marketplace, and agents search with pictures on every plan, Free included. The Claude and ChatGPT directory connectors don’t offer it, because their chats can’t send a file that way. For setting up each agent, see website inspiration for Claude Code, Cursor and Codex.

How is this different from other image search?

Searching by image isn’t new. Mobbin, Refero, Savee and Cosmos all take a picture, and Behance does on Pro, as our comparisons found in October 2026. Image search mostly finds what looks like your picture, and often that’s what you want.

regards answers more than “like this”. Ask for alternatives and you get real sections that do the same job better; ask for the opposite and you get what yours isn’t. Each answer is a section of a real website, from 3,060+ sections of 330+ sites picked by hand for how they look. Better inspiration, not more of it.

What do you do with what comes back?

Treat it as references, not answers. Pick the two or three that matter and write down what each does well: rows instead of cards, tabs that read as a sentence, a warm ground. Then go deeper by hand, in Figma, in code or in a sketch: the type, the spacing, the rhythm, the small details no search returns. That’s where the section becomes yours. And when you’re stuck, photograph the sketch and search with that.

For how we compare in more detail, see regards vs Mobbin and regards vs Refero. Or try it with a picture of your own.