Somny · 2025–2026

A character that actually looks like you

Somny turns a handful of someone's photos into a character you can put in any scene you can describe: you, your family, even the dog. I named it, designed it, and built the brand and the marketing around it.

Founder · Brand and identity · Design system · Interface · Motion
19Marketing pages live on somny.com
12Visual styles, each with its own look
37Brand guideline pages, dark and light
Twenty-four portraits of the same handful of people and pets, each in a different visual style: anime, art nouveau, painterly, cubist, 3D render, fantasy armour, autumn photography, an 1880s cabinet card, a 1950s advertisement, renaissance oil, synthwave and retro-futurism.

Everyone else gave you back a stranger

Apps that promise to put you in a costume mostly hand back someone who only half looks like you. The face drifts. You notice straight away, because it is your own face.

Somny settles the person first. You build a character out of your own photos once, and every image after that is drawn from it, so the likeness holds across a hundred different scenes.

What I made by hand

I animated this by hand in After Effects. The portraits in it are the product’s own.

A single prompt line reading "@Daisy starring as the main character in a movie", with a row of small character portraits underneath and the dog's portrait ringed to show it is selected. Below sit four generated images of the same dog in different scenes.

Put the character in the sentence

Who the picture is of goes inside the sentence, as an @ mention. It could have been a field above the prompt, the way a settings form would do it, but what someone is composing is a single thought and splitting it across two controls makes them assemble it twice.

The row underneath swaps who the sentence is about, and the sentence survives the swap.

A decision I'd defend

Twelve named styles, not a slider. People choose a look, they don’t tune one.

Designed twelve looks people can name

Each style is a visual language rather than a filter: its own palette, lighting, framing and subject vocabulary, tuned so the same character still reads as the same person inside it. That is the harder half. A look is easy to build and easy to build badly, and the way it fails is that the face stops being yours. A slider would have offered a thousand positions and no opinion.

A portrait in the anime style, flat cel shading and saturated color.
A portrait in the art nouveau style, ornamental border and flowing organic line work.
A portrait in the brushed light style, soft painterly edges and diffused daylight.
A portrait in the conceptualist style, geometric abstraction and flat blocks of color.
A portrait in the dreampop 3D style, rounded forms and pastel volumetric light.
A portrait in the epic fantasy style, armour and dramatic landscape lighting.
A portrait in the everyday style, plain photography in ordinary daylight.
A portrait in the heritage style, a monochrome nineteenth-century studio photograph.
A portrait in the mid-century modern style, pastel palette and period styling.
A portrait in the renaissance style, oil paint handling and warm directional light.
A portrait in the retro thunder style, neon rim light against a night city.
A portrait in the velvet future style, 1960s retro-futurism and warm synthetic texture.
The Somny logo mark, two cyan strokes reading as an S and as two figures turned away from each other, above the ten steps of the primary color ramp and a short lavender secondary bar.

Named it, drew the mark, set the palette

I chose Somny off a long list because it carries sleep and dreaming without spelling either one out. I drew the mark as two strokes that read at once as an S and as two figures turned away from each other. I let the cyan do the identifying on its own and kept the rest of the palette quiet, because a brand sitting behind photographs cannot also compete with them.

color pallet

Built the system every screen runs on

I built Somny’s design system as a semantic layer: 112 color tokens across three themes, 159 text styles, 42 effect styles. Change a token and every screen moves with it, which is how one person kept a whole product looking like one thing. HeroUI gave me the component foundation. The palette, the type, the icon set and a third dark theme that never shipped are mine.

A live marketing page headed "Create Your AI Superhero Look", with a generated comic-book portrait beside it and a small photograph of the person it was built from.

Designed the pages that sell it

Nineteen marketing pages are live, and each makes the same argument at a different search: put yourself in this particular world. I designed the sections they are built from and wrote the front-end for them, so a new page is composed now rather than drawn from nothing. That is the only reason there are nineteen of them, and it is what the pipeline in the next case study was built on top of.

The styles page on somny.com, headed "Bold Styles. Bold Photos.", with an offset collage of portraits in six different styles.
The same marketing page at phone width, the headline stacked above the portrait.

The whole surface is mine

I named it, drew the mark, built the design system, designed the product and wrote the front-end for the pages that sell it. Nineteen of those pages are live.

One person can hold a whole product’s look together only if the system underneath is built to let them. The 112 tokens are that system.

What I did

I own design, brand and marketing: the name, the identity, the design system, the interface, the marketing pages and the film.