Image generation from a text prompt

Describe the frame — the AI draws it in any style and in the size your layout needs. A photograph, an illustration, a 3D render or a retro poster: one model instead of stock sites and photo shoots.

Create an image from a description

Portrait of an old fisherman in a yellow raincoat against the morning sea, close-up, soft diffused light, photorealism, detailed skin

One description — a finished picture

Everything below was generated from prompts like the ones printed under the images — ordinary sentences, no special syntax.

Photorealism · 1024 × 1280
Portrait of an old fisherman in a yellow raincoat against the morning sea, close-up, soft diffused light, photorealism, detailed skin

Cyberpunk · 1536 × 864
A cyberpunk megacity street at night, neon signs glowing as abstract shapes with no readable lettering, rain and reflections on the asphalt, a silhouette under an umbrella, cinematic shot

Isometric · 1024 × 1024
Isometric 3D illustration of a cosy coffee shop in cross-section, pastel colors, soft shadows, neat interior details

Food photography · 1280 × 1024
A plate of pasta with tomatoes and basil on a dark wooden table, top view, steam over the dish, side light from a window, food photography

Poster · 1024 × 1408
Retro travel poster of Lake Baikal in a mid-century style, flat shapes, limited palette, a large BAIKAL caption at the bottom

Interior · 1536 × 1024
A Scandinavian living room interior: a light sofa, oak floor, a big window, plenty of daylight, minimalism, architectural photography

Product shot · 1024 × 1280
A frosted glass perfume bottle on a sand-colored pedestal, stony texture, soft shadow, warm beige background, an ad product shot

Watercolor · 1280 × 1024
Watercolor illustration of a mountain lake at dawn, soft washes, white paper along the edges, calm cool palette

Anime · 1024 × 1280
Anime illustration of a girl with headphones on a crosswalk in an evening city, warm lights, cinematic perspective

Macro · 1408 × 1024
Macro shot of dew drops on a spiderweb, morning backlight, blurred green background, maximum detail

Fashion · 1024 × 1408
Fashion shoot: a model in a burgundy coat against an ochre paper backdrop, hard directional light, a long shadow, magazine aesthetics

Aerial · 1536 × 864
Aerial shot of a rocky coastline with turquoise water and a white boat, straight top-down view, sunny day

Children's book · 1280 × 1024
Children's book illustration: a red fox cub with a lantern in a night forest, soft texture, warm firefly lights

3D render · 1024 × 1024
3D render of an abstract composition of glossy and matte shapes, soft studio light, peach-lavender palette

Pixel art · 1280 × 1024
Pixel-art scene of a small seaside town at sunset, 16-bit palette, crisp pixels, nostalgic mood

Documentary · 1024 × 1280
Black-and-white documentary photograph: an elderly couple on a park bench, grainy film, a natural moment

Visualization · 1536 × 1024
Architectural visualization of a wood-and-concrete country house on a slope, panoramic glazing, evening light from inside

Advertising · 1024 × 1280
An unbranded soda can in water splashes and ice cubes, bright orange background, frozen motion, an ad shot

Double exposure · 1024 × 1280
Double exposure: the silhouette of a person's profile filled with a foggy conifer forest, monochrome with a cold tint

Greeting card · 1024 × 1408
Holiday illustration: a snow-covered cabin in the forest, warm light in the window, falling snow, gouache texture, deep blue night

Flat lay · 1280 × 1024
Flat lay of a designer's desk: a notebook, pencils, a mug of coffee and a plant on a light tabletop, top view, soft shadows

Generate your own image

Why it replaces stock

A picture made for your layout, not whatever turned up in someone else's library.

Any style by name

Photorealism, watercolor, isometric, pixel art, retro poster, 3D — the style is set with words in the same request. No separate model per genre.

The size your layout needs

Portrait for a story, landscape for a banner, square for a post. The proportions are set before the generation, so the frame is composed for them instead of being cropped later.

Holds text and fine detail

A caption on a poster, labels on packaging, neat typography inside the picture — the model draws letters, not an imitation of them.

Straight onto the canvas

The image appears in your project: crop it, drop it into a template, send it to prompt-based editing or cut out the background — without leaving the tab.

How to generate an image

It all happens in the browser: no installs, no queue, no model settings.

1
Sign up for SUPA and create a project of the size you need
2
Open the AI section and pick image generation
3
Describe the frame: what's in it, where it happens, in what style and light
4
Choose the proportions and run the generation. Keep the option you like on the canvas, regenerate the rest

How to write a prompt that gets the right picture

Start with the main subject

The first words weigh more than the rest. First what's in the frame, then the surroundings, then the style and the light.

Bad: “beautiful, stylish, expensive, coffee shop”
Good: “isometric 3D illustration of a cosy coffee shop in cross-section, pastel colors, soft shadows”

Describe the light — it does half the work

“Soft diffused light”, “hard directional light and a long shadow”, “morning backlight” change the frame more than three more adjectives about the mood.

Name the genre of the shot or the drawing

The words “food photography”, “architectural visualization”, “documentary shot”, “children's book illustration” set everything at once: the angle, the lens, the palette and the level of detail.

Specify the angle and the framing

Close-up, top view, strict side view, shot from below, aerial — one such phrase removes half of the failed attempts.

Put on-image text in quotes

If the frame needs a caption, give it exactly and say where it goes: “a large BAIKAL caption at the bottom”. The shorter the line, the cleaner the result.

Ban the unwanted explicitly

“No people”, “no lettering”, “no logos”, “plain background” — direct bans work better than hoping the model won't add them on its own.

Change the wording, not the number of attempts

If the result is consistently off, add the missing dimension: the material, the time of day, the lens, the palette. Repeating the same request gives variations of the same idea.