Text to Image AI
Describe it and get a picture back. Several models, one composer, and a library that keeps everything — so the good result is still there tomorrow when you need a second version of it.
- Multiple image models, one prompt
- Reusable characters and references
- Any aspect ratio you publish in
- Straight from image into video
How to generate an image from text
Three steps, and the interesting decisions are all in the second one. Most of what separates a usable image from a near-miss is model choice and shape, not prompt length.
- 01
Describe the picture, not the idea
Name the subject, the framing, the lighting, and the style. Adjectives about mood do less work than concrete nouns about what is in the frame and where the light comes from.
- 02
Pick a model and an aspect ratio
Models differ sharply in what they are good at — photographic realism, illustration, text inside the image, faithful product shapes. The form lists the ratios each one supports, so choose the shape you will actually publish in before generating, not after.
- 03
Generate, compare, refine
Runs go to your library, so you can fire several phrasings and judge them side by side. Editing the winner in image-to-image is usually faster than re-prompting from scratch.
What the image studio adds over a bare prompt box
Model switching without rewrites
The prompt, references, and settings carry when you change model, so testing the same idea across two models is a click rather than a rebuild.
Characters that stay the same person
Save a character once from your own photos and reference it in every prompt. Describing a face in words produces a different face on every run, however precise the description.
Reference images, not just words
Attach a reference for style, composition, or a product's real shape. It is the difference between asking for 'our packaging' and showing it.
Publish-shaped output
Vertical, square, 4:5, and wide are all first-class. You are not cropping a square down to a story frame afterwards and losing the composition.
Finishing in place
Upscale, relight, or edit a generated image without exporting it to another tool and re-uploading it somewhere else.
One step from motion
Any image in your library can become the first frame of a video, which is the shortest route from a still concept to a moving one.
What people generate from text
Concept art and mood
Look development for a campaign or a scene, generated in a dozen directions before anything is commissioned.
Social creative
Posts and story frames in the exact ratio each platform wants, without a design round trip.
Backgrounds and plates
Environments to composite a product or a person into, generated to the lighting you already have.
Thumbnails
Attention-shaped stills for video, tested in variants rather than designed once and hoped for.
Illustration for writing
Article and deck imagery that matches a house style, generated instead of licensed.
First frames for video
A still generated deliberately so the video that grows out of it starts on the right composition.
Prompting for images is a different craft to prompting for video
An image model has one frame to satisfy, so it rewards detail that a video model cannot use. Lens language, light direction, surface texture, and colour palette all land — where a video prompt loaded with the same detail tends to produce a clip that keeps changing its mind. If you have come to images from video prompting, the useful adjustment is to be more specific about the frame and less specific about the story.
Negatives are usually a symptom rather than a fix. When a result keeps including something you do not want, the cause is more often an ambiguous positive description than a missing exclusion — 'a plain concrete wall' does more than a long list of things the background should not contain. Reach for the negative field after the positive prompt has stopped improving, not before.
Model choice does more than any prompt rewrite. Photographic realism, clean vector-like illustration, legible text inside the image, and faithful product geometry are four different strengths, and no single model leads on all of them. Running one prompt across two models costs a few credits and answers the question directly, which is faster than another round of adjectives.
Finally, save what works. A prompt that produced a good result is a reusable asset, and so is the character or reference behind it. The people who get consistent output are not writing better prompts each morning; they are re-running a small set of prompts they have already proved, with the subject swapped.
Text to image FAQ
- Is text to image free?
- Yes, within your free credits. Every account starts with credits that work across image, video, and audio generation, and no card is required. Paid plans add more credits, higher resolutions, and watermark-free exports.
- What resolution do I get?
- It depends on the model, and the upscaler takes any result higher afterwards. For print or large display, generate at the model's best size and then upscale rather than asking for an unusual size up front.
- Can I generate the same character twice?
- Yes. Save a character from reference photos and mention it in each prompt. That is the reliable route to a consistent face across a set of images.
- Can models write text inside the image?
- Some can, most only approximately. For anything where the wording has to be exact — a logo, a price, a headline — generate the image without the text and add it in a layout tool.
- Can I use the images commercially?
- Yes, under the terms of service, on a paid plan. You remain responsible for the prompt — in particular for not generating a real person's likeness without their agreement.
Where to go next
Write one line and see what comes back
Free credits on sign-up, every image model in the catalog, and a library that keeps the results you liked.