Text to image, without writing a prompt like a programmer
Type what you want in plain language. The prompt is rewritten into the structure image models actually respond to, then rendered in five to ten seconds.
Try one of these
Text to image is the simplest way to use a generative model: you describe a scene, the model returns a picture. The hard part has never been the model. It is the gap between how people describe things and how image models read them.
This page explains what actually changes the output — prompt order, lighting words, aspect ratio, and when a reference photo beats any amount of description.
What the model hears when you type
Image models do not parse a sentence the way a person does. They weight tokens, and the tokens near the front of the prompt tend to dominate. A prompt that starts with "a beautiful image of" spends its strongest position on a word that describes nothing. Starting with the subject — "a red fox curled in snow" — puts the weight where it matters.
The second thing models respond to is medium and light. "Photoreal" and "oil painting" pull the output in completely different directions, and words like "golden hour", "overcast" or "hard flash" change the shadows more than any adjective about mood. Specific beats poetic: "rim light on wet asphalt" lands, "moody vibes" does not.
This generator rewrites your input into that structure before it reaches the image model: subject first, then style, then light, then lens or medium, with a short negative list appended. You can see the rewritten prompt in the response, which makes the tool useful as a learning aid as well as a renderer.
- Lead with the subject, not with praise for the image
- Name the light source before naming the mood
- Keep one idea per clause — models blend competing ideas into mush
- Use the style chips for look, and the text box for content
Choosing an aspect ratio before you generate
Aspect ratio is not a crop applied afterwards. It changes what the model composes. A square frame tends to centre the subject and fill the edges; a 9:16 frame leaves headroom for text overlays and pushes the subject lower; a 16:9 frame wants a horizon.
If the image is destined for a phone screen, generate at 9:16 rather than generating square and cropping — you will get a composition that was designed for the frame instead of one that was squeezed into it. Portrait 3:4 is the safest default for people, because it matches how portrait photography has been framed for a century.
When words are not enough
Text alone cannot reliably reproduce a specific person. You can describe silver hair, green eyes and freckles, and the model will produce someone who matches the description — but not someone who matches the photograph. If identity matters, switch to image to image and supply a reference.
The same applies to products, logos and anything with exact typography. Describe the scene in words, then attach the subject as a reference image so the model has something concrete to preserve.
Which mode to use for a given job
| You have | Use | Why |
|---|---|---|
| Only an idea | Text to image | No reference to preserve; fastest path to a result |
| A photo you want restyled | Image to image | Keeps composition and identity, changes treatment |
| Several photos of one person | Character | Identity is enforced across the set of references |
| Exact text or a logo | Image to image | Letterforms survive far better when supplied visually |
3 example prompts are preloaded above — tap one to try it.
Questions, answered
About fidelity, credits and output sizes on this tool.
Other ways to generate
Ready to make something that looks like the person you had in mind?
Start generating