01
Text-to-Image
Describe what you want in ordinary language. Composition, lighting and detail are the model’s job, not yours.
Home / Product
ProductSix things you stop
doing by hand.
Write a sentence, choose a size, get a file you can ship. Everything below is one request away.
Capabilities
What the platform does
02
Style Lock
Save a look once and apply it to everything after. Sets hold together across seeds, subjects and aspect ratios.
03
Inpainting & extend
Change one region, or push a composition past its edges, without regenerating what already worked.
04
Upscaling
Any generation to full delivery resolution with detail intact — print, packaging, full-bleed hero.
05
Batches & variations
One prompt across many seeds and ratios in a single call. Keep the frames that work, bin the rest.
06
API access
Three endpoints, documented, with the settings on every response.
The controls
Enough dials to be useful. Few enough to stay usable
Six parameters cover everything. There is no node graph, and there is nothing you have to learn before your first good result.
Full reference| Control | Range | Changes |
|---|---|---|
| prompt | text | The image, described plainly. |
| size | ≤ 2048 px | Output dimensions and ratio. |
| steps | 4 – 50 | Denoising passes. More detail, more time. |
| guidance | 1 – 12 | Prompt adherence versus model judgement. |
| seed | integer | Fixes randomness. Same seed, same picture. |
| style | preset id | Applies a locked look to the generation. |
In practice
What the three editing tools
What the three editing tools
actually do
Every one of them exists because regenerating a whole image to fix one corner of it is the wrong answer.
Style Lock
Save a look once, then apply it by id. The subject changes; the treatment does not.
Inpainting
Mask a region and regenerate only that. Everything outside the mask is left byte-for-byte alone.
Upscaling
Take a draft to full size, resolving detail rather than interpolating pixels.
Questions
Before you ask
Can I use the output commercially?
That is the intent, and the terms are written to allow it. Read the terms before you rely on it for a client project.
Do you train on what I generate?
No. Prompts and outputs run the service and help us debug it; they are not training data. If that ever changes it will be opt-in.
What happens to a prompt you will not run?
It is refused before generation and logged. The categories are listed in the acceptable use policy.
How close is Style Lock to a real brand guide?
It holds palette, lighting and treatment across a set. It will not reproduce a specific logo or typeface — that is layout work, not generation.
Put it against your worst brief
The prompts that break are the ones we want. Bring one and see.