From prompt to finished image

Pick the model and the mode

The image workspace lists every model the catalogue marks as available and callable for this site, so the list matches what the service will actually accept. A model is offered only in the modes it declares.

If a mode is missing from the workspace, that model does not declare it — it is not hidden by the interface.

Write the prompt

Describe the picture you want in the order you would describe it to a colleague: subject, setting, framing, light, then any style instruction. Naming a camera or a film stock narrows the result more than a long list of adjectives.

The prompt stays in the field after a run, so the fastest way to improve a result is to change one clause and generate again with the same settings.

1. State the subject first.
2. Add the framing and the light.
3. Keep style instructions to one or two.
4. Change one clause per attempt and compare.

A ceramic mug on a matte stone counter, three-quarter view, soft window light from the left, shallow depth of field, muted colour palette.

A courier in a worn canvas jacket standing in a rainy alley, chest-up framing, cool evening light, cinematic colour grading.

Both prompts above are written for this guide; they are not quoted from any other source.

Read the quote before you submit

The credit cost is quoted by the server for the settings you have chosen, and the ledger records what was actually charged when the job finishes. This is why the price on a landing page and the price of a task can differ.

1. Set the aspect ratio, quality and reference images first — the quote follows the settings.
2. Read the quoted credits next to the generate button.
3. Generate, then check the ledger entry if the balance looks wrong.

• Changing the settings after the quote refreshes the amount.
• A failed job still appears in the ledger, so read the entry before assuming credits vanished.

Review the result

Generated images are a starting point rather than a finished asset. Check the hands, the text in the image and the edges of a reference subject first, because those are the places a model most often gets wrong.

If the result is close, keep the seed and the settings and change one thing. If it is far off, changing the mode is usually faster than adding more adjectives.

• Open the image at full size before you use it.
• Check any text in the image character by character.
• Keep the prompt with the result so an edit is repeatable.