Skip to content

Creating images

The image studio turns a written prompt into finished images. You pick a brand, describe what you want, and generate. By default the Auto model picker chooses the best model for the job, so you can start without knowing anything about models.

This page walks the whole flow. For deeper topics, see Choosing a model, Writing better prompts, and Generation settings.

The brand selector sits at the top of the composer. The brand you pick decides which brand the images belong to and which brand kit is applied.

Every brand has a kit — a prompt prefix and suffix, banned terms, default styles, and anchor references. When a brand with a kit is active, a panel shows what it adds, marked applies on submit. The kit is applied automatically. To generate this batch without it, turn on the Override toggle in that panel.

Every image goes into a brand. Owners and admins on All brands pick one first: Generate reads Pick a brand to generate until they do. New images start private to you until you share them.

Type what you want in the prompt box. Be specific about subject, setting, mood, and composition. See Writing better prompts for what works.

Two helpers sit at the bottom of the prompt box:

  • Enhance rewrites your prompt into a richer, more detailed version (keyboard shortcut shown in the tooltip). After it runs, an Undo chip appears for a few seconds so you can revert.
  • Dictate — the microphone — lets you speak your prompt instead of typing. Tap the mic, talk, and the text is inserted for you. It biases toward your active brand and model names so jargon comes back spelled correctly.

A character counter shows in the corner.

By default the model is set to Auto, which reads “AI picks what’s best” and resolves to the right model when you generate. You don’t have to touch it.

To choose a model yourself, click the model row to open the browser. Models are listed one row per version, each with capability chips:

ChipMeaning
ReferencesAccepts reference images as input
FastQuicker, lower-cost renders
Imported filesTakes imported files from the brand’s Training tab

Filter by provider, by capability, or by Best for (the kind of result a model is suited to), or search by name. Pick a model to pin it; the row then shows its name and its cost per image, in credits. To return to Auto, click the Auto link above the model row, or pick the Auto row in the browser.

For a fuller breakdown of what each model does, see Choosing a model.

Reference images guide the look of a result — match a style, a subject, or a vibe. Click Add reference to open the picker. You can:

  • Upload a file or drop one into the grid
  • Browse History — your recent generations
  • Browse Gallery — images across your team
  • Browse Uploads — files you’ve added

Search within any source, select one or more, and click Done. Selected references show as a removable thumbnail strip above the button. Most models accept up to four references. To learn more, see References.

To change a specific part of an existing image rather than guide a new one, use Editing images instead.

  • Aspect ratio — choose from the grid below the prompt (square, wide, tall, and any others the model supports).
  • Images — use the stepper to make 1 to 4 images per run. The estimated cost updates as you change the count.

Open Advanced settings below the prompt for more control. What appears depends on the active model and can include style, negative prompt, resolution, seed, guidance, steps, and output format. Everything here is optional — defaults are sensible. See Generation settings for what each control does.

If your studio admin has enabled mature content, you’ll see an optional toggle.

Click Generate (or Generate 4 images when the count is higher) to start. Results stream into the feed as they finish. Click any result to open it full size.

Next to Generate, the footer shows what the run costs, about how long it takes, and your studio’s balance, for example About 4 credits · ~25s · You have 212. If the run costs more than you have, that line turns amber, the note under it reads Not enough credits for this run, and Generate waits until you lower the count, pick a cheaper model, or add credits. Generate also waits until you write a prompt; hover or Tab to it to see why.

From the open result you can:

  • Download the image
  • Regenerate with the same settings
  • Run quick tools — Upscale, Remove BG, Vectorize, or Edit
  • Tag it with a brand, or file it into a project
  • Set it as a reference for your next prompt, to build on the result

From any feed tile you can also download it or reuse it as a reference without opening it.

If a result misses, give it a thumbs-down from the tile or from the result’s panel in the Library and pick what went wrong: wrong product, bad quality, ignored the prompt, unsafe or off-brand, or something else in your own words. A thumbs-up marks the ones that worked. Your rating stays on the result, and it helps the team see where generation falls short.

Your newest result also has a row of next steps underneath it: Upscale, Remove background, Extend, Animate, Reuse prompt, and Use as reference. Upscale, Remove background, and Extend open that tool with the image already loaded. Animate opens the video studio with the image as the first frame. Reuse prompt puts the image’s prompt back in the prompt box. Use as reference adds it to your references. A step that doesn’t fit the result is left out: an image that’s already been cut out has no Remove background, and Animate only shows when you can make video.

Some results need more than a single step — stacking an imported file on a base model, guiding with several references at once, or running a multi-step template. These appear as added options and workflows right in the composer when the model you’ve chosen supports them. To save a model, prompt, and settings you want to reuse, see Templates; for imported files, see Imported files.