Chat to generate (the prompt composer)

Chat to generate is the free-form half of Generate Image: a rich prompt editor where you build a scene by combining text with your model, predefined styles, and reusable custom elements. It's the most flexible way to make a brand-new image, as opposed to picking a ready-made Complete set.

Both live inside Generate Image (tabs on desktop, sub-tools of Generate image in the mobile Choose a tool sheet), but they work differently:

Chat to generate is for when you know what you want and want to describe it precisely; Complete set is for a fast, on-theme starting point.

The prompt editor accepts several kinds of content, each shown as a pill or plain text:

You mix these together in one prompt; APOB reads the whole thing to build the image.

With a portrait model selected, an empty prompt's muted example demonstrates that structure directly: "Describe your image with @influencer and styles. E.g., A polished editorial portrait set in a Cafe on a Rainy Day, with @influencer wearing a Hot summer clothing and relaxing beside a rain-streaked window, rendered in the Realistic with soft ambient light, natural skin texture, and rich reflections across the glass." Both @influencer references appear as model pills; Cafe, Rainy Day, Hot summer, and Realistic appear as modifier pills with grayscale catalog-image backgrounds. Placeholder pills use the same dimensions as pills entered in the editor.

In No Model mode, the example contains detailed character features instead of model pills: "Describe your image with detailed character features and styles. E.g., A poised young woman with warm olive skin, shoulder-length wavy auburn hair, green eyes, and light freckles sits beside a window in a Cafe, wearing a tailored charcoal Hot summer clothing on a Rainy Day, rendered in the Realistic art style with soft window light and reflections across the glass." Cafe, Hot summer, Rainy Day, and Realistic remain modifier pills. Both examples are placeholders only; they are not submitted.

Use the editor's model/mention control to insert your model. The pill shows the model's name and avatar. The muted model pill in the empty-state example is only a placeholder and is not submitted with your prompt; insert the real pill when you want the generated image to feature that model.

Elements are custom references (a person, an object, a look) that you slot into the prompt and reuse. The number of element slots you can actually fill depends on your selected Quality:

The editor shows every slot your quality allows, plus one locked slot when a higher tier unlocks more (so UltraS 4K, already at the 10-slot maximum, shows no locked slot); on desktop the pink Element header also shows your used count out of the allowance, e.g. "(2/8)". Hover the locked slot to see which qualities unlock it, e.g. "Element 5 only available for UltraS 4K / Ultra 4K." Clicking it on desktop switches automatically to the lowest quality that unlocks it. Mobile follows the same allowance-plus-one rule; tapping the locked card shows the hint as a toast, switches to the lowest unlocking quality, and reveals that tier's full allowance plus its next locked card. The mobile card popup scrolls vertically when the list exceeds its height. To learn how to create and manage elements, see Elements.

If you've filled more slots than a lower tier allows and then switch quality, APOB warns you first. A dialog titled "Some elements will be removed" explains, for example, that the new tier "supports up to N elements. Switching now will remove X extra elements and their mentions from the prompt." Choose Continue to switch and drop the overflow, or Cancel to keep your current quality and all elements.

Yes — there's a single large character limit for the prompt (the same across every quality). If your input goes over it, APOB trims to the limit and shows a one-time toast: "Your prompt exceeded the 20000 character limit and was truncated." You can keep typing; only the trimmed text is used when you generate.

Template is the pill button at the bottom-right of the prompt editor (when the character counter is visible, it sits just below it). It opens a picker of images made with Generate Image — public ones under Community, your own creations under Generation. The picker always opens on the Community tab (templates are inspiration-first, even when you have generations of your own); switch to Generation to pick from yours. Every card shows the image with a preview of the prompt that picking it would apply: the prompt text with its style and element pills, and your currently selected model in the model pills — a template is applied onto your model, so the preview shows your model even on a community image made with someone else's. Hover a card on desktop, or tap the card's info icon on mobile, to expand the preview over the whole card; a prompt longer than the card still ends trimmed with an ellipsis.

Clicking a card uses that image as a template, and this completely replaces what you currently have in the composer: the prompt (text plus its model, style, and element pills), the selected Quality, the Aspect ratio, the filled element slots, and the stored negative prompt are all overwritten with the template's stored generation settings — the negative prompt has no visible field in this workstation, but one carried over from a previous remix or template is replaced (or cleared) along with everything else. Your selected model and Quantity are kept. A "Content settings applied." toast confirms it and the picker closes — so save anything you want to keep before picking a template.

Some cards can't be applied: one that is still generating shows "Generating. Can't select right now."; a Complete set image has no free-form prompt to copy, so it shows "This content can't be reused." and the picker stays open for another pick. The same message appears for an old image whose prompt has nothing left to apply — for example one built only from styles that have since been removed from the catalog. If a template's details fail to load, a "Failed to load content detail…" error appears and nothing in your composer changes.

They're separate things. Chat to generate is the main composer for making a new image of your model. The Generate result image dialog is a smaller, focused popup used inside other flows (like creating a video's last frame from a reference) — it has just a description box, an optional extra image, and a single result, with no element slots or style picker. If you're building a scene from scratch, you want Chat to generate.

Those live in the bottom bar of the workstation, not inside the editor. The order is cost, then Quantity (only on the Ultra tiers), then Aspect ratio, then Quality, then Generate. See Quality, resolution & duration and the main Generate Image page for what each control does.

Generate needs the generation context loaded and some content in the prompt. An empty prompt, or a selected personal model that hasn't finished loading, keeps the button greyed out. No Model mode supplies the default generation model behind the scenes, so it only needs a non-empty prompt once that context is ready. The rate depends on your plan and quality: Fast is priced as one four-image batch, while the higher tiers are billed per image times your quantity.

The composer fills the panel and grows when you tap into it — the on-screen keyboard pushes the layout up and the editor resizes to stay usable. Adding a model, styles, or elements works the same way through the editor's controls. The Template button sits at the editor's bottom-right corner here too; its picker opens as a full-screen page where tapping a card's info icon expands the prompt preview, and tapping the card applies the template. Quality, aspect ratio, Cost, and quantity are in the mobile bottom bar. Fast's Cost uses the credits / 4 images batch unit; the higher qualities show credits / image. The hint banner's tutorial opens by tapping the "?" button.

Related: Generate Image · Complete set · Elements · Credits & billing