Composed slide prompt — per slide

Carousel

A carousel = an ordered set of single-image slides — flat images, no per-resource checklist (unlike video). Each slide is now the deep host of the shared media engine: a Higgsfield-only ProviderModelPicker, an editable composed image prompt + guide, and refine-with-history. The pink SlideCard is the studio chrome; the violet panel inside is the engine — a deliberate, contained seam. The composed / refined prompt pins to a per-slide snapshot so Generate below runs exactly it.

Saved

Slides (4) 4:5

Hook

Image selected

Benefit

Composing · this slide

Each slide owns its own provider/model + composed prompt + refine history. Switching slides remounts the panel on the right.

This is the per-slide GeneratePanel in detail. Below is Slide 2 · Benefit's engine section — model picker → editable composed IMAGE prompt + guide → refine row with revertable history. It composes on the free text path; the candidate grid, Upload & Deliver-to-ad stay unchanged below (coexistence) and spend the media credit on the pinned snapshot.

Slide 2 · Benefit 4:5

Mid-carousel card · close-up hero benefit shot · one flat image · pins to this slide's prompt snapshot

Image Video · soon

Shared media engine · Slide 2

Media provider

higgsfield

Per-slide composed generation is Higgsfield-only — google-flow stays in the manual / DIY path. This picker is the slide's own override — it replaces the old single whole-carousel model Select.

composed on platform key

Higgsfield image model — Slide 2's own pick

Nano Banana Pro

Picked for this slide → other slides can carry a different model. The whole-carousel Select is gone; each slide overrides per-slide.

Composed slide prompt — Slide 2 (image)

Prompt guide — image

1.Describe this slide as one flat still — scene, light, palette and the exact product — as if briefing a photographer for a single card.
2.Reference the real assets (product / logo) so the model places them, not look-alikes. A slide is a flat image — there's no per-resource checklist here, unlike video.
3.Keep negative space for the card headline — the headline & link live on the ad, not in the image. Hold the carousel's 4:5 frame.

Prompt composed by your AI provider; the slide image is delivered by your Higgsfield connector. Composing & refining never spends a media credit.

Refine the prompt

Refine uses the free text path (basePrompt-chained) — not a media credit; Generate spends one. History is per slide — the panel remounts when you switch slides, so each slide keeps its own composed prompt + refinements.

1brighter palette

2add the product centered

3more negative space for the headline

Pinned to Slide 2's snapshot. The composed & refined prompt above is the exact text Generate runs — edits here re-pin the snapshot; the candidate grid below renders from it.

The panel remounts per slide — Slide 2's provider/model, composed prompt and these three refinements are kept apart from Slide 1's and Slide 3's. Open another slide in the rail and it brings its own composed prompt + history.

Unchanged Candidate grid, Upload & Deliver-to-ad are exactly the carousel Studio controls — they run the prompt pinned to this slide's snapshot.

Valid for Nano Banana Pro · 4:5 · 3 variations · prompt pinned

Runs Slide 2's pinned snapshot on your Higgsfield connector — the only step that spends a media credit.

Candidates (3)

Generated

Upload your own image

Skip generation — bring an existing 4:5 image for this slide. Fills the slide directly.

Card headline & link? Those live on the ad — set them in Ad · card copy after delivery. This studio produces the slide image only.

Deliver to ad — Summer Sale → Ad set “Warm — Sales”

2 of 4 slides ready · each delivered image becomes one card. Atomic deliver once every slide has a selected image.

Refine uses the free text path (basePrompt-chained) — not a media credit. Generate runs Slide 2's pinned composed prompt on your Higgsfield connector and spends the media credit; resolveMediaProviderKey resolves your connector.

Back to slide work

Compose & refine per slide, then generate against the pinned snapshot. Next slide → Proof