Image Studio — deep engine host (E4b)
ImageThe shipped Image Studio now hosts the shared media engine inside Generate — compose the AI prompt, refine it with history, then generate. Accent #7c3aed marks the mounted engine. Light app theme · clickable. No caching — refresh for latest.
What changed in E4b. Generate now mounts the shared media engine: preview the AI-composed generation prompt + its prompt guide, chat-refine it with full turn history (revert any turn), and generate straight from the composed/refined prompt. The bare prompt textarea is demoted to a seed in Brief. The standalone model picker is replaced by the engine's ProviderModelPicker (Higgsfield · google-flow). The studio's shadcn chrome stays the same — the engine is a distinct surface hosted inside it.
Pages — Summer Splash · Hero Image example
01
Brief & direction
Read the brief from the planner, shape the visual direction — and the generation prompt now becomes a seed the engine composes from downstream, not the thing you generate from directly.
Open
02
Generate (deep)
The mounted engine end-to-end: compose the prompt + guide → refine-with-history → generate from the composed/refined prompt. ProviderModelPicker, editable composed prompt, refinement turns with revert.
Flagship
Open
03
Engine states
Every engine condition side by side: composing, fallback provider, resolved scope, not-connected, and the google-flow manual hand-off path.
Open
04
Spec round-trip
How the prompt travels: pre-fill from the spec → refine in the engine → write-back. Clarifies Draft-with-AI vs Compose/Refine — two distinct AI surfaces.
Open
The seam. These pages show studio shadcn chrome (cards, stepper, segmented controls) hosting the gua-* engine panel — one shared media engine, deep-hosted by the Image Studio. Start at Generate (deep) for the full flow.