The old per-clip “Reference ingredients” block is re-skinned as an actionable ResourceChecklist. Each row is a GenerationResource node for this clip — a video role · status · a per-row mode toggle (⚡ on-platform vs ✦ prompt / DIY) · its actions. The role set is driven by the clip's frames-vs-ingredients method. On-platform generation is Higgsfield-only and lives in the engine surface embedded inside this pink studio card — a deliberate, contained seam.
This re-skins the existing scene.ingredients[]. Every checklist row maps to one ingredient on this clip's scene — the candidates / takes pipeline below it is unchanged. The video roles offered differ by method: frames-to-video needs a start frame (+ optional end frame); ingredients-to-video needs typed image / video / audio ingredients. Flip any row between ⚡ on-platform and ✦ DIY independently.
1 Clip 1 · Full reel
Frames Ingredients
Frames-to-video — the model animates between a start frame and an optional end frame. The checklist asks for exactly those frame roles.
on-platform generation · resolves your Higgsfield connection
Generating
On-platform Prompt / DIY
Generate this start frame on-platform, then “Use this” to fill the node. One asset for one node — flat, no nested checklist.
Media provider
higgsfield
On-platform per-resource generation is Higgsfield-only — google-flow stays in the manual / DIY path.
resolveMediaProviderKey
Higgsfield model
Nano Banana Pro · still frame
Resource prompt — start frame
Prompt guide
1.Describe the first frame as a single still — the model animates motion out of it, so frame the subject and composition precisely.
2.Attach reference chips (product / persona / logo) so the frame is on-brand; the result fills this start-frame node directly.
Prompt composed by your AI provider; the frame asset is delivered by your Higgsfield connector.
Refine the prompt
1warmer morning light
2tighter on the fabric
Candidates (3)
“Use this” on the selected candidate fills the start frame node and flips it to Have.
End frame end_image· optional
leave empty to let the model resolve the closing frame
Needed
On-platform Prompt / DIY
End frame is optional for frames-to-video — provide one to lock the closing composition, or skip it. Generate opens the same Higgsfield generator inline for this node only.
Compose & refine use the free text path (your AI key / platform key) — they never spend a media credit. Generate spends a media credit and resolves your Higgsfield connection via resolveMediaProviderKey — never the platform key. Filled frames are written straight back to scene.ingredients[].
2 Clip 2 · Market walk
Frames Ingredients
Ingredients-to-video — the model composes from typed ingredients. The checklist asks for image and audio roles for this clip.
Generate opens the inline Higgsfield generator for this image node only — credentials resolve via the Higgsfield connector and this spends a media credit.
Voiceover audio
DIY — generate with your BYOK voice (ElevenLabs), then upload back
Needed
On-platform Prompt / DIY
Composed VO script — copy & run in your voice tool
1.Copy the script into your BYOK voice — ElevenLabs (your key, your cloned brand voice). Audio is not generated on-platform.
2.Render the VO there, then Upload the clip back — or Pick one already in your library. It fills this audio node and is synced as the clip's audio ingredient.
DIY mode has no Generate button — nothing runs on the platform, so no media credit is spent here. Composing / refining the script above is free text. Voice stays BYOK-only (ElevenLabs), outside the Higgsfield resolveMediaProviderKey boundary.
Persona — Aanya identity image
via Pick — persona-aanya.png · matched in for character continuity
Have
Filled nodes write straight to scene.ingredients[] — the candidates / takes pipeline below is untouched.
3 Clip 3 · Packshot CTA
5s
A clip's composed prompt GeneratePanel and its checklist live in the engine surface too — but stay disabled until a clip model is picked. The pin is per-clip.
Shared media engine · Composed prompt NEW · per-clip
Gated. No clip model is picked yet for Clip 3, so the composed VIDEO prompt + Higgsfield picker + refine row below are disabled. Pick a model in the clip's Model & settings to arm it — the composed / refined prompt then pins to a per-clip snapshot so Generate-clip runs exactly that text.
Media provider
higgsfield
awaiting clip model
Higgsfield model
— pick a model —
Composed VIDEO prompt — Clip 3
Coexistence. Everything above is new on top of the existing Stage-4 chrome. Below each clip, the shots editor, Generate-clip row and takes grid are unchanged — the resource checklist only feeds scene.ingredients[]; the pinned composed prompt is what Generate clip sends to Seedance / your media connector.