⭐ Storyboard MCP · Flagship · Face-swap
One Face · Six LivesOne identity photo · six cinematic scenes · the fastest face-personalized flagship in the catalog
A user uploads ONE selfie. Storyboard generates six 10-second cinematic scenes with a stand-in actress, runs an identity quality gate on the first scene, then fans out face-swap-video across all six in parallel — replacing the actress with the user's face in every shot. ~6 minutes wall time. ~$1.50 end-to-end. The cheapest path from "I want to see myself in a film" to a finished cinema reel anywhere in the AI-video category.
The proof · same face across six sceneskrea-2-large hero · the showcase reel ships in PR B
Real-use scenarios — who needs this, what it unlocks
Face-swap personalization is not a vanity feature. It's the missing layer between "make a generic ad" and "make THIS customer's ad." These are the personas who ask for this every week.
DTC brand marketer
"I want to send every customer the same product reel — but with their face. Can we do it at scale?"
Today: $1.50 per customer · 6 minutes · zero re-shoot. Previously: impossible at any reasonable cost.
Indie filmmaker
"I'm pitching an actor for the lead. Can you put their face in 6 cinematic scenes so I can show what they'd look like in the role?"
Today: one headshot from the actor → six scenes in six minutes. Previously: hire a pre-vis studio.
Family historian
"Can you put my grandmother's face into a video that shows her on the beach she used to talk about?"
Today: a memorial reel with disclosure watermark, made privately. Previously: not possible.
Brand-influencer agency
"We have one influencer signed for six brand contexts this quarter. Can you generate the kit and we'll face-swap to her face?"
Today: one influencer kit, six scenes, customer can swap THEIR face onto any of them. Previously: 30 days of agency back-and-forth.
Vtuber / streamer (post-launch, with the live extension)
"My vtuber model is good but I want my streams to feel like cinematic shorts during downtime — with MY face on a movie protagonist."
Today: Six pre-rendered reels with your face. After PR C (Scope graph node for live face-swap): live LV2V face-swap on the stream itself, sub-second latency, your face on any character in real time. The post-credit demo for OPUS 1 / MAYA / ATELIER.
The orchestration — two caps working together (plus a quality gate)
This showcase is the minimum-orchestration flagship — only two caps in the main loop. seedance-i2v generates the cinematic context, face-swap-video replaces the stand-in's face with the user's. The "wow" isn't capability count; it's the one-photo input + the six-scene fanout in 6 minutes for $1.50.
| Stage | Capability | What it does | Cost |
|---|---|---|---|
| Stage 1 · Generic scenes | seedance-i2v × 6 | 6 parallel renders of cinematic 10s clips with a stand-in actress in your 6 scene contexts (skate park · foggy bridge · neon alley · sunset · midnight Chinatown · dawn beach) | ~$2.16 |
| Stage 2 · Identity gate | face-swap-image | One test swap on scene 1's first frame. Show user pre/post side-by-side. If poor → pick new selfie BEFORE the 5 video swaps fire (saves $1.20) | ~$0.009 |
| Stage 3 · Fanout face-swap | face-swap-video × 6 | 6 parallel face-swap passes. Each swaps the stand-in's face with the user's identity photo while preserving the scene's pose, lighting, body, clothing | ~$1.44 |
| Stage 4 · Finishing | ffmpeg-concat (+ optional music + narration) | Stitch 6 swapped clips into one 60s cinema reel. Optional minimax-music + gemini-tts | ~$0.30–0.60 |
Pipeline trace — honest end-to-end with the identity quality gate
Stage 1 — GENERIC SCENES (parallel)
6 ×
Stage 2 — IDENTITY QUALITY GATE (your attention, ~30s) 1 ×
Stage 3 — FANOUT FACE-SWAP (parallel) 6 ×
Stage 4 — FINISHING
SHIP → viewer URL with pre-swap / swapped toggle per scene
create_media action=animate, model=seedance-i2v, duration=10s
Each prompt: [scene setting + camera + action] with generic actress as stand-in
P1 skate park golden hour · P2 foggy Embarcadero bridge · P3 Mission neon alley
P4 Golden Gate sunset wide · P5 midnight Chinatown rain · P6 dawn beach soft pink
Returns → 6 × 10s clips in ~90s parallel
Stage 2 — IDENTITY QUALITY GATE (your attention, ~30s) 1 ×
create_media action=generate, model=face-swap-image
source = scene-1 first-frame · face_ref = user's selfie
Show pre-swap (actress) vs post-swap (you) side by side
If poor (occluded face, wrong lighting, "stuck on" look) → pick new selfie
Cost of catching here: $0.009. Cost of not catching here: $1.20 in re-fires.
Stage 3 — FANOUT FACE-SWAP (parallel) 6 ×
create_media action=generate, model=face-swap-video
source_url = each scene's 10s clip · face_ref_url = user's selfie
Each swap preserves pose, body, lighting, scene — replaces only the face
Returns → 6 × swapped clips in ~120s parallel
Stage 4 — FINISHING
ffmpeg-concat 6 swapped clips → 1920×1080 cinema
Optional minimax-music (palette-matched bed at 0.55 vol)
Optional gemini-tts (3-stanza poetic narration ABOUT the user)
Returns → Final 60s reel ~60s wall
SHIP → viewer URL with pre-swap / swapped toggle per scene
Cost — honest end-to-end
Stage 1 scenes
~$2.16
6 × seedance-i2v 10s @ $0.36
Stage 2 gate
~$0.009
1 × face-swap-image (saves $1.20 if it catches a bad selfie)
Stage 3 swaps
~$1.44
6 × face-swap-video @ $0.24 (10s × $0.024/s)
Stage 4 finishing
~$0.30–0.60
optional music + narration; ffmpeg-concat free
Baseline cost: ~$3.85 with all bells. With fast siblings (seedance-i2v-fast): ~$1.50–2.00. Cheapest face-personalized cinema in the catalog.
Why this is the flagship for face-swap — and not just a cap registration
What this showcase proves
- One-photo onboarding is the wedge. No LoRA training, no multi-image upload, no 10-minute wait. Single selfie → six cinematic scenes. Lowest friction point for personalized video in the AI-video category.
- The identity quality gate is the trust gate. Stage 2 (one $0.009 image-swap test) catches bad selfies BEFORE $1.20 in video swaps fire. The cheapest insurance policy in the catalog.
- The toggle viewer is the proof. Side-by-side pre-swap (actress) vs swapped (you) per scene. No magic, no marketing — the buyer sees exactly what the cap does.
- Per-customer marginal cost of $1.50. Means brands can offer "personalized cinematic reel" as a $5–10 product. Storyboard becomes the engine behind a customer-facing offering.
- The pattern generalizes. Substitute the 6 scene briefs and you have: a tourist promo (your face in 6 city scenes), an actor reel (one headshot in 6 genre tones), a memorial (a loved one in 6 scenes from their life), a brand-influencer kit. Same chassis.
- The live extension is queued. When the Scope
face_swapgraph node lands (PR-C running inswapper-512-live ≥30 FPS @ 512×512), this same identity photo will drive both the VOD reel AND a real-time LV2V stream. The post-credit demo for the OPUS 1 / MAYA / ATELIER flagships.
Open invariants — what the playbook enforces
Quality gates (auto-applied)
- Cost upfront. Before any inference: estimated total shown, user approves.
- Identity quality gate AFTER scene 1, BEFORE the other 5 swaps fire. The single most cost-effective control point in the pipeline.
- Per-scene face-detection check. Half-moon returns a flag if no face detected; that scene gets a warning chip on the canvas card; original (pre-swap) URL is kept as backup.
- Auto-chunk long clips. If any scene extends >10s, face-swap-video auto-chunks to 5s segments + ffmpeg-concat. Single-call timeout safety.
- Consent gate on first run. Storyboard asks the user to confirm rights to the identity photo. Default: yes for own face; explicit confirmation for any other identity.
- AI-altered watermark via pillow-watermark on public-publish outputs (default-on; user can opt out + acknowledge responsibility).
Cross-surface parity — webapp · MCP · CLI · slash
| Surface | Entry point | Status |
|---|---|---|
| MCP | generate_project({ playbook: "one-face-six-lives", identity_photo_url, ... }) | Phase 2 — registered playbook |
| Webapp chat | "swap my face onto card-1" → agent routes via MCP | Live (via natural language) |
| Webapp right-click | image card → "Six Lives reel with this face" → modal picks 6 scene briefs OR uses defaults | PR B — context menu entry |
| Webapp slash | /six-lives <identity-card-ref> | PR B — new entry in lib/skills/commands.ts |
| CLI | livepeer playbook one-face-six-lives --identity <url-or-card> --piece-slug <slug> | PR B |
Reproduce — the playbook is open
Pipeline lives in the one-face-six-lives playbook · capability spec is in the face-swap skill. The two new caps (face-swap-image at $0.009/gen + face-swap-video at $0.024/s) replace the deprecated easel backend. The bare face-swap alias still works for backward compatibility. Live LV2V face-swap is queued for the Scope graph-node sprint.