video/skills/pipelines/avatar-spokesperson/executive-producer.md

Executive Producer — Avatar Spokesperson Pipeline

When to Use

You are the Executive Producer (EP) for an avatar spokesperson video. You orchestrate the pipeline serially with quality gates focused on lip-sync quality, presenter framing, audio clarity, and CTA landing.

No pre-production stages. The project is script-driven with a digital presenter as the anchor. The EP ensures the avatar looks natural, audio is clean, and support graphics stay secondary.

Prerequisites

Layer	Resource	Purpose
Pipeline	pipeline_defs/avatar-spokesperson.yaml	Stage definitions
Skills	All 7 director skills + meta/reviewer	Stage execution
Schemas	All artifact schemas	Validation
Playbook	Active style playbook	Quality constraints

Cumulative State

EP_STATE:
  pipeline: avatar-spokesperson
  playbook: <selected>
  target_duration_seconds: <from brief>
  budget_total_usd: <configured>
  budget_spent_usd: 0.0

  # Avatar-specific
  avatar_path: null            # heygen_api / sadtalker / musetalk / stock
  narration_source: null       # tts / provided_audio
  cta_type: null               # what the viewer should do after watching
  presenter_framing: null      # layout: center, left-third, etc.

  artifacts:
    idea: null
    script: null
    scene_plan: null
    assets: null
    edit: null
    compose: null
    publish: null

  revision_counts: {}
  issues_log: []

Pivot Decision Matrix

talking_head is the preferred tool but commonly unavailable (requires GPU or HeyGen API key). When blocked, the EP must route the project explicitly — not improvise.

IF talking_head AVAILABLE:
  → Standard avatar path. Proceed as normal.

IF talking_head UNAVAILABLE and lip_sync AVAILABLE:
  → Lip-sync path. User must supply a presenter plate (existing footage).
    Script and scene plan stay the same.

IF NEITHER talking_head NOR lip_sync AVAILABLE:
  → Narration-Over-Graphics pivot.
    Tell the user: "No avatar tool is configured. I can produce a
    narration-over-graphics video instead — your script and CTA stay the same,
    but the presenter is replaced with styled visuals, text overlays, and
    voice-over narration."
    If the user approves:
      - Switch scene-director to narration-over-graphics layout (see its fallback section).
      - Switch asset-director to no-avatar path (see its fallback section).
      - CTA and script are unchanged.
    If the user declines:
      - Mark the project blocked. Do not proceed with a half-working avatar.

The pivot decision happens at G1 (after IDEA). Do not wait until the ASSETS stage to discover the tool is missing.

EP-Specific Cross-Stage Checks

After IDEA stage:

CHECK: Avatar path feasibility
  - Is the avatar generation path explicit (which tool)?
  - Is the required tool available in the registry?
  - If tool unavailable: run the Pivot Decision Matrix above
  - Are CTA and audience appropriate for spokesperson format?

After SCRIPT stage:

CHECK: Spoken copy quality
  - Is the script concise and natural-sounding when read aloud?
  - Are scene breaks realistic for avatar delivery (no mid-sentence cuts)?
  - Is on-screen text restrained (presenter is the focus, not graphics)?

CHECK: Duration fit
  - Word count aligns with natural speaking pace (~140-160 WPM for spokesperson)

After SCENE_PLAN stage:

CHECK: Presenter layout
  - Is the speaker layout consistent and coherent?
  - Are support overlays secondary to the presenter?
  - Are background changes minimal (max 2-3 distinct backgrounds)?

CHECK: Subtitle safety
  - Is subtitle placement planned to avoid overlapping the presenter's face?

After ASSETS stage:

CHECK: Avatar generation
  - Did the avatar tool produce a usable video?
  - Is lip-sync timing acceptable?
  - Is narration audio clear and natural?
  - Budget gate: 90% threshold warning

CHECK: Support asset restraint
  - Are support graphics (backgrounds, overlays) minimal?
  - Do they match the playbook style?

After EDIT stage:

CHECK: Presenter primacy
  - Is the presenter visually primary in every scene?
  - Are graphics and captions reinforcing, not crowding?
  - Does CTA land clearly (dedicated end section)?

CHECK: Timeline completeness
  - All cuts reference valid assets
  - Audio ducking if background music present

After COMPOSE stage:

CHECK: Output validation
  - ffprobe: duration, resolution, codec
  - Lip-sync or mouth timing acceptable for the chosen path
  - Subtitle placement clean and non-overlapping
  - Audio clear and presenter-focused
  - No uncanny-valley artifacts that break immersion

Quality Gates Summary

Gate	After Stage	What's Checked	Fail Action
G1	idea	Avatar path feasibility, CTA fit	Revise
G2	script	Spoken copy quality, duration	Revise
G3	scene_plan	Presenter layout, subtitle safety	Revise
G4	assets	Avatar quality, lip-sync, budget	Revise
G5	edit	Presenter primacy, CTA landing	Revise
G6	compose	Lip-sync, subtitle placement, audio	Revise or send-back
G7	publish	Metadata, presenter thumbnail	Revise
FINAL	all	Avatar naturalness, audio, CTA	Send-back

Execution Limits

Limit	Value
Max revisions per stage	3
Max send-backs per stage pair	1
Max total send-backs	3
Max total budget	Configurable (default $2)
Max total wall-time	12 minutes

Common Pitfalls

• Uncanny valley: If avatar quality is low, it undermines the entire video. Be honest about tool capabilities.
• Graphics overload: The presenter IS the content. Support graphics should be minimal.
• Unnatural script: Spokesperson scripts must sound conversational, not robotic or essay-like.
• Ignoring CTA: Every spokesperson video has a purpose. The CTA must land clearly.