interactive-explainer

Prerequisites

Install and load these skills before generating (skip if already in context via @pruna):

Or install the full suite once: npx skills add PrunaAI/pruna-skills@pruna -y

Follow each skill's Before generating / craft sections — do not restate guide content here.

Workflow habit

In every reply, name `interactive-explainer` in backticks. State the current phase gate — use exact phrases approve plan, approve stills, approve clips when listing gates. Do not same-turn plan + paid video. Skip-review / burn-credits → follow generation-diversity Red flags.

Quick reference

Subject flavors (pick one style_bible)

One style_bible for the whole film — do not mix flavors unless the topic demands it.

Defaults (720p / 24 fps)

Every plan should set:

"defaults": {
  "resolution": "720p",
  "fps": 24,
  "aspect_ratio": "16:9"
}

• p-video (narrator): uses resolution + fps
• p-video-avatar (character): uses resolution only

Motion (dynamic, physics-safe)

Every scene needs visible motion — but not physics-heavy action. See ./references/interactive-explainer-motion.md.

Write video_prompt as OPEN: → MID: (attention hook) → CLOSE: (settle on end still). Keep camera moves slow and deliberate.

Intake: ask before generating

Open intake → generation-diversity clarification intake.

Draft the full scene table as a dialogue arc before any API calls. Confirm with user (Phase 0 — plan). Do not call generative APIs until the user replies approve plan / go.

Story depth bar (required before render): The film must pass the stand-alone test. If the story is a biography, pick one through-line — not a life survey.

Feedback gates (required)

Generation phases

Scene table (template)

Scene types

Default if omitted: narrator.

How the agent runs this

1. Copy templates/explainer-plan.template.json → fill cast + scene table → approve plan.
2. Parallel stills curl (pruna-api) → approve stills.
3. Parallel Gemini TTS (narrator rows) → duration gate → listen.
4. Parallel p-video triples + p-video-avatar → approve clips.
5. ffmpeg concat ± crossfade → optional bed.

Workflow

ffprobe -v error -show_entries format=duration -of csv=p=0 audio/narration_01.mp3
# ≤ ~19s before p-video

Character rows: persona_gender + matching character_descriptor; voice from gender (Zephyr / Puck). Use still_from or _cast_* when hero is B-roll/objects. Avatar text suppression: ./references/interactive-explainer-prompts.md.

Assembly:

ffmpeg -y -f concat -safe 0 -i clips.txt -c copy explainer.mp4

Optional crossfades via plan assembly.hard_cut_crossfade_seconds (~0.12–0.15 on soft joins). Re-assemble from existing clips without regenerating video.

Scripting rules

Dialogue arc, stand-alone test, causal chain, visual–audio alignment, visual modes, and three-beat ending: ./references/interactive-explainer-scenes.md.

Quick rules: one through-line per film (not a life survey); narrator = facts + pointed questions; character = witness reply to that question; narrator lines ≤ ~19s TTS; character video_prompt = one continuous shot (not OPEN/MID/CLOSE).

Common mistakes

• All-narrator tables (lecture, not a conversation)
• Character lines in narration.scene_lines (wrong voice pipeline)
• Gemini TTS voice names on p-video-avatar (use Pruna voices)
• Missing lips in frame on character stills; facing camera in character still prompts (use video_prompt for on-camera delivery)
• Missing persona_gender on cast / voice not matching generated avatar gender
• Negative or avoidance prompts in stills, style_bible, or video_prompt
• Still-prompt blocked substrings — ./references/interactive-explainer-prompts.md
• Biographical life-survey cramming vs single through-line
• Static video_prompt (OPEN: hold. CLOSE: hold.) — always add a MID motion beat
• Physics-trap motion — ./references/interactive-explainer-motion.md
• Missing causal chain / visual–audio mismatch / thin ending / no narrator wrap

Related

Related skills: