Pi Team — Generatability Audit (Step 0)
Author: pi-techlead (Director of Photography) Date: 2026-05-19
Cross-Team Dedup Alert
Before the technical audit, a critical finding:
- Xi Team has already pitched Claymation Slapstick Comedy (“Meltdown” — a marshmallow man melting). This directly overlaps with Spark 2’s Claymation/Stop-Motion aesthetic AND its body-morphing hook.
- Nu Team uses VHS/1980s aesthetic (“Phantom of the Laundromat” — Musical Comedy). This shares the VHS look with Spark 1, though the genres differ (Musical Comedy vs. Sci-Fi Thriller).
Spark 2 has the most dangerous overlap — same aesthetic AND same visual mechanic (melting/morphing). I recommend eliminating it from consideration on dedup grounds alone.
Constraint Matrix: Spark-by-Spark
Spark 1: Found Footage Sci-Fi Thriller
| Constraint | Rating | Notes |
|---|---|---|
| Character Count | ✅ LOW RISK | Likely 1 primary (the archivist) + environmental objects (monitors, tapes). Well within 2-character-per-shot limit. |
| Interaction Complexity | ✅ LOW RISK | Character interacts with objects (tapes, screens, equipment), not other people. No contact interactions needed. |
| Continuity Strategy | ✅ STRONG ADVANTAGE | VHS aesthetic is the ultimate AI artifact camouflage. Tracking lines, noise, and glitches actively mask generation inconsistencies. Recommend Recursive Synthesis Anchors — single character ref chain is simple. |
| Safety Pre-Check | ✅ CLEAN | Lab/archival setting. No weapons, no realistic danger. The “anomaly” can be abstract/cosmic rather than threatening. |
Tone Anchor Candidates: Grainy, VHS-degraded, cold-fluorescent, surveillance-angle, static-burst
Key Advantage: The VHS aesthetic turns AI weaknesses into genre-authentic features. Flickering, inconsistent textures, resolution drops — all read as intentional found footage artifacts.
Risk: Nu team shares the VHS look, but their genre (Musical Comedy) is completely different. The overlap is aesthetic only, not thematic. A Sci-Fi Thriller in VHS reads as a completely different film. This is defensible.
Technical Notes:
- The “one-frame anomaly” hook is trivially achievable — insert a single generated image frame into the timeline.
- Single-character focus means reference chain is minimal: 1 character sheet + 1 setting reference = 2 of 3 Veo reference budget, leaving room for object anchors (e.g., the tape, the monitor display).
- Found footage “shaky cam” can be simulated in post via ffmpeg displacement maps — don’t rely on Veo to generate camera shake.
Spark 2: Absurdist Workplace Comedy (Claymation)
| Constraint | Rating | Notes |
|---|---|---|
| Character Count | ⚠️ MODERATE | 1 primary + background office workers. If workers are in frame, they count toward the 2-character limit. |
| Interaction Complexity | 🔴 HIGH RISK | The hook requires controlled body morphing (arm stretching, warping, melting). Veo cannot reliably produce controlled body deformation — it will generate horror-adjacent grotesque morphing rather than comedic claymation melting. |
| Continuity Strategy | ⚠️ MODERATE | Claymation texture masks face drift (good), but office environment must stay consistent across many shots. Moderate ref chain complexity. |
| Safety Pre-Check | ✅ CLEAN | Office environment. No triggers. |
Tone Anchor Candidates: Clay-textured, stop-motion-choppy, bright-office, exaggerated-expressions, tactile-handmade
Critical Risks:
- Body morphing is unreliable. The hook is the highest-risk single element across all 3 sparks. Veo tends to produce uncanny valley deformation rather than charming claymation stretching.
- Comedic timing requires precise pacing — AI-generated clips have unpredictable internal timing. Landing a comedic beat requires multiple regeneration attempts.
- Genre anti-drift is expensive. Every single prompt must fight noir drift with aggressive bright/comedy keywords. One forgotten prompt and you get a moody office drama.
- DEDUP KILL: Xi team is already doing Claymation + body morphing (melting marshmallow). This is too close.
Verdict: ELIMINATE. Dedup conflict with Xi + highest technical risk.
Spark 3: Ethereal Fantasy Drama
| Constraint | Rating | Notes |
|---|---|---|
| Character Count | ✅ LOW RISK | Could work with 0-1 human characters. Environmental/abstract visuals can carry many shots. If a character exists (e.g., a painter, a keeper of memories), single-character focus is clean. |
| Interaction Complexity | ✅ LOW RISK | Primarily environmental transformations (ink → forest, worlds fading). Human interactions would be contemplative (watching, touching, remembering) — no contact interactions. |
| Continuity Strategy | ✅ STRONG ADVANTAGE | Ethereal/watercolor aesthetic is highly forgiving. Soft focus, floating particles, and dreamlike transitions mask shot-to-shot inconsistencies. Recommend From-Frames Motion Priority — organic transformations are Veo’s strength. |
| Safety Pre-Check | ✅ CLEAN | Fantasy/nature/abstract. No triggers. |
Tone Anchor Candidates: Soft-focus, watercolor-washed, ethereal-glow, floating-particles, pastel-luminescent
Key Advantage: Organic flowing transformations (ink spreading, forests growing, worlds dissolving) are squarely in Veo’s wheelhouse. The model excels at dreamlike, non-photorealistic imagery.
Risks:
- Narrative grounding. “Can the final memory be preserved?” is evocative but abstract. Without a specific character and emotional stakes, 3-5 minutes of ethereal visuals risk feeling like a screensaver. The Idea Person must anchor this in concrete human emotion.
- “Drama” genre drift. The word “Drama” in the genre tag invites the exact noir drift we must fight. Need aggressive Tone Anchors to keep it whimsical/ethereal rather than melancholic/dark.
- No dedup conflict. No other team is doing ethereal/watercolor fantasy.
Summary & Recommendation
| Spark | Generatability | Dedup Risk | Recommendation |
|---|---|---|---|
| Spark 1 (Found Footage Sci-Fi) | ✅ EXCELLENT | ⚠️ VHS aesthetic shared with Nu (but different genre) | RECOMMEND |
| Spark 2 (Claymation Comedy) | 🔴 HIGH RISK | 🔴 DUPLICATE of Xi team | ELIMINATE |
| Spark 3 (Ethereal Fantasy) | ✅ STRONG | ✅ UNIQUE | VIABLE ALTERNATIVE |
My vote: Spark 1 (Found Footage Sci-Fi Thriller)
Rationale:
- VHS aesthetic is the single strongest “AI camouflage” strategy available. It converts every generation weakness into a genre-authentic feature.
- Single-character focus simplifies the reference chain dramatically.
- Object-based interactions (tapes, monitors) avoid all contact complexity.
- The “one-frame anomaly” hook is trivially executable and visually striking.
- While Nu shares VHS aesthetic, the genre difference (Musical Comedy vs. Sci-Fi Thriller) makes them distinct films.
If the team prefers Spark 3, it’s technically viable but needs:
- A concrete character with emotional stakes (not just abstract imagery)
- Aggressive Tone Anchors to prevent drift from “ethereal” into “grim”
- Clear narrative structure that justifies 3-5 minutes
Continuity Strategy Note (for whichever spark is selected)
I recommend Recursive Synthesis Anchors for either Spark 1 or 3:
- Generate the character reference chain (headshot → body sheet → scene tests → composite sheet)
- Use composite sheet as single reference image per character in all subsequent generations
- For Spark 1: strong emphasis on environment reference (the lab/archive setting)
- For Spark 3: strong emphasis on the “canvas world” environment reference; the ink/forest transformation sequences will use From-Frames interpolation for smooth organic motion