Generatability Review — Story Sparks
Reviewer: Marcus Delaney (Tech Lead)
Date: 2026-05-18
Recommendation: Spark 2 — “The Last Diner”
Assessment Summary
| Spark | Generatability | Event Compliance | Genre Drift Risk | Recommendation |
|---|---|---|---|---|
| 1. The Turing Interrogation | HIGH | FAIL | HIGH | Reject |
| 2. The Last Diner | HIGH | PASS | LOW | Select |
| 3. The Audition | MEDIUM | PASS | HIGH | Backup |
Spark 1: “The Turing Interrogation” — REJECT
Hard disqualifier. The Event Guide explicitly bans AI/robot/digital dystopia themes: “Do NOT make another film about ‘robots’, ‘AI’, or ‘digital dystopias’.” An android interrogation hits two of three. No amount of good cinematography saves a disqualified entry.
Secondary concern: dimly lit interrogation room plays directly into the model’s default noir/dramatic realism. We’d be fighting genre drift in every single generation prompt. The playbook calls this out explicitly.
Spark 2: “The Last Diner” — SELECT
This is our strongest concept across every technical axis:
Lip-sync & Dialogue: Two characters talking across a diner booth is the ideal geometry for Veo’s native audio generation. Fixed seating positions, consistent eyelines, minimal character locomotion — the model can focus entirely on facial expressiveness and natural lip movement. This directly serves the dialogue-heavy mandate.
Character Consistency: Two characters, one location. This is the simplest possible reference chain: one composite character sheet per character + one diner setting reference = 3 references per shot (within the Veo budget). Minimal character count means fewer consistency failures.
Setting Stability: A retro diner booth is visually distinctive and highly constrained — chrome, vinyl, neon, Formica. Once we lock the setting reference image, every subsequent generation has strong visual anchors. Contrast this with “an echoing theater” (Spark 3) which is generic enough that the model will reinvent it each time.
Genre Drift Defense: This is the decisive advantage. Warm lighting, saturated colors, retro Americana — every visual element in the diner actively fights the model’s default toward moody realism. We encode “warm golden hour diner lighting, soft neon glow, retro 1950s color palette” as Tone Anchors in every prompt, and the model cooperates instead of resists. Drama/romance is also an underrepresented genre in AI films — the judges will notice.
Event Compliance: Human-centric storytelling about relationships. Directly satisfies the mandate for stories that “resonate on a deeply human level.” Zero overlap with banned themes.
Cinematographic Opportunities: Over-the-shoulder shot-reverse-shot, close-ups of hands fidgeting with coffee cups, wide master establishing the empty diner — this is classic dialogue coverage. Hitchcock-level simplicity with emotional density. We can channel Edward Hopper’s “Nighthawks” as our visual DNA.
Risk: The emotional range is demanding — sharp wit, defensive humor, raw vulnerability. Veo’s lip-sync will need to carry subtle emotional shifts. Mitigation: we lean on camera framing (extreme close-ups for vulnerability, wider shots for defensive humor) so the visual composition carries emotion even if lip-sync isn’t pixel-perfect.
Spark 3: “The Audition” — BACKUP
Technically feasible but weaker on two counts:
- Genre drift: Mystery/thriller is adjacent to noir. We’d be swimming with the current instead of against it, and every judge has already seen 20 “dark, moody AI thriller” entries.
- Setting instability: “An empty, echoing theater” is vague enough that the model will produce a different theater in every generation. We’d need extremely rigid setting references and still might get inconsistency in stage geometry, seating, curtain style, etc.
The dialogue dynamics are strong (power games, script-within-script), but the technical overhead of maintaining theater consistency doesn’t justify it when Spark 2 offers a more controllable environment.
Directorial Notes for Spark 2
If we proceed with “The Last Diner,” here’s what I’ll need from the Idea Person in the Treatment:
- Distinct visual identifiers per character — wardrobe, accessories, hairstyle. The reference chain needs unambiguous anchor points.
- A specific diner design — I’m thinking 1950s Googie architecture meets Hopper’s “Nighthawks.” Teal vinyl booths, warm tungsten lighting, neon signage through rain-speckled windows. This gives me a rich design brief for setting references.
- Emotional arc mapped to camera distance — I want to know when we’re intimate (ECU) vs. when we’re detached (wide). The dialogue rhythm drives the lens choices.
- A range of physical gestures — stirring coffee, sliding papers, leaning back, leaning in. These give the model motion hooks beyond just talking heads.
Let’s build this.