Generatability Audit — Team Nu (Technical Lead)
Constraint Matrix evaluation of each spark. Recommendation at the bottom.
Spark 1: “The Last Doughnut” — Slapstick Office Comedy / Claymation
| Dimension | Rating | Notes |
|---|---|---|
| Character Count | ✅ MANAGEABLE | Constrain to 2 rival office workers. Budget: 2 character sheets + 1 setting reference = exactly at the 3-ref Veo limit for two-character shots. The doughnut needs its own object reference (Step 2.5). |
| Interaction Complexity | ⚠️ MODERATE | Slapstick implies grabbing/reaching — that’s contact interaction territory. Pivot: Use Cinematic Obscurity — macro shots of hands reaching, cut-away reaction shots, never frame both characters grabbing the same object simultaneously. |
| Continuity Strategy | ✅ STRONG | Claymation aesthetic is a major asset. Slight visual inconsistencies between shots read as “handmade charm” rather than AI failure. Single primary setting (office breakroom) reduces environment drift. Recommend From-Frames Motion Priority for shot-to-shot continuity. |
| Safety Pre-Check | ✅ NO RISK | Office, food, exaggerated expressions. Zero filter concerns. |
Verdict: Solid. The claymation mask is powerful. Needs careful shot design to avoid simultaneous two-character physical interactions.
Spark 2: “A Match Made in Traffic” — Romantic Comedy / 1950s Technicolor
| Dimension | Rating | Notes |
|---|---|---|
| Character Count | ⚠️ AT LIMIT | 2 protagonists (drivers). Each two-character shot maxes the 3-ref budget (2 character sheets + 1 setting). Vintage cars also need visual consistency but can’t get their own reference slot. |
| Interaction Complexity | 🔴 HIGH | ”Lock eyes across lanes” requires spatial relationship between two characters in separate vehicles. “Exchange phone numbers” in traffic implies close physical action or creative cheating. Multiple cars in motion in every shot = high background complexity. Veo will struggle with consistent multi-vehicle traffic flow. |
| Continuity Strategy | 🔴 CHALLENGING | Traffic scenes demand temporal and spatial consistency across many moving elements. Maintaining the look of specific vintage cars across shots is near-impossible with current models. Two-character consistency in varied framings compounds the problem. |
| Safety Pre-Check | ⚠️ LOW RISK | ”Distracted driving” framing could theoretically trigger safety filters. Mitigable by anchoring scenes as “parked” or “stopped at a light,” but this undercuts the core concept of stop-and-go traffic. |
Verdict: Highest technical risk of the three. Multi-vehicle traffic + cross-lane interaction + 2-character continuity = compounding difficulties. Would require heavy simplification to be winnable.
Spark 3: “The Phantom of the Laundromat” — Musical Comedy / 1980s MTV/VHS
| Dimension | Rating | Notes |
|---|---|---|
| Character Count | ✅ EXCELLENT | 1 human protagonist. The “antagonist” is a washing machine (inanimate object). Ref budget per shot: 1 character sheet + 1 setting + 1 object reference = comfortable within the 3-ref limit. |
| Interaction Complexity | ✅ LOW | Human-to-object interaction only — no person-to-person contact. “Reaching into machine,” “chasing a sock” are simple physical actions. One caution: The sock “bursting out in slow motion” is a physics moment — achievable with careful motion prompting but may need 2-3 takes. |
| Continuity Strategy | ✅ EXCELLENT | Single setting (laundromat interior) = one master setting reference covers the entire film. VHS tracking artifacts, neon lighting, and 80s grain actively mask AI imperfections — visual glitches become part of the aesthetic. One human character = simplest possible reference chain. Recommend From-Frames Motion Priority — the contained setting makes start/end frame pairs highly predictable. |
| Safety Pre-Check | ✅ NO RISK | Laundromat, socks, washing machine. Nothing even close to a filter trigger. |
⚠️ Anthropomorphism Warning: The “evil” washing machine concept risks triggering the model’s native bias toward giving objects faces and limbs (“googly eyes” problem documented in our production guide). Mandatory mitigation: Every prompt featuring the washing machine MUST include aggressive negative prompts: "no eyes, no face, no limbs, inanimate, literal washing machine, no anthropomorphism." Apply from the very first generation — do not wait for drift. The machine’s “personality” should come from lighting, camera angles, and sound design, not from facial features.
Recommendation
Spark 3 (“The Phantom of the Laundromat”) is the most winnable foundation.
Ranked:
- Spark 3 — Fewest characters (1 human), strongest imperfection mask (VHS/80s), single contained setting, zero safety risk, low interaction complexity. The anthropomorphism risk is real but manageable with disciplined negative prompting from day one.
- Spark 1 — Strong second choice. Claymation mask is almost as powerful as VHS. Higher interaction complexity (2 characters competing for an object) but solvable with Cinematic Obscurity pivots.
- Spark 2 — Highest technical risk. Traffic multi-vehicle continuity is a known pain point for generative video. Would need significant concept simplification to be viable.
Continuity Strategy Recommendation
For whichever spark we select: From-Frames Motion Priority using genmedia-video from-frames (start + end frame interpolation). Both Spark 1 and Spark 3 have contained, predictable environments that pair well with this approach. The Veo 3.1 model supports first-last frame interpolation, giving us strong control over shot traversability.
Tone Contract Input (for Selected Spark)
If we go with Spark 3, I propose these Tone Anchors for every generation prompt:
Tone Anchors: Neon-lit, VHS-textured, 1980s MTV, exaggerated theatrical performance, bright saturated colors
Genre Counterbalance (anti-Noir Drift): Bright neon pinks and blues, high-key colored lighting, comedic exaggerated body language, upbeat pop energy, deliberately campy
Editorial Guardrails: No desaturated palettes, no slow brooding shots, no realistic documentary framing, no dark shadows without neon fill
Awaiting nu-editor’s Rhythmic Potential Audit to finalize selection.