Continuity Law: Composition, Chains, and Character Design
Four small rules, locked once at the start of a production, are the difference between dozens of independent generations and one film that holds together.
Four rules that hold a production together
A locked character bible solves identity within a single shot. It does not, by itself, solve continuity across dozens of shots generated independently, sometimes hours or days apart, sometimes by different people. That's a different problem, and it needs its own set of rules. This lesson covers four of them: prior-frame chaining, the board-not-video rule, screen-direction law, and archetype-driven character design. None of these are complicated individually. What makes them work is that they get locked once, early, and then held without exception for the rest of the production.
Prior-frame chaining
The first rule governs how consecutive scenes connect to each other. When a production is built scene by scene, each scene's start frame should be the prior scene's exact final frame, combined with the locked character reference.
This is the mechanism that produces a seamless cut between scenes without any post-production patching. If scene 2 starts from a fresh description of "the character walking into a room," it will generate a plausible walking-into-a-room shot — but it won't necessarily match the exact pose, lighting, and framing scene 1 ended on. Chain from scene 1's precise final frame instead, and scene 2 picks up exactly where scene 1 left off, because that's literally the image it started from. The locked character reference travels alongside the chain the whole way, so identity stays anchored even as the specific pose and environment evolve scene to scene.
The board, not the video
The second rule is a close relative of the first, and it matters most in longer productions built from storyboard-grid scenes chained across many beats. When it's time to re-anchor continuity into the next board, the source has to be the previous board image — never a frame extracted from an already-rendered video clip.
This distinction is easy to skip past because both options look like "a reference frame from the last scene." They are not equivalent. A board image is a still that has been through exactly one generation step — it's crisp, full detail, nothing lost. A frame pulled out of a rendered video clip has already been through generation, then video synthesis, then often compression — and extracting a still from it and feeding that still into another generation compounds all of that loss. Do this across several chained scenes and the compounding becomes visible: softening detail, drifting texture, a face that looks slightly less locked with every hop.
Screen-direction and composition law
The third rule governs spatial logic across a two-sided scene — any production with rivals, factions, or a dialogue between two characters. Lock one consistent screen-direction rule for the entire production: for example, side A is always screen-left, side B is always screen-right, and any clash or meeting between them always resolves center-frame. Then hold that rule for every single shot, without exception, for the rest of the production.
Viewers build a spatial map almost instantly — usually within the first two or three shots that establish the pattern — and once that map is locked in their heads, they read every subsequent shot against it automatically. Break the pattern later, even in a single technically well-composed shot, and it reads as a continuity error, not as a stylistic choice. This is true even though nothing about the individual shot is wrong in isolation; the error only exists relative to the pattern the earlier shots established. That's exactly why the rule has to be decided once, before the first shot is generated, and documented somewhere every future collaborator can check it — not re-derived from instinct on shot six.
Archetype-driven character design
The fourth rule is less about continuity between shots and more about consistency of character across a longer production — and it borrows a trick from strategy-game character design. Games that need players to read a character's personality at a glance often lean on archetypes: a compact label that implies an entire visual language. A "control" archetype reads through counters, defensive stances, summoned-proxy elements, a cool color palette, and measured pacing. An "aggro" archetype reads through brute strikes, relentless forward motion, a hot color palette, and fast pacing.
The same trick works as a directing tool. Instead of writing a long free-form personality description that has to be re-interpreted consistently by every future prompt, map each character onto an archetype at the character-bible stage — the same stage that lesson locks identity. From that point forward, "in character" has a compact, checkable meaning: does this shot's palette, pacing, and motion language match the archetype that character was assigned? A defensive, measured character who suddenly reads hot-palette and fast-paced in shot nine is off-model in the same way a face that drifted would be — it's just drifting personality instead of drifting identity.
This is exactly the logic behind a real produced piece — a crimson-vs-blue twin-rivalry action short, where one twin was locked to a cool, defensive, control-coded visual grammar and the other to a hot, relentless, aggro-coded one from the character-bible stage onward. Every shot generated afterward, across multiple sessions, inherited that mapping automatically, because the archetype had already done the translation work from personality into visual grammar.
Why all four rules get locked at the same stage
Notice where each of these rules gets decided: prior-frame chaining and the board-not-video rule are procedural — they govern how you generate each new scene, and they apply from the very first chain onward. Screen-direction law and archetype mapping are closer to character-bible-stage decisions — they get set once, early, alongside the identity lock covered earlier in this track, precisely because everything downstream depends on them staying fixed.
The common thread across all four: none of them are expensive to follow, but all four are expensive to fix retroactively. A production that locks its chaining source, its board-not-video discipline, its screen-direction rule, and its archetype mapping before the first shot generates will hold together as a single coherent piece — even if it's built across a dozen sessions by more than one person. A production that improvises any of the four per shot will accumulate small, hard-to-diagnose inconsistencies that, individually, look like nothing, but collectively read as "something's off" to any viewer watching the finished piece.
A worked failure, end to end
It's worth tracing what actually breaks when these rules get skipped, because the failure is rarely dramatic in any single shot — it accumulates quietly. Imagine an eight-scene rivalry sequence built without any of the four rules locked in advance. Scene 3 chains from a frame extracted out of scene 2's rendered clip instead of scene 2's board image, because it was the file sitting closest to hand — a small softening in detail, barely visible on its own. Scene 5 places the aggressive character on screen-left for one shot because that composition happened to look strong in isolation, breaking the screen-right pattern established in scenes 1 through 4. Scene 7 is generated by a second collaborator who wasn't told which character was meant to read as defensive and which as relentless, so both characters drift toward a generic middle-ground energy.
No individual scene in that sequence is broken. Watched as a finished eight-scene piece, though, it reads as subtly wrong throughout — soft in places for no clear reason, disorienting at the one flipped composition, and character-inconsistent by the second half. Every one of those three defects traces back to one of the four rules in this lesson being treated as optional rather than locked. The fix in each case would have cost nothing at generation time — reference the board image, hold the established side, tell the second collaborator which archetype is which — which is exactly why the rules are worth writing down once, before scene 1, rather than trusting them to memory across a multi-session production.
What comes next
Identity locked at the bible stage, coverage built through action grids, and now continuity held across chained scenes with a fixed composition law — these are the core control surfaces this track has built toward. What remains is diagnosis (reading failure modes when a generation goes wrong), credit economics (budgeting reroll spend deliberately instead of gambling it away), and the capstone: assembling every control surface covered in this track into one finished, multi-scene production.