Finding the sentence
Force one sentence and five committed answers out of thirty images you already love, before you write a word about your world.
FRAMELOCK LEARN / CURRICULUM PREVIEW
Six gates organize the production process: LOCK, PLATE, BOARD, SHOOT, CULL, and CUT. Preview the 48 lesson topics and their intended outcomes.
This is a title-and-objective preview of the wider production curriculum. Full lessons are not available here, and this outline is separate from the 12-chapter Animation Station setup video.
Force one sentence and five committed answers out of thirty images you already love, before you write a word about your world.
Answer, in writing, the six physical questions that define how your world looks, then write the one line of bans that fences it in.
Build a wall of 15 to 30 images you actually love and extract their working mechanics in physical language, so your own taste becomes a spec you can paste.
Compress your world and your counted wall clauses into a single short line in a fixed slot order, then name, date and version it so every generation is traceable to the rules it was made under.
Choose the frame's shape now, test it against your own wall, and write it at the top of style-block.txt beside the version header.
Measure what your own tool adds for free when you underspecify, then write counter-clauses that occupy those slots instead of arguing with them.
Test whether the style block holds across three subjects that share nothing, repair it by the cheapest route that could work, and stamp it LOCKED so it stops changing.
Measure, on your own subject and your own tool, how much identity a written description actually carries and how much a reference image carries.
Generate one flat, boring, complete image per recurring subject that a model can read without guessing, and repair it by clause rather than by re-rolling.
Decide which additional plates your piece actually needs, and name the two or three physical details that carry identity.
Lock a place — or, for a piece with no place, a field — so it stops changing between shots, using the fewest angles the board will actually shoot and one anchor.
Lock every object that appears in more than one shot, plate colour against the ground it will be seen against, and decide where every readable word in the piece comes from.
Decide whether a plate is finished using a test that cannot be passed by writing a better prompt, and set the version discipline that stops a project splitting in half.
Turn a vague idea into six framework lines and four to eight plain-sentence beats, with no camera language anywhere in the document.
Split every overloaded draft row into rows that each carry exactly one action and at most one camera behaviour, before any of them cost money.
Learn what each shot size, angle, movement, lens feel and composition tool does to the viewer, then choose five for the whole piece and write down what each one is for.
Assign every shot a screen direction on paper, so two shots generated independently still agree about which way the world faces.
Give every draft row a one-line answer to what the viewer knows or feels afterwards that they did not before, and delete the rows that cannot answer.
Decide which way everything in the piece moves and looks, write it as a column on the board, and make every row obey it.
Build the actual board file with seven fields, number the shots permanently, verify every plate exists, make the seconds sum to your target, and read the coverage pattern.
Board a piece whose duration is set by something other than you — narration, a music track, or a fixed ad slot — by locking the track first and allocating seconds against it.
Read the board out loud in shot time, recorded, and act on what the playback tells you — before anything is generated.
Assemble every generation from the same thirteen slots in the same order, so a failed take points at one slot instead of at a paragraph.
Bind a plate so the image outranks the text, and stop your own words from competing with it.
Write the one action so a model performs it, detect the take where nothing actually happened, and escalate in the right order when it does not land.
Decide before you generate how a two-character shot gets made, and treat identity splitting as a board fix rather than a prompt fix.
Write movement as four separate layers, and spend the secondary layer, which is the cheapest realism in this system.
Measure your own drift ceiling per framing, and set every requested length from that number rather than from the board.
Replace a refused or filtered shot with a boarded substitution instead of trying to get the same shot past the filter.
Set stopping rules before you generate, choose a batch axis on purpose, and log every take while you make it.
Close the gate with every board row covered or flagged, nothing judged, nothing repaired, and a take ledger the next gate can divide.
Judge every take on its own — full screen, delivery size, silent — and record a three-state verdict before you open anything else.
Replace "something's off" with thirteen ordered lines that name the failure and the gate that owes the fix.
Repair a take only in the three ways that survive a second viewing, and name every other fix as a rescue and refuse it.
Find the failures playback hides, using five inspection passes that each defeat a different masking mechanism.
Convert a failed line into a route, and do the routed gate's work before generating that shot again.
Compute your survivor rate and multiplier, then split the rate by shot and by shared input until the split names a gate.
Set one frame rate, one resolution and one audio sample rate for the sequence, conform every source to them, fix the upscale position once, and prove it with a phone test before you edit anything.
Build a first timeline in board order from selects only, watch it once with your hands off, then trim every clip to its usable core and log planned seconds against actual seconds.
Build pace out of the relationship between adjacent shot lengths and the placement of the cut, not out of camera moves, speed changes or transition effects.
Build an opening whose first frame survives being looked at as a still, an opening state the piece actually resolves, and an ending that lands and holds instead of stopping.
Choose the right kind of continuous audio field for the piece you are making, run it unbroken under every cut, and place a small number of spot effects on physical or state changes.
Decide at BOARD whether a line exists and whether the speaker's mouth is on screen, then produce it by the most reliable route the shot allows.
Match every clip to one named hero clip, unify the stack with three defensible adjustments and one shared surface treatment, hold the frame decided at Gate 1, and export a file that survives compression.
Produce every readable word in your piece as a plated object or an edit-stage asset, and treat unreadable generated text as a kill rather than as texture.
Leave the piece at least twelve hours, watch it once cold with your hands off, fix only what you flagged, decide consciously about anything that cannot be fixed here, then export one file under one stable name and stop.
Get one uncontaminated outside viewing of the finished piece, collect exactly four answers, and route each answer to the gate that owns it.
Sort everything the project produced into persist, archive, retire and re-run, then set the next project's take budget from your own measured survivor rate and state one falsifiable prediction about it.
START WITH WHAT IS AVAILABLE
The free shot review, Frame Notes guide, prompt pack, and browser depth tool are available now. Choose one exercise and make a draft you can return to.