The Nine Wins Stack On The Still First Lock

9 wins sit on this list. They are not 9 ideas. They are 9 closed loops, and every 1 of them was paid for with a render that failed first. The structural decision underneath all of them is the still-first lock. Generate the still. Approve the still. Only then animate it. Never text-to-video on anything that has to hold an identity. The arithmetic is the whole argument. A still costs cents and seconds. A wrong still burned into video costs the video. Zero video credits are spent on a bad still, because the bad still is caught while it is still cheap. 2 phases hold the order. Render and validate every still first, archive the rejects, and only then animate the stills that passed. The order is written down in 1 line: do all the images first, check them, and only if they are perfect create the videos. The lock is mechanical rather than editorial. The approved still's CDN URL becomes the input image for the video, and the video prompt then carries motion, expression and camera only. It never re-describes appearance, because anything extra steals attention and breaks identity. 1 decision, and it moves the cost of every mistake to the cheapest place it can happen. That is why it sits underneath the other 8 rather than beside them.

The second win is the identity lock, and it is 3 locks stacked in 1 prompt block. Ethnicity, skin and hair are named verbatim in every prompt. Without them the model defaults to its training majority. On the Brazil brief the first render came back as a subject who was not Brazilian at all, and the fix was the explicit line: Afro-Brazilian, deep brown skin, honeyed sheen, long dark wavy hair. From that render on the line never left the block. 1 canonical image per subject is the face anchor, and every render is reference-locked to it. Facial features are never described in the prompt, because describing them invites drift. The anchor carries them. The body lock extends to anatomy. Breast, hip and build are defined once and held verbatim, and the second negative tier backs them up. Then the lock stops moving. The lens that opens a scene closes it. Wardrobe never changes colour. The temperature stays constant per room. 1 file holds every render of a subject, and the rule sits above the file. A render rig that hits the face 90 per cent of the time and ships the other 10 per cent is not at 90 per cent. On the frames that go out, it is at 0.

The third win is the one that made the merge possible, and it is a set of numbers rather than a set of adjectives. Exact Kelvin, position and shadow behaviour are pinned per shot. Moody adjectives were the failure. Warm intimate light produced footage that could never be matched shot to shot, because 2 people reading warm intimate light will build 2 different frames. The locked spec from the Sal y Miel production runs in 4 sources. A 2000 Kelvin match flare. A 2700 Kelvin candle as a single source, low camera-left, 45 degrees off-axis, hard shadows and negative fill. A 5600 Kelvin moon, high camera-right, with a silver rim. Then dawn amber. Because every layer graded to the same constants, separately generated shots cut together with no seam at all. That is the mechanic that let the Wiro body footage and the OpenArt face layer merge into 1 shot. 4 numbers replace a paragraph of direction, and the paragraph is what used to cost a day of matching. Write the numbers down once and the light becomes repeatable by a hand that has never met the subject.

The fourth win is 1 phrase swap that removes the gloss. Realism-biased models get pushed glossy and uncanny by the word photorealistic and by a blanket cinematic. The line that fixed it reads: documentary film, 35 mm, natural skin texture, visible pores, unretouched, natural light, negative fill. The subject stops reading as a render and starts reading as a photograph. The fifth win is the negative prompt, and it runs in 2 tiers on every render. Tier 1 is global and applies to everything: waxy and plastic skin, CGI, 3D render, beauty filter, uncanny, extra fingers, fused digits, morphing face, identity drift, text, watermark, jitter and fourth-wall breaks. Tier 2 is the body layer, and it applies to every anatomy-critical shot: deformed anatomy, mismatched size, mutated limbs, asymmetric face, elongated neck and distorted torso. The difference between looks real and looks AI is 1 block of text, and it is the cheapest lever on the whole list. Neither tier is optional. A render that ships with only the global tier is the render that comes back with 6 fingers in it.

9 wins sit on this list. They are not 9 ideas. They are 9 closed loops, and every 1 of them was paid for with a render that failed first.

The sixth win is continuity, and it is a header rather than a habit. Consecutive renders used to cut like slides. Nothing matched and the sequence jumped. The fix is the Shot DNA header: 1 identical block carrying character, location, light and colour temperature, axis, screen direction, shot, eyeline and continuity state, copied verbatim into every prompt with only the shot and the action changing. Under the header sit 13 rules. The 30 degree rule, the 180 degree rule, screen direction, eyeline, cut on action, match cut, the J and L cuts, the size ladder, motion continuity, light continuity, object continuity, and the meta-rule that ties the other 12 together. The seventh win covers the case where a face guard blocks a shot the scene still needs. The answer is the two window treatment: the frame is built in 2 passes and the locked anchor holds both of them, instead of 1 compromise being forced across the whole shot. The eighth win is the merge. When a shot cannot be produced in 1 render, 2 renders are built separately and joined, and the join only holds because both sides were graded to the same locked Kelvin set. The grade is the joint. 1 constant, or the seam shows. 8 wins so far and not 1 of them is a taste call. The ledger is a header, 13 rules, 2 windows and 1 grade.

The ninth win is the gate, and it is the reason the other 8 survive contact with a deadline. Every checker is folded into 1 pass or fail protocol. A shot is not done until all 9 benchmarks are green: identity, lens, light, location, the house look, the anti-AI block, continuity, impact and grade. 1 fail blocks delivery, and it blocks at the still rather than the video, because a broken still becomes a broken video and the still is the cheap fix. The same gate runs on the writing. 95 out of 100 is the delivery bar, and it is measured rather than felt. The numbers underneath the list are the point of the list. The probability that a first-pass render ships clean without the identity lock and the 2 negative tiers sits near 1 in 10. The probability that the same shot ships when the still is approved first, the anchor is reference-locked, the Kelvin is pinned and the gate runs before the credit is spent sits near 9 in 10. 9 wins, 1 stack. Still first, identity held, light numbered, language documentary, negatives on by default, continuity in 1 header, the guard answered in 2 windows, the join carried by 1 grade, and a gate that runs before anything is rendered for money. The list is short because the loops are closed. 12 countries of company building, 8 brands across Germany and Spain, 210 energy systems deployed across Africa and Asia and 200 wood-gasification machines built across the UK and Europe all ran on the same discipline: decide once, write it down, hold it. A pipeline that forgets does the same work twice at twice the price.