The Prompt Is Subject Plus Motion Plus Camera

The Wan 2.7 prompt that ships is a 5-part formula and the 5 parts have to land in this order every single time. Subject plus a fixed identity block. Action plus motion. Camera plus framing. Lighting plus atmosphere. A quality trigger that names documentary film and 35 mm. Subject is the first 15 words of the prompt. The first 15 words name the person or the place or the object that the camera is pointed at, and the first 15 words set the eye on the right noun. A prompt that opens with adjectives instead of a noun is the failure that loses the eye in the first 200 milliseconds, and the lost eye is the failure. Action plus motion is the next 10 to 20 words. The next 10 to 20 words name what the subject is doing and how the subject is moving, and the verb is held in the present tense so the closed box reads the verb as a present-tense instruction. A prompt that names the action in the past tense is the failure that the closed box reads as a memory, and the memory drifts the action. Camera plus framing is the next 5 to 15 words. The next 5 to 15 words name the lens and the framing (35 mm, wide close, locked-off tripod, handheld, push in, pull out, dolly left, dolly right), and the camera is what the operator can choose per shot. A prompt that names the camera only at the end is the failure that loses the framing, and the lost framing is the failure. Lighting plus atmosphere is the next 5 to 15 words. The next 5 to 15 words name the light source and the climate (tropical sunrise, neon alley, sodium-vapour streetlamp, overcast morning, warm window light, candle at dusk), and the named light is what the closed box reaches for. A prompt that names the light as bright or moody is the failure that the closed box reads as a guess, and the guessed light drifts the look. The quality trigger is the last 5 to 10 words. The last 5 to 10 words name the medium and the grade (documentary film, 35 mm, natural skin texture, natural light, handheld, unretouched), and the named medium is what the eye reads as a film still. A prompt that lands without a quality trigger is the failure that the closed box fills with its default look, and the default look is the glossy super-AI look that the eye flags as fake.

The fixed identity block is the second noun in the formula, and the block goes in every character prompt verbatim. The block names the ethnicity, the skin tone, the hair texture, the age, and the wardrobe in a single sentence, and the sentence is held consistent across 8 shots and 100 shots and the whole film. Ethnicity is named first because Wan 2.7 defaults to its training-data majority look on any character prompt that leaves ethnicity unspecified. The default is often Asian-featured, and the Asian-featured default on a Salvador shoot is the failure that flips an Afro-Brazilian girl into a character that does not match the film. The fix is to name the ethnicity on every character prompt with 1 of 3 phrases: Afro-Brazilian, Black Brazilian, or deep brown skin. The named ethnicity is the cheap gate. Skin tone is named second, and the skin tone is the named gate that holds the chroma below the clip on every face. The named phrases are deep brown skin, warm brown skin, and the named phrase is held the same way the named ethnicity is held. A character prompt that names the ethnicity without naming the skin tone is the failure that drifts the chroma toward orange, and the orange chroma is the failure. Hair texture is named third, and the hair texture is the named gate that holds the silhouette consistent across cuts. The named phrases are tightly coiled dark curls, gray curls, and dark wavy hair, and the named phrase is held across every shot of the same character. A character prompt that names the ethnicity without naming the hair texture is the failure that drifts the silhouette, and the drifted silhouette is the failure. Age and wardrobe are named fourth, and the named pair is the gate that holds the casting consistent across cuts. The named phrases are 6 years old and white linen dress, and the named pair is held verbatim across every shot of the same character. A character prompt that names the ethnicity without naming the age and the wardrobe is the failure that drifts the casting, and the drifted casting is the failure. The full block is prepended to every character prompt, and the block reads as 1 sentence: Anna, an Afro-Brazilian girl, deep brown skin, tightly coiled dark curls, 6 years old, white linen dress, gold hoops, barefoot. The 1-sentence block is held verbatim across every shot, and the held block is the cheap gate that holds the face across 8 shots.

The first failure mode that bit the Salvador v1 render was ethnicity drift. The first render left ethnicity unspecified (a barefoot girl, a grandmother), and the closed box defaulted to its training-data majority look, and the majority look was Asian-featured. The Asian-featured default is the failure that flipped an Afro-Brazilian girl into a character that does not match the film. The root cause is the closed box has no memory between generations. A character prompt that names the ethnicity in shot 1 is forgotten by shot 2 unless shot 2 re-states the ethnicity, and the re-state is the gate. A film that re-states the ethnicity on every character prompt is a film that holds the ethnicity across the whole cut, and the held ethnicity is the cheap gate. The fix is the ETHNICITY LOCK rule. The rule says ethnicity is named on every character prompt without exceptions, and the rule is held the same way the 30 degree rule is held on every cut. A character prompt that omits the ethnicity is a prompt that ships at the model's default, and the model's default is the failure. Skin tone is added because the named skin tone holds the chroma below the clip on every face. The named phrases are deep brown skin, warm brown skin, and the named phrase is held the same way the named ethnicity is held. A character prompt that names the ethnicity without naming the skin tone is the failure that drifts the chroma toward orange, and the orange chroma is the failure that reads as fake on every face. Hair texture is added because the named hair texture holds the silhouette consistent across cuts. The named phrases are tightly coiled dark curls, gray curls, and dark wavy hair, and the named phrase is held across every shot of the same character. A character prompt that names the ethnicity without naming the hair texture is the failure that drifts the silhouette, and the drifted silhouette is the failure. The default for the Salvador shoot is Afro-Brazilian because Bahia is the heart of Afro-Brazilian culture, and the default is held across Anna's younger self and the grandmother and every supporting character. A film that holds 1 default across the whole cast is a film that reads as 1 cast, and the held cast is the cheap gate.

The second failure mode that bit the Salvador v1 render was the super-AI glossy look. The first render carried the word photorealistic and the blanket phrase cinematic film still and the blanket phrase film grain, and the 3 phrases pushed the render toward a polished but obviously synthetic look. The polished look is the failure that the eye flags as a digital painting. The root cause is Wan 2.7 is already realism-biased. The closed box defaults toward realism on every prompt, and the word photorealistic pushes the closed box toward a hyper-realism that reads as a digital painting. The hyper-realism is the failure that loses the documentary naturalism, and the lost naturalism is the failure. The fix is to drop photorealistic. Drop cinematic as a blanket adjective. Use documentary language instead: documentary film, candid, handheld 35 mm, natural skin texture, visible pores, fine laugh lines, unretouched, natural light. The documentary language names the camera and the light and the skin, and the named look is what the eye reads as a film still. The blanket adjectives are dropped because the blanket adjectives push the closed box toward the model's default look. The model's default look is the polished glossy look, and the polished glossy look is the failure that the eye flags as fake on the first frame. The probability that a prompt carrying photorealistic and cinematic together ships a film still sits at about 5 per cent against the probability that a prompt carrying documentary language and natural light ships a film still at about 90 per cent, because the closed box defaults toward realism already. A prompt that lands with documentary language and no blanket adjectives is a prompt that the closed box reads as a film still, and the film still is the cheap gate. Tears and emotional shots get their own fix. The glossy tear (eyes wet, crying) is the failure that flips the face into a beauty filter. The fix is to name the tear at restraint: eyes glistening, holding back, or a single tear caught in the light. The restrained tear is what the eye reads as a real tear, and the real tear is the cheap gate. The negative block carries the rest of the fix. The block lists glossy, airbrushed, plastic skin, CGI, 3D render, digital painting, over-smooth, beauty filter, uncanny, doll-like, wax figure, cartoon, anime. The block is appended to every character prompt, and the appended block is the cheap gate that holds the closed box below the polished look. Face close-ups get negative fill on top of the documentary language. Negative fill shapes the face with shadow and contrast, and the shadow-and-contrast face is what the eye reads as a 3D face on a 2D image. A face close-up that ships without negative fill is a face close-up that ships flat, and the flat face is the failure.

The Wan prompt that works is a 5-part formula, and the 5 parts have to land in this order every single time.

The revised prompt pack for the Salvador retry holds 8 scenes across 1 identity, and the 8 scenes share the same 5-part formula. Scene 1 is the city aerial: documentary aerial drifting over Salvador, Bahia, terracotta rooftops stacked down to a glittering Atlantic, heat shimmer, distant green hills, gulls, natural color, warm amber light, slow drift, thin haze, 35 mm film. The aerial holds 8 nouns in 30 words and the 30 words land the 5 parts. Scene 2 is the kitchen: an Afro-Brazilian grandmother's weathered hands kneading dough, deep brown skin, mango halves on a worn wooden table, steam in warm window light, a small Black Brazilian girl's bare feet swinging under the table. Documentary film, 35 mm, natural skin texture, natural light, handheld. The kitchen holds 2 characters and 1 climate, and the named climate is what the closed box reads as the kitchen light. Scene 3 is the street: a Black Brazilian girl, deep brown skin, tightly curled dark hair, running barefoot down a sunlit dirt street, jasmine in the air, laundry lines overhead, other children laughing, a drum circle beginning. Candid documentary, 35 mm, natural light, motion. The street holds 1 character in motion, and the named motion is what the closed box reads as the running gait. Scene 4 is the lesson: an Afro-Brazilian grandmother, brown skin, gray curls, leaning in and cupping a young Black Brazilian girl's face, teaching her to smile, the girl's face learning it, eyes crinkling. Natural window light, documentary 35 mm, natural skin texture, negative fill on faces. The lesson holds 2 faces in close-up, and the negative fill is what the eye reads as the cup. Scene 5 is leaving: a young Afro-Brazilian woman's face reflected in a bus window at dusk, deep brown skin, dark curls, eyes glistening but steady, a small smile forming, the lights of Salvador sliding past. Documentary, 35 mm, natural skin texture, not airbrushed, unretouched. The leaving holds 1 face and 1 reflection, and the reflection is what the eye reads as the inside. Scenes 6 to 8 are the Anna reference shots, and the reference shots use 00_unsorted_unsorted_anna_reference.0V1.0.jpg as the start frame for image-to-video identity lock. Shot 6 is Anna turning to camera and smiling wide and genuine with laugh lines showing. Shot 7 is her hands mid-gesture with a gold necklace catching the light. Shot 8 is a slow push-in on her smile with the eyes first and then the mouth, held. The 3 reference shots hold the identity across the cut. The 8 scenes share the same 5-part formula and the same fixed identity block, and the shared formula is the cheap gate that holds the film across 8 shots. A film that ships 8 shots on 1 formula is a film that reads as 1 film, and the 1 film is the cheap gate.

The production notes are held across the retry and the notes are 4 lines and the 4 lines name the wardrobe, the light, the camera, and the grade. A retry that holds the 4 lines is a retry that holds the look, and the held look is the cheap gate. Wardrobe is white linen plus gold hoops because the Salvador shoot sits in the heat of Bahia, and the white linen is what the eye reads as the warmth. The Sal y Miel champagne silk slip stays on the Sal y Miel film. A wardrobe that drifts from white linen to champagne silk on the Salvador retry is a wardrobe that flips the film, and the flipped wardrobe is the failure. Light is 2 sources: window moon plus table lamp. The 2 sources are warm and never cold, and the warm light is what the eye reads as the domestic interior. A light that ships cold is a light that flips the interior into a hospital, and the flipped interior is the failure. Camera is the lens that never rushes. She moves, the lens leans in. The lean is held at a 35 mm prime on a locked-off tripod or a handheld, and the lens holds the eye on the subject without rushing the subject's motion. A camera that rushes the subject is a camera that loses the eye, and the lost eye is the failure. Grade is teal-shadow plus amber-honey, grain about 20 per cent, no green anywhere. The teal-and-amber grade is the celluloid look, and the grain at about 20 per cent is what the eye reads as a 35 mm print. A grade that lands without the grain is a grade that lands digital, and the digital grade is the failure. Operating rule is 1 sentence and the sentence holds the formula. Render with the 5-part formula, prepend the fixed identity block on every character prompt, drop photorealistic and cinematic, name documentary language and 35 mm, append the Tier 1 negative block, and grade to teal-and-amber at about 20 per cent grain. The held formula is what makes 8 shots read as 1 film.