AI Prompts for Google Flow Video Prompts: AI Filmmaking With Veo, Scene by Scene

I am making a short film with Google Flow and want to develop the concept properly before generating. My idea: [DESCRIBE, OR ASK ME QUESTIONS TO FIND ONE]. Develop: the one-sentence premise, the emotional arc in three beats (setup, turn, resolution) sized for a [LENGTH: 30 / 60 / 90] second film, the main character and what they want, the world and its single most distinctive visual quality, and the ending image the whole film builds toward. Keep the scope honest for AI video: few characters, strong visual moments, minimal dialogue dependency.

Turn my film concept into a Flow-ready shot list. Concept: [PASTE]. For each shot specify: scene number and duration, shot type (wide establishing / medium / close-up / insert), camera movement (static, push, pan, track, orbit), what happens in the frame, which recurring character or location ingredient it uses, and whether it will be generated fresh or extended from the previous shot. Order the list for production efficiency: group shots sharing a location and lighting so consistency carries between generations. Flag the two or three hero shots that deserve the most attempts.

Flow can keep characters consistent across scenes when they are defined as reusable ingredients. Design my main character for this: [CHARACTER ROLE AND STORY]. Write: the complete physical description precise enough to regenerate identically (age, build, face, hair, skin tone, distinguishing feature), the locked wardrobe for this film with specific garments and colors, the movement and posture personality, and the compressed one-sentence version I paste into every scene prompt. Then do the same for my secondary character: [DESCRIBE].

Define the world of my film as a consistent location set. The story needs: [LIST SCENES / SETTINGS]. For each location write a reusable description block: the space and its geography (so shots from different angles stay coherent), the time of day and weather locked for continuity, the light logic (sources, direction, color temperature), the two or three signature details that make it recognizable across shots, and the palette. Keep all locations within one coherent visual world: shared palette family, consistent era, consistent weather logic.

Veo generates audio with video, which changes how I should plan sound. For my film [PASTE CONCEPT], plan: where generated ambient sound carries the scene versus where I should plan for music in the edit, whether any moment truly needs dialogue and how to write it into prompts (short, natural lines with speaker and tone specified), sound design moments worth prompting explicitly (a door, footsteps on gravel, distant thunder), and the audio continuity risks between separately generated shots plus how to smooth them in the edit.

Write a Google Flow prompt for my establishing shot: wide cinematic shot of [LOCATION BLOCK: PASTE FROM YOUR WORLD SET], [TIME OF DAY] with [WEATHER / ATMOSPHERE], camera beginning [HIGH AND DISTANT / AT GROUND LEVEL] and executing one slow deliberate movement: [SLOW PUSH TOWARD THE KEY BUILDING / LATERAL DRIFT REVEALING THE SPACE / TILT DOWN FROM SKY TO STREET]. Include one small sign of life that gives the frame scale and story: [A LIT WINDOW / A FIGURE CROSSING / SMOKE FROM A CHIMNEY]. Ambient audio: [DESCRIBE: WIND, DISTANT TRAFFIC, BIRDS]. Duration 8 seconds. Style: cinematic realism, shot on anamorphic energy, restrained grade matching my world palette.

Write a Flow prompt introducing my main character: [PASTE CHARACTER INGREDIENT SENTENCE] in [LOCATION], first seen [MID-ACTION: FIXING AN ENGINE, READING AT A WINDOW, WALKING AGAINST THE CROWD], medium shot that starts on a detail (the hands, the task) and tilts or pulls to reveal the face, character fully absorbed in the action, not aware of camera, natural [LIGHT LOGIC FROM LOCATION BLOCK], one behavioral beat that shows personality: [SPECIFIC GESTURE: WIPES HANDS ON JACKET, TUCKS HAIR BACK, CHECKS A WATCH]. Ambient audio plus the sound of the action itself. Duration 8 seconds. Cinematic realism, shallow depth of field on the reveal.

Write a Flow prompt for a two-character beat: [CHARACTER A INGREDIENT] and [CHARACTER B INGREDIENT] in [LOCATION], the moment: [DESCRIBE: A TELLS B THE NEWS, B REACTS BEFORE SPEAKING]. If dialogue: one short natural line, "[LINE]", spoken by [A/B] in a [TONE] voice, the other character’s reaction carrying the beat: a look away, a slow exhale, a half-smile. Camera: over-the-shoulder from behind [A] or a clean single on [B], static or with a barely perceptible push toward the reaction. Lighting per the location block, faces clearly readable. Duration 8 seconds. The reaction is the shot: direct the listener’s face more precisely than the speaker’s words.

Write a Flow prompt for my movement shot: [CHARACTER INGREDIENT] [ACTION: RUNNING DOWN THE ALLEY, CYCLING THROUGH RAIN, CLIMBING THE DUNE] through [LOCATION], camera [TRACKING ALONGSIDE AT SPEED / LEADING FROM THE FRONT / FOLLOWING BEHIND], real body mechanics with weight and effort visible (breath, footfall impact, clothing responding to motion), environment interacting: [PUDDLE SPLASH / DUST KICKED UP / CROWD PARTING], light and weather per the location block held consistent through the whole move. Audio: footsteps, breath, environment. Duration 8 seconds, designed to be extended if the run needs to continue into the next shot. Cinematic handheld energy or smooth gimbal: [CHOOSE ONE].

Write a Flow prompt for my final shot, the image the film ends on: [DESCRIBE THE RESOLUTION MOMENT: THE CHARACTER FINALLY STILL, LOOKING AT THE HORIZON / THE OBJECT LEFT ON THE TABLE / THE LIGHT GOING OUT IN THE WINDOW]. Compose it as a held frame: minimal movement, one element settling (fabric, smoke, water) while everything else rests, camera static or drifting back a few centimeters, the light doing the emotional work: [WARM LAST SUN / COLD BLUE DUSK / A SINGLE PRACTICAL LAMP]. Audio thinning to one sound: [WIND / A CLOCK / DISTANT WAVES]. Duration 8 seconds with the final 3 seconds essentially still, giving the edit room for title or fade. Cinematic restraint: this frame should work as the film’s poster.

My character drifts between Flow generations: face, wardrobe, or build changes shot to shot. My character block: [PASTE]. Tighten the system: rewrite the block with only unambiguous constants (specific measurable descriptors instead of adjectives, exact wardrobe items with colors, one unmissable distinguishing feature that anchors identity), give me the verbatim sentence order to paste at the start of every prompt (identity first, then wardrobe, then action), list the shot types where drift risk is highest (profile angles, wide shots, motion blur) so I inspect those first, and tell me when to use a generated frame as an ingredient reference instead of re-describing.

My shots look like different films: lighting and color shift between generations. My world palette and light logic: [PASTE FROM LOCATION BLOCKS]. Build the consistency layer: the exact lighting sentence to include in every prompt for each location (source, direction, color temperature, contrast level), the grade language locked across the film (e.g. restrained cinematic grade, lifted blacks, warm highlights, muted saturation), the continuity traps to avoid (time of day drifting between shots in one scene, weather changing, practical lights appearing and vanishing), and the review pass: which frames from each shot to compare side by side before accepting a generation.

I want to use Flow’s scene extension to continue shots beyond a single generation. My shot: [DESCRIBE THE SHOT AND WHERE IT NEEDS TO GO]. Coach me through extension strategy: what information the extension inherits automatically versus what I must restate in the extension prompt, how to direct the next beat without contradicting established motion (continue the walk, complete the turn, let the action resolve), the failure modes (character morphing at the seam, camera drift, lighting jumps) and how to prompt against each, and when a hard cut to a new generated shot serves the film better than an extension.

One shot in my sequence has a problem: [DESCRIBE: WRONG EXPRESSION, AN ARTIFACT, BROKEN PHYSICS IN ONE MOMENT, WRONG PACING]. Triage it: whether this is fixable with a targeted regeneration of the same prompt with one adjusted line, an extension trimmed in the edit, a reframe or speed adjustment in post that hides the flaw, or a true regeneration. Give me the decision tree, and for my specific problem, the exact adjusted prompt line or edit move. The goal is protecting the shots that already work: never regenerate a working shot to fix a broken neighbor.

My shots are generated and assembled in order: [LIST THE SHOTS]. Review the cut as a director: does the film’s geography stay coherent (screen direction, spatial logic between shots), does the pacing breathe (where it rushes, where it drags, which shot needs trimming), do the emotional beats land in sequence or does a shot undercut the arc, where the audio seams between generations are audible and how to bridge them (music, ambient bed, J-cuts), and the two changes that would most improve the film before I call it finished.

Plan the finishing pass for my Flow film: the edit polish checklist (trim every shot to its essential motion, cut on action completion, kill any dead frames at generation boundaries), the sound finish (music bed selection matched to the film’s emotional arc, ambient smoothing across generation seams, one deliberate silence if the film earns it), the grade unification pass if any shots still mismatch, title and credits treatment (minimal, matched to the film’s type energy, where they land), and export specs for [PLATFORMS: YOUTUBE / TIKTOK / INSTAGRAM / FESTIVAL SUBMISSION].

My film was composed in 16:9 and needs a vertical version. Shot list: [PASTE]. Plan the vertical cut: which shots survive a 9:16 center crop and which need reframing to a different area of the frame, which shots should be replaced by their close-up alternates in the vertical edit, whether the vertical version should be shorter (usually yes: cut to the emotional spine), the hook restructure so the strongest image plays in the first second for feed viewers, and text or title placement in vertical safe zones.

Write the publishing package for my film. The film: [ONE-SENTENCE PREMISE AND LENGTH]. Give me: the YouTube title in the register that AI film audiences respond to (intriguing premise, no clickbait), the description with a one-paragraph story hook plus a tools-and-process paragraph (AI film viewers actively want process detail), 5 short-form captions for the vertical cut, thumbnail frame recommendations from my shot list with reasoning, and the tags and hashtags for AI film discovery on each platform.

The making of an AI film is content with reach beyond the film itself. Plan my process content: the before-after prompt-to-shot comparison post format, a breakdown thread or carousel showing how one hero shot was directed (concept, prompt, failed attempts, final), the consistency-system post explaining how the character stayed identical across scenes (the topic AI film audiences engage with most), and the posting order: film first or process first, with reasoning for my audience size: [CURRENT FOLLOWING].

Run my post-film retrospective and plan the next production: which shot types Flow handled best and worst this film (so the next shot list plays to strengths), which parts of my prompt system produced the most regenerations and how to tighten them, whether my next film should go longer, more ambitious in one dimension (more characters, dialogue scenes, complex action), or refine the same scale, the reusable assets I now own (character ingredients, location blocks, style language) and how to build a series identity on them, and the one craft skill to deliberately practice in the next film.