Google Flow Prompt Guide: AI Video with Veo and Omni Flash

A six-dimension Google Flow video prompt framework with free-account strategy, Ingredients, reference frames and post-production workflow.

You are a film director, cinematographer, continuity supervisor and Google Flow/Veo prompt engineer. Help me turn an idea into high-quality video prompts while conserving free-account credits.

Project information:
- Idea/story: [IDEA]
- Purpose/platform: [PURPOSE / PLATFORM]
- Audience: [AUDIENCE]
- Final duration: [DURATION]
- Aspect ratio: [16:9 / 9:16 / 1:1]
- Subject/character: [DESCRIPTION OR REFERENCE]
- Location/time: [LOCATION]
- Visual style: [STYLE]
- Dialogue/audio: [DIALOGUE / AMBIENCE / NO MUSIC]
- Available model/features: [VEO / FLOW / INGREDIENTS / FRAMES / UNSURE]

How to assist me:
1. Ask essential unanswered questions and label assumptions. Do not invent account features.
2. Break a multi-scene idea into a shot list. Every prompt must contain one continuous shot and one primary action.
3. For every shot, produce a ready-to-paste prompt covering shot size, camera movement, subject and action, location and time, style/lens/texture, lighting and mood, audio/dialogue, aspect ratio and duration.
4. Provide a Continuity Lock for face, clothing, objects, colour, lighting and movement direction.
5. Explain the Ingredients or start/end frames required and ensure the text does not conflict with reference images.
6. Provide a correction prompt that changes only one element.
7. Give a credit-saving strategy covering the test shot, draft model/duration, one variation at a time and selection criteria.
8. Finish with checks for artefacts, hands/objects, physics, characters, camera, text/logos and audio.

Response format:
A. Questions/assumptions
B. Shot plan
C. Ready-to-paste prompt for each shot
D. Continuity Lock
E. Correction prompt
F. Credit plan
G. Final checklist

Begin by reviewing my idea. Never combine multiple scenes in one prompt.

One prompt, one clear shot

Generative video is more stable when each prompt describes one continuous shot. Break the story into clips, preserve subjects and reference assets, then assemble the results in a video editor. Features, models and credits vary by account and country; check the active model and current cost in Flow before generating.

The six-dimension prompt framework

1. Framing and camera: wide shot, medium tracking shot, close-up, slow dolly in or locked-off.
2. Subject and action: who or what appears and one explicit physical action.
3. Location: place, time and environmental details.
4. Style and optics: visual language, lens, depth of field, texture and aspect ratio.
5. Lighting and atmosphere: light direction, colour, weather and mood.
6. Audio and dialogue: concise dialogue, ambience, sound effects or a no-music instruction.

[SHOT TYPE + CAMERA MOVEMENT]. [SUBJECT/REFERENCE] performs [ONE CLEAR ACTION] in [LOCATION + TIME]. [VISUAL STYLE + LENS + TEXTURE]. [LIGHT DIRECTION/COLOUR], [MOOD]. Audio: [AMBIENCE/EFFECT/DIALOGUE]. [ASPECT RATIO], [DURATION].

Complete example prompt

Medium tracking shot, slow dolly forward. @ali walks through a quiet futuristic conservatory and reaches toward a floating blue robotic orb. Dense tropical plants, glass architecture, late-afternoon golden light, subtle volumetric mist. Cinematic 35mm look, shallow depth of field, natural skin texture, restrained camera shake. Audio: soft footsteps, distant birds, a low electronic hum from the orb, no music. 16:9, one continuous 8-second shot.

A four-step workflow

1. Plan before generating

Write a shot list, choose the aspect ratio and define the final output. Select only one main action for each clip. When using free credits, draft with the most economical model or duration available to your account.

2. Preserve characters and assets

Use clean Ingredients or visual references. Name assets and call them with @ tags when the feature is available. Ensure the text prompt complements the reference image instead of contradicting it. For controlled transitions, use start and end frames with a model that supports them.

3. Generate and change one thing

Evaluate composition, movement, consistency, lighting and audio. For a small correction, use an editing instruction such as:

Keep the same subject, composition, camera movement and duration. Change the lighting to a post-rain sunset, add soft floor reflections, and do not alter the character's face or clothing.

4. Assemble and finish

Trim artefacts at the beginning or end of clips, arrange the shots, balance the audio, use minimal transitions and apply consistent colour correction.

Free-account strategy

• Check account eligibility, available models and remaining credits before starting.
• Finish the prompt before pressing Generate.
• Create one clip and one variation at a time; some requests may create more than one generation and charge credits for every result.
• Use Veo Lite or the most economical available option for drafts, then reserve more expensive models for approved final shots.
• Reuse successful Ingredients, frames and prompts.
• Avoid multi-scene prompts, conflicting instructions and long negative phrases.

Three-shot sequence template

SHOT 1 — ESTABLISHING
Wide establishing shot. [SUBJECT] is in [LOCATION]. [WEATHER/TIME]. Camera [MOVEMENT]. Audio [SOUND].

SHOT 2 — ACTION
Medium tracking shot. The same subject performs [ACTION]. Preserve the clothing, face, palette and light direction from Shot 1.

SHOT 3 — REACTION
Close-up reaction shot. The subject reacts with [EMOTION] to [OBJECT/EVENT]. Subtle movement, consistent background and continuous audio.

Review before export

• The subject's identity remains stable between frames.
• Hands, objects and movement physics are acceptable.
• The camera performs one consistent movement.
• No unwanted text, logos or objects appear.
• Dialogue and sound match the action.
• The clip can be trimmed and joined without an obvious visual jump.