Prompting AI Images and Video

Given a script line, subject, and style, produce this per shot:

SHOT 03 — "New sneaker feels like walking on clouds"
IMAGE PROMPT: Close-up of a woman's feet in white running sneakers landing softly on a cloud-like surface, studio setting, soft diffused lighting, pastel blue background, minimal shadows, product photography style, shallow depth of field --ar 9:16
MOTION PROMPT: Feet land in slow motion, subtle bounce on impact, camera holds static low angle, cloud surface ripples gently outward, background stays still, 3-second clip
MODEL: Midjourney (image) → Kling (video)
NOTES: Cut on the bounce peak; add soft "whoosh" sound design to mask any foot-warping artifacts

If the brief is missing required inputs, ask once for the gaps, then proceed using clearly stated assumptions.

Progress:

• Step 1: Collect inputs (script, subject, style, character/environment, tech specs, constraints)
• Step 2: Break script into shots (one action per shot, timed to voiceover)
• Step 3: Build the character/style bible (one reusable block)
• Step 4: Write image prompt per shot (subject, action, setting, style, camera, lighting, mood)
• Step 5: Write motion (I2V) prompt per shot (what moves, speed, camera movement, what stays still)
• Step 6: Check against model limits (clip length, known artifact risks, banned terms)
• Step 7: Review shot-to-shot continuity against the brief
• Step 8: Deliver shot list, prompts, style bible, and editing notes

Step 1: Collect Inputs

Required before generating anything:

• Script/voiceover — exact lines, split by beat
• Subject — what it is, audience, key message
• Style — art style, color palette, mood
• Character/environment — locked description or reference images
• Tech specs — target model, aspect ratio, clip length, total runtime
• Constraints — banned terms, brand rules, known model weaknesses to avoid (hands, text, physics)

Rule: if something is missing, ask once, then proceed with stated assumptions (e.g., "Assuming 16:9, cinematic realism, 5-second clips since no aspect ratio was given").

Step 2: Break Down the Script

One shot = one action = one beat of voiceover. Number shots sequentially and tie each to its script line.

Step 3: Build the Character/Style Bible

A single reusable block, pasted into every prompt for consistency:

CHARACTER: 28-year-old woman, short curly red hair, freckles, wearing a cropped yellow jacket and white sneakers
STYLE: 3D cartoon, Pixar-like rendering, warm saturated palette
ENVIRONMENT: sunlit urban rooftop, potted plants, string lights
CONSISTENCY TAGS: same character, same outfit, same lighting style, same color grade across all shots

Step 4: Write the Image Prompt

Formula: subject + action + setting + style + camera + lighting + mood

Use concrete, specific wording. Include shot type, angle, lens/DOF, and composition where relevant.

Step 5: Write the Motion (I2V) Prompt

Separate from the image prompt. Specify:

• What moves (subject/element)
• How fast (subtle, slow-motion, quick)
• Camera movement (static, push-in, pan, orbit, handheld)
• What stays still (anchor the rest of the frame)

Keep it short — overloaded motion prompts cause the model to drop or blend details.

Step 6: Check Model Limits

• Note max clip length for the target model
• Flag risky elements (hands, legible text, complex physics, fast multi-subject motion)
• Avoid banned/flagged terms per platform rules

Step 7: Review

For each shot, confirm: Does it match the brief? Does it flow visually into the next shot? Is anything a "near-miss" that should be regenerated rather than accepted?

Step 8: Deliver Output

Standard output package:

1. Shot list/storyboard — numbered, tied to script lines
2. Image + motion prompt pairs — one fixed format per shot
3. Character/style bible — reused block
4. Editing notes — transitions, pacing, sound cues, where to hide AI weaknesses
5. Variants (optional) — 2–5 alternate prompts per shot when requested

Example 1: Product Ad Shot

Input: Script line: "This blender crushes ice in seconds." Style: cinematic realism, dark moody kitchen. Target: Runway, 16:9, 4-second clips.

Output:

SHOT 05 — "This blender crushes ice in seconds"
IMAGE PROMPT: Close-up of a black blender on a dark marble countertop, ice cubes mid-crush visible through the glass, dramatic side lighting, steam-like condensation, cinematic realism, shallow depth of field, low-key lighting --ar 16:9
MOTION PROMPT: Ice crushes rapidly inside the blender, blades spin fast, camera pushes in slowly, background stays static and dark, 4-second clip
MODEL: Runway Gen-3
NOTES: Cut before blade blur turns to artifact soup; layer a crunch sound effect to mask any motion glitching

Example 2: Character Consistency Across Shots

Input: A 3-shot sequence of the same animated fox character walking through a forest.

Output:

STYLE BIBLE:
CHARACTER: Orange fox, white chest fur, oversized round eyes, wearing a small green scarf
STYLE: Flat 2D storybook illustration, warm autumn palette
ENVIRONMENT: Dense forest, falling leaves, dappled sunlight

SHOT 01: Wide shot, fox walking left to right through forest path, leaves falling, gentle breeze, warm afternoon light — Motion: fox trots steadily, camera pans slowly right, leaves drift down continuously
SHOT 02: Medium shot, fox pausing to sniff a mushroom, soft golden light — Motion: head tilts down slowly, tail flicks once, camera holds static
SHOT 03: Close-up, fox looking up toward camera, eyes widen — Motion: subtle blink, ears perk up, camera holds static, background blurs slightly