Prompt Director: A Practical Guide

What it does

Prompt-director converts a novel, script, scene description, or creative brief into an executable prompt. When essential settings are missing, it asks questions first instead of guessing. This prevents a polished prompt from being built on the wrong aspect ratio, tool, output type, or shot density.

Two operating modes

Mode A: Creative
- Input: a novel, script, scene, or creative brief
- Flow: confirm gates -> storyboard draft -> prompts only for approved shots

Mode B: Diagnosis
- Input: an existing prompt that produces weak results
- Flow: inspect the debug-prompt workflow -> identify the failure -> rewrite the prompt

Standard workflow

1. Read cheatsheet.md
2. Read confirmation-gates.md
3. Ask for aspect ratio, tool, style, shot density, and other missing inputs
4. Return a storyboard draft
5. Wait for the user to confirm shot numbers
6. Generate prompts only for confirmed shots
7. Open at most one or two references when needed

The important sequence is confirm, draft, approve, and then generate. It avoids producing twenty prompts when the user only wanted to test two shots.

Why confirmation gates matter

A short plot summary does not contain enough information for a final prompt. Confirm the output form, aspect ratio, target tool, visual style, camera tone, and shot granularity first.

Required confirmations:
- output_type: image or video
- aspect_ratio: 16:9 / 9:16 / 1:1
- target_tool: Midjourney / GPT Image / Gemini / Seedance / Kling
- visual_style: realistic / anime / Chinese-inspired / commercial / documentary
- camera_tone: candid / cinematic / emotional / product still life
- shot_granularity: direct prompts or storyboard first

• Ask first when the user did not say whether the result is an image or video.
• Ask first when landscape or portrait output is unclear.
• Recommend a storyboard draft when the plot has no camera direction.

How to layer references

Do not load every reference into the context at once. Read the cheatsheet for frequent rules, use workflows for task-specific procedures, and open references for focused topics such as composition, lighting, camera movement, consistency, or color.

Prompt precision rules

Replace abstract adjectives with visible constraints. Specify camera language, light direction, color temperature, composition, depth of field, and exclusions. Write the subject and base lighting before adding atmosphere words; negative prompts should target a concrete failure such as plastic skin, dirty light, or unwanted framing.

• Pair camera angle with focal length so the visual intent is testable.
• Use small physical actions to make expressions and performances feel alive.
• Manage character, scene, position, and scale consistency as separate constraints.
• Build a color palette or visual DNA before generating a series.

Writing image-to-video prompts

An image-to-video prompt should describe what changes after the first frame, not rewrite the first frame. Keep it focused on action, camera movement, and timeline changes.

For image-to-video, write only:
- the main action
- the camera movement
- changes over time

Do not repeat:
- 8K / cinematic / hyper-realistic claims
- lighting and style that already exist in the first frame
- conflicting camera instructions

For still-image prompts, use this six-part skeleton:

1. Camera position, shot size, and focal length
2. Light direction and time of day
3. Environment plus foreground, middle ground, and background
4. Subject action or state change
5. Small but important visual details
6. Negative constraints and exclusions

Debugging and recovery

A failed take is not always a reason to restart. Keep the composition, lighting, or color that already works, then repair the specific failure.

• Fake expression: replace broad emotions with a small emotional cue and a physical action chain.
• Fake camera move: keep one dominant movement instead of mixing a smooth push-in with aggressive handheld motion.
• Drifting scene: build a spatial master first, then split it into shots.
• Broken product: redraw keyframes and send the keyframe sequence back to the video model.
• Dirty lighting: preserve a well-composed failed take and repair the action instead of regenerating the whole frame.

For a focused image-to-video prompt workflow, continue with the I2V prompt guide. For narrative production, see ai-video-director; for ecommerce and advertising, see ecommerce.