Veo Prompt Director

Veo Prompt Director

Domain knowledge for generating Google Veo 3.1 video prompts based on the Universal Prompt Formula.

Prompt Input Elements

Required Elements

| Element | Description | Example | |---------|-------------|--------| | Subject | The main subject. Noun + state/appearance combination | "a man standing in a dark alley" | | Action | An observable, physical action performed by the subject. Emotions are not actions | "lowers his head and slowly steps backward" | | Style | The overall tone/genre. 1–2 clear, non-conflicting choices | "film noir", "retro aesthetic" |

Optional Elements (Quality Enhancers)

| Element | Description | Example | |---------|-------------|--------| | Cinematography | Camera movement + composition + lens/focus | "dolly shot, close-up, shallow focus" | | Context | Location, time of day, background elements | "a rainy city alley at night" | | Mood | Color tone, lighting, atmosphere | "blue-toned night lighting" | | Audio | Dialogue, sound effects, ambient noise | See [audiodirecting.md](references/audiodirecting.md) | | Duration | Clip length | 4s, 6s, 8s | | Aspect Ratio | Frame orientation | 16:9 (landscape), 9:16 (portrait) | | Negative Prompt | Elements to exclude (use affirmative descriptions of the desired opposite) | "clear sky, dry pavement" |

Input Characteristics

• A valid prompt requires all three required elements (Subject, Action, Style)
• Emotional descriptions ("looks sad") are not understood by Veo — the equivalent observable action is needed ("lowers his head and slowly steps backward"). See emotion-to-action table in [promptformula.md](references/promptformula.md)
• Conflicting style combinations (e.g., "film noir" + "bright cartoon") degrade output quality
• Optional elements default to model inference when omitted

Prompt Assembly

Formula: [Cinematography] + [Subject] + [Action] + [Context] + [Style & Ambiance]

Full spec, composition rules, and examples → [promptformula.md](references/promptformula.md)

Output Format

A generated prompt consists of:

components:
  - settings: "{duration} | {aspect_ratio} | {resolution}"
  - prompt: "assembled prompt string following the Universal Prompt Formula"
  - negative_prompt: "optional — affirmative descriptions of elements to exclude"
  - checklist: "per-element confirmation of which elements are included"

Platform Constraints

safety_filter:
  scope: "Applies to both generated video and uploaded images"
  mechanism: "Uses the same safety filters as the Gemini family"
  effect: "Prompts violating policy produce no output — the video is not generated"
  guidance: "Action/state/environment descriptions produce stable results; provocative language triggers blocking"

digital_watermark:
  technology: SynthID
  scope: "All Veo-generated videos are automatically watermarked"
  purpose: "Indicates the content is AI-generated"

Technical References

• [promptformula.md](references/promptformula.md): Universal Prompt Formula structure, composition rules, emotion-to-action table, negative prompt patterns, and model settings
• [cinematography.md](references/cinematography.md): Camera movements, composition types, lens/focus effects, and Veo-specific notes
• [audiodirecting.md](references/audiodirecting.md): Dialogue, sound effects, and ambient noise — formats and combination patterns
• [advancedworkflows.md](references/advancedworkflows.md): Timestamp prompting, First/Last Frame, Ingredients to Video, Image to Video, and Prompt Enhancement with Gemini

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: buYoung
• Source: buYoung/skills
• License: MIT

Install and usage instructions live in the source repository linked above.