Video Model Veo
This Skill is a Google Veo model adapter layer. Before starting, adhere to the unified model adapter contract of the video-generation dependency; this file only saves Veo-specific differences. It compiles existing SHOT.md and reused assets into executable prompts for the current Veo entry point; it does not own shot creativity, character facts, or product facts.
1. First Determine the Entry Point and Model
First, clarify the specific Google entry point the user is actually using (e.g., Vertex AI, Google's creative products, or another explicit usage entry) and record the specific model / mode.
Capabilities may differ between Google entry points and stable/preview models. When involving reference images, first/last frames, video extension, synchronized audio, resolution, duration, or input quantities, always verify against current official documentation in references/MODEL-GUIDE.md; do not infer based solely on the family name "Veo 3.1".
2. Basic Prompts
The general structure from the current Veo creation guide can be condensed into:
Cinematography language → Subject → Action → Environment / Context → Style and Atmosphere.
Write only what is truly needed for the current shot. Describe actions and performances as visible behaviors, not abstract psychological terms.
3. Image to Video
When an input image has already fixed the subject and starting composition, the prompt focuses on describing the motion, cinematography, and environmental changes that occur from this starting point.
If the current entry point supports first/last frames, treat the first frame / last frame as two explicit endpoints, and the prompt primarily describes how to move, transition, and how audio continues between them; do not let the prompt contradict the two endpoint images.
4. Reference Ingredients
Only use character, scene, object, or style reference images when the selected entry point and model currently explicitly support them. For each reference, specify its purpose, and do not let multiple assets compete for the same identity.
If the current official entry point does not support this mode, fall back to available Image to Video, first frame/last frame, or standard Text to Video paths; do not replace current facts with historical preview capabilities.
5. Audio
When the current mode explicitly supports audio generation, you may include dialogue, ambient sounds, specific sound effects, and the timing of audio in the prompt.
Audio requirements still follow the video script and shot, not adding unnecessary dialogue or sound effects to showcase model capabilities.
6. Return to video-generation
Output the selected entry point / model / mode, current prompt, input images or reference purposes, and duration / aspect ratio / resolution / audio options that the user needs to confirm in the interface; explicitly note when a capability differs by entry point.
Long-term prompts and one-shot generation packages are still managed by video-generation.
Completion Criteria
Do not conflate capabilities across different Google entry points; the prompt matches the current Veo mode; first/last frames, reference ingredients, and audio are only enabled when officially supported.