Write Veo prompts with an agent
Turn a video idea into a shot list, clear camera directions and editable prompts, with reference roles and revision choices defined before generation.
Write Veo prompts with an agent by deciding how a shot should be made before polishing its wording. The ai-cinematographer skill plans coverage, continuity targets and prompts for Veo 3.1 and Nano Banana 2. It distinguishes text-led generation, image-led motion, reference-based work and start/end-frame planning instead of treating every idea as the same kind of request.
Its output is a production plan and prompt text. The skill does not itself generate an image or video. A still-image prompt can prepare a starting frame, while the video prompt concentrates on motion. Available modes and limits depend on your chosen tool and account. Reference and preservation instructions guide the model but do not guarantee consistent faces, products or style across clips.
What it does
- Turn an idea into a shot list
- Choose a suitable generation mode
- Describe camera movement clearly
- Separate changes from preservation constraints
- Plan a still frame before animation
- Prioritise revisions after reviewing output
Install and use
- Install the skill and describe the aim, audience and output format. Writing prompts needs no generation account; producing media needs separate compatible model access and may cost money.
- Give a concrete brief: “Plan three shots for this product. Specify camera motion, reference roles and what must stay fixed, with a revision path for each shot.”
- Review the selected modes before generating. Use references you are entitled to supply and make the permitted changes clear for sensitive faces or products.
- Inspect generated results and change one or two important controls at a time. When a limitation comes from the model or tool, a longer prompt may not solve it; a stronger reference or simpler shot may help.
FAQ
Source
Made by TheWillMundy. Original skill: ai-cinematographer, published on GitHub under the MIT licence.
View on GitHub