AI Film Prompts for Cinematic Shots

Recommended Visuals

None of these should be presented as the author’s own generated footage or as screenshots of work the author produced.

1. Cinematic prompt anatomy diagram — one prompt sentence with each part colour-coded and labelled (Cinematography / Subject / Action / Context / Style & Ambiance). Place directly under the “Basic Formula” H2. Original graphic.
   
2. Camera movement diagram — simple top-down icons showing push-in, pull back, truck, pan, tilt, crane. Place inside the Camera Movement H3, under the table. Original graphic.
   
3. Shot size ladder — silhouette of a figure at extreme wide → wide → medium → close-up framing. Place in the Shot Size H3. Original graphic.
   
4. Lighting types panel — the same subject rendered under soft, hard, backlit, and low-key light. Place in the Lighting H3. Original graphic, or properly licensed stock photography.
   
5. Weak vs better prompt comparison card — two side-by-side text blocks with the added elements highlighted. Place in the “Weak Prompt vs Better” H2. Original graphic.
   
6. Composition overlay examples — four frames showing symmetry, leading lines, negative space, frame within frame. Place in the Composition H3. Licensed stock imagery with original overlay graphics.
   
7. Model guidance comparison table graphic — Runway vs Veo on prompt length and negative phrasing. Place in the model-differences H2. Original graphic, sourced from the official docs listed below.
   

Optional: an official screenshot of the Runway or Google prompting documentation page could support the model-differences section, clearly captioned as a screenshot of the vendor’s documentation.

Sources Used

Primary (official documentation):

• Google — Video generation prompt guide (Gemini Enterprise Agent Platform / Vertex AI docs): https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/video/video-gen-prompt-guide — used for the prompt anatomy (subject, action, scene and context, cinematic elements).
• Google — Best practices for generating videos: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/video/best-practice — used for clear-vs-vague prompting, the single-scene rule, and character/voice consistency via repeated descriptions and a fixed seed.
• Google Cloud Blog — The ultimate prompting guide for Veo 3.1 (published Oct 16, 2025): https://cloud.google.com/blog/products/ai-machine-learning/ultimate-prompting-guide-for-veo-3-1 — used for the five-part formula and the camera / composition / lens vocabulary.
• Runway — Gen-4 Video Prompting Guide: https://help.runwayml.com/hc/en-us/articles/39789879462419-Gen-4-Video-Prompting-Guide — used for prompt simplicity, positive-phrasing-only guidance, motion-focused prompting, and the single-scene recommendation.
• Runway — Camera Terms, Prompts, & Examples: https://help.runwayml.com/hc/en-us/articles/47313504791059-Camera-Terms-Prompts-Examples — used for shot size, camera angle, composition, camera movement and focus definitions, and the note on static-camera difficulty.

All prompt examples in the article are original and were written for this piece. No prompts were copied from any source.

Fact-Check Notes

Things that could change and should be re-checked before republishing or updating:

1. Model versions. As of the August 2026 check, Google’s documentation listed Veo 2, Veo 3 and Veo 3.1; Runway’s help centre covered Gen-4 and Gen-4.5. The article

1. deliberately refers to guidance rather than version-specific behaviour wherever possible, but the two named references (“Veo 3.1 prompting guide”, “Runway’s Gen-4 guide”) will date. Re-check both docs before any refresh.
   
2. A live contradiction inside Google’s own material. The Vertex/Gemini docs’ best-practices page advises avoiding quotation marks and using a colon after the speaker’s action to denote speech. The October 2025 Google Cloud blog post advises using quotation marks for dialogue. The article does not make any claim about dialogue punctuation, so nothing needs correcting — but avoid asserting either convention in future articles without re-checking which is current, and prefer the docs page over the blog post.
   
3. Negative prompting guidance. Runway currently states negative phrasing is unsupported; Google discusses describing exclusions. This is the kind of guidance vendors revise between model releases. Verify before reuse.
   
4. Seed parameter. Google’s consistency advice references using the same seed. Seed control is not exposed in every consumer interface, so the Final Tips wording says “holding the seed steady” rather than promising the control exists in a given app.
   
5. Deliberately excluded as too volatile. Clip lengths, resolutions, aspect ratios, pricing, credit systems, free tiers and commercial-use terms were all left out of the article. These change frequently and would date the piece within weeks. If you want a version that includes them, they should be verified against official pricing and model pages on the day of publication and given an explicit “as of” date.
   
6. Unverified claims avoided. No performance benchmarks, no comparative quality scores, and no claims about which model is “best” appear in the article, because no reliable independent evidence was located for those. The lens-behaviour caveat (“these models aren’t simulating optics”) is stated as a general limitation rather than a measured finding.
   
7. No first-hand experience is claimed anywhere. The article does not state that anyone generated, tested, or paid for any tool. The “Why it works” lines explain which prompt elements are doing the work — they are craft analysis, not test results.