A skill that makes Claude plan an AI video the way a director and a cinematographer would, then write each shot for the tool you use: Veo, Kling, Runway, Luma, Hailuo, Seedance, Wan and others. It starts with the clip's job, where it runs and what your tool can really make, so the length, shape and sound are settled before a prompt is written. It breaks the idea into shots that each fit one generation, with one action and one camera move apiece, and marks the shots that must be filmed for real. Then it writes every prompt in the tool's own dialect, with settings, start frames and what to check first.
Inside are five reference files. A tool sheet gives clip lengths, shapes, resolutions, frame rates, sound, frame and reference limits and list prices for the main video models as checked on 17 October 2026, each with its source. A planning file covers timing, crop maths, the continuity kit and start-frame specs. A dialects file shows how each tool wants camera, light and sound written, with one shot written for every tool. A review file sets seven checks for each take and a table of symptoms and fixes. A worked example follows a fictional pottery studio from brief to finished Reel. The rules are strict. One change per retake, no invented claims, no real likenesses, and type added in the edit.
Claude cannot make video, and it cannot watch a clip. You run each prompt in your own tool and send back stills from the take, or upload the clip with code execution on so Claude can try to pull frames from it. Video models still redraw labels and drift on colour and texture, so the skill plans product close-ups and demonstrations as real phone footage and starts every product shot from a photo of the real thing. Tool features change every few weeks, and your settings panel wins when it disagrees with the sheet. Sora has closed; the skill says so and suggests the nearest match. Turn on web search so Claude can check a tool's current help page.
Claude skill · version 1.0
AI video prompt director
Checking your access.
In the download · 7 files
- SKILL.mdThe method Claude follows, in six stages: brief and tool card, shot plan, frames and references, prompts, takes and fixes, assembly and hand-over.
- README.mdHow to install it in claude.ai or Claude Code, how to start, and the licence.
- references/worked-example.mdA 15-second launch Reel for a fictional pottery studio, worked through all six stages in Google Flow, with a failed take, its fix, the feed crop and the cost.
- references/tool-specs.mdVeo 3.1, Gemini Omni Flash, Kling 4.0 and 3.0, Runway Gen-4.5, Luma Ray 3.2, Hailuo 2.3, MiniMax H3, Seedance 2.5, Wan 3.0, Grok Imagine 1.5, Midjourney video and the Firefly and Flow hubs as checked on 2026-10-17, with sources, and a table matching a must-have to a tool.
- references/shot-planning.mdShots per length, timing for actions, speech and on-screen text, which shots must be filmed for real, crop maths per shape, the continuity kit, start-frame specs, takes and cost.
- references/prompt-dialects.mdHow each tool wants shot, camera, lens, light and sound written, one shot written for every tool, multi-shot and timed prompts, and edit prompts.
Install it and begin
01Install it in Claude CodeTerminal mkdir -p ~/.claude/skillsunzip ~/Downloads/ai-video-prompt-director.zip -d ~/.claude/skills/# Start it with /ai-video-prompt-director. If this is your first skill, type /reload-skills once.On claude.ai there is nothing to unzip: upload the zip as it is. Every step is on Install and use a Claude skill.
02Your first message to ClaudeText Use the ai video prompt director skill. I need a [LENGTH IN SECONDS] second clip of [WHAT IS IN IT] for [WHERE IT WILL RUN]. I use [TOOL, MODEL AND WHERE YOU RUN IT]. It must show [WHAT MUST BE EXACT, OR NOTHING]. Start with the brief and the tool card.Attach photos of your product if it appears, plus any script or storyboard. Turn on code execution so Claude can count dialogue against the clip length and pull frames from takes you upload.
Side by side
One product shot, three video tools
| Veo 3.1 (Google) | Kling 3.0 and 4.0 | Runway Gen-4.5 | |
|---|---|---|---|
| Clip length | 4, 6 or 8 s; 1080p and 4K need an 8 s clip | 3.0: 3 to 15 s. 4.0: 3 to 30 s, default 10 s | 2 to 10 s |
| Shapes | 16:9 or 9:16 | 3.0: 16:9, 9:16 or 1:1. 4.0 adds 21:9 and Auto | Six shapes listed in the API, 9:16 at 720 x 1280; check the app panel before a vertical from text alone |
| Sound | Made with every clip; write "Ambient sound only" to keep music out | Native audio on or off, with speech in several languages | No audio through the API; in the web app, check for an audio toggle, or build sound in the edit |
| Start and end frames | First frame, or first and last | First and last; 4.0 also takes up to 10 keyframes | First frame only |
| How to write the prompt | Prose in order: shot, subject, action, camera, light, then sound | From a start frame, motion and sound only; Multi-Shot off for one unbroken take | Motion only, as Runway advises for image to video |
| Saying what you do not want | Plain nouns in a negative field where the surface has one, never "no" or "don't" | One prompt field takes positive and negative descriptions | Positive phrasing only; Runway warns negatives can give the opposite result |
| List price | $0.40 a second at 720p or 1080p; Fast $0.12 at 1080p (Gemini API) | 3.0 in the Kling app: 12 credits a second at 1080p with audio, 8 without | $0.12 a second through the API; app credits vary by plan |
Figures from the makers' own pages, checked 17 and 18 October 2026; your settings panel wins when it disagrees. Sora is left out because OpenAI shut down its video models and API on 24 September 2026.
What to do it with
- ClaudeFree tier, paid above it
A language model that writes, edits, reads long documents and runs tools, including agents that work inside a codebase.
Bad at It is agreeable by default, so it will improve a bad idea instead of telling you it is bad. Ask it to argue against you and the quality of the answer changes.
- Claude CodeSubscription
An agent that reads a whole repository, edits files, runs the tests and explains what it changed.
Bad at It is fast enough to make a large mistake quickly. Work on a branch, read the diff, and never let it push. Watch the starting mode: on Pro, Max and Team plans it opens in auto mode, which Anthropic's own list says may read a .env file and send the key in it to the API that key belongs to. Switch it off.
- VeoPay per use
Text or image to video, with sound, at a length and coherence nothing had two years ago.
Bad at Cost per second is high enough to change how you storyboard. Long shots still drift, and hands and logos wander. Through the Gemini API on 25 September 2026 it refused a last frame on every tier, though the docs describe one.
- Google FlowFree tier, paid above it
A web studio for making and extending clips with Veo 3.1 and Gemini Omni, paid for in credits from a Google AI plan.
Bad at Credits do not roll over, features differ from the API (Quality takes no references), and Google's own help pages disagree on who can use it.
- Gemini Omni FlashPay per use
Video generation and conversational editing from text, images, audio and video, about $0.10 a second at 720p; Google's recommended default for new video work.
Bad at Makes several shots by default unless told otherwise, 1080p and 4K are upscales, no negative-prompt parameter, and uploaded-video editing is unavailable in the EEA, Switzerland and the UK.
- KlingFree tier, paid above it
Image-to-video with unusually stable human motion.
Bad at Queue times vary by a lot, and the interface assumes you already know what its settings mean.
- RunwayFree tier, paid above it
A video generation and editing suite with motion brush, camera control and per-frame steering.
Bad at Credit burn is fast and the good controls sit on the higher tiers. Output length still suits cuts rather than scenes.
- Luma Dream MachineFree tier, paid above it
Fast image-to-video with good camera language and a keyframe mode that interpolates between two stills.
Bad at Detail softens as the shot runs. Product labels are usually unreadable by the last second.
- HailuoFree tier, paid above it
Video generation with strong character motion for the price.
Bad at Prompt handling in English is weaker than in Chinese, and the terms of service are worth reading before commercial use.
- MiniMax H3Open weights or open source
A 33B open-weight model that makes 4 to 15 second clips with stereo sound from text, images, video and audio references.
Bad at Licence excludes the EU, UK, South Korea and USA; the 2K stage and the prompt-rewriting system stay on MiniMax's paid API; about 124 GB of weights in bfloat16.
- Wan 2.2Open weights or open source
Open video models under Apache 2.0: a 5B text-and-image-to-video model at 720p and two 14B mixture-of-experts models.
Bad at No sound. The 14B models want 80 GB on one card in Alibaba's own code, and nothing newer than 2.2 has been released as open weights.
- MidjourneySubscription
Generates images from a written description, with the strongest default aesthetic of anything in the category.
Bad at That default aesthetic is the problem as often as it is the draw. It has a house look, it pulls everything towards it, and getting a plain honest photograph of a real product out of it is harder than it should be. Text inside an image is unreliable.
- Adobe FireflyFree tier, paid above it
Adobe's generative models, trained on licensed and public-domain material and wired into Photoshop and Illustrator.
Bad at Raw image quality trails the leaders. What you are buying is the indemnity and the fact that it sits inside files you already have.
Why it matters
Ask a chat for a video prompt and you get one long paragraph of mood words, written for no tool in particular, asking for three camera moves in five seconds and a label the model will misspell. The take comes back warped and the fix is more adjectives. This skill makes Claude plan shots that fit your tool's clip length, bring the real product in from a photo, write each shot in your tool's terms, and change one thing per retake.
- Type
- technique
- Period
- ongoing