Cinematic AI Ads
Claude walks you through a scripted AI ad the way a director would, stage by stage, and hands you prompts you can paste straight into the model.
What the flow actually looks like
9 stages. Each one makes the thing the next one needs, and every prompt comes labelled with the model it goes into.
- 01
Setup
YouYour product and who buys it
ClaudeSix questions, asked before a word is written
- 02
Script
YouYour answers
ClaudeHook, pain, and dialogue built to be spoken
- 03
Tone
YouPick one of eight
ClaudeCamera, lens, film stock, lighting, grade
- 04
Characters
YouA face, or nothing
ClaudeCharacter sheet prompts, era matched to the tone
- 05
Product
YouOne product photo
ClaudeA reference sheet that keeps the label readable
- 06
Location
YouThe character sheets
ClaudeBase image, wide shot, location sheet, new angles
- 07
Coverage
YouThe base image
ClaudeOver the shoulder, reverse and profile stills
- 08
Video
YouAll of the above
ClaudeA shot list split across Kling and Seedance
- 09
Post
YouYour renders
ClaudeThe grade, the grain, and where the cuts hide
Made with it
More of the same pipeline. No camera, no crew, no location.
The aerobics class
80s American, warm push, heavy halation
A room full of people who all have to stay the same person for the whole shot. This is the case that breaks most AI video.
The garage
Night exterior, practical light only
Dark scenes are where AI video usually falls apart into noise. The tone block names the stock and the lighting so it holds.
What is in the download
The pipeline
Nine stages in order, each one producing the asset the next stage needs.
Eight visual tone blocks
Copy-paste blocks for IMAX, A24, 60s Italian, 80s American, 80s summer comedy, Hong Kong studio, Wong Kar-Wai and 90s American. Every camera, lens, film stock and cinematographer checked against ARRI, Kodak, Panavision and the ASC.
Character realism
How to build a character sheet that survives twenty generations, and how to era-adapt an existing face without losing it.
Model routing
Which shot goes to which model and why. Faces go one way, movement goes another.
Video prompting
Shot list format, prompt structure, and the lines you must include or the model adds music and subtitles you cannot remove.
Continuity and post
Keeping a scene consistent across separate generations, then the colour grade and grain pass that sells it as film.
Buy this if
- You can already generate images and video, and your results still look like AI
- You want scripted ads with characters, not single pretty frames
- You have Claude and know how to install a skill
- You would rather learn the pipeline than pay per finished video
Do not buy this if
- You have never generated an image before. Start with the free prompts instead
- You want one image, not a whole ad
- You want the ad made for you. That is the service, not this
The problem this solves
Most AI ads fail in the same place. The prompt is fine. The character changes face between shots, the room rearranges itself, and the whole thing reads as AI in the first second.
That is not a prompting problem. It is a preparation problem. Output realism is set by input realism: the character sheet, the base image and the location sheet decide how every clip looks. Build those right and the prompts downstream behave. Skip them and no prompt rescues the shot.
This skill is that preparation, written down as a pipeline Claude runs with you.
What makes the tone blocks worth the price
A visual tone is not a filter. It is an engineered prompt block naming a real camera body, a real lens, a real film stock, a real lighting approach and a real cinematographer. It speaks to the model in the language of its training data, which is millions of real films, so the generation inherits a real production’s look instead of a generic AI one.
Getting them wrong is easy and invisible. A lens that never covered that sensor, a camera that did not exist in that decade, a cinematographer who never shot in that style: the model reads the contradiction and gives you mush.
Every block here was checked. The 80s American drama block names the Panavision Panaflex with C Series anamorphics, which is the package Vilmos Zsigmond actually shot Blow Out on. The Italian block names Eastman Color Negative 5254 with a Technicolor dye-transfer print, because by the mid-60s Technicolor was a print process and not a film stock. The Hong Kong studio tone and the Wong Kar-Wai tone have different references, because they are different traditions.
What it is not
Not a course. There are no videos and no modules. It is working material that Claude executes.
Not a prompt pack. The prompts it produces are written for your product, not copied from a list.
Not a shortcut past learning. It will make you faster at something you already partly know how to do. If you have never generated an image, the free prompts are the right place to start, and they cost nothing.
Questions
Do I need a Claude subscription?
Yes. This is a Claude skill. It installs into Claude and runs there.
Which generation platforms do I need?
Whatever carries the models it routes to. The platform comparisons cover what each one costs and which models it carries, so you can pick one rather than subscribe to five.
Do I get updates?
Yes. Model routing is the part that ages fastest, and updates are included.
Can I use the output commercially?
The skill is yours to use on your own and your clients’ work. You cannot resell or redistribute the skill itself. Whether a given generation is licensed for commercial use depends on the platform you generated it on, not on this skill, and the comparisons say which plans include a commercial licence.
Cinematic AI Ads
11 files, 1,065 lines. Instant download. Updates included.