Claude Code skill that makes launch videos, ads and music videos from one line: idea, script, visuals, voice, music, edit and review, then a review room where your notes make the next take
Concept check: before anything is built
A storyboard costs minutes. A film costs hours. This check sits between them, and it is the cheapest place to catch a film nobody wants to watch.
Why it exists. Two launch films for one tool were built unattended from a delegated brief on the same day, by two separate runs. Both passed the render gate, the text check and the audio meter. The owner watched each once. Of the first: "it doesn't say anything". Of the second: "almost nothing". Of both: "the storyboard or the concept isn't that good and feels like slop". One had taken two hours. Nothing in the process had asked, before building, whether the idea was worth a film.
Run it on the leading idea before the pitch, and again whenever the storyboard changes materially. It applies in full to launch, product, feature and brand films. For a story film or a music video, use steps 1, 3, 5 and 6.
1. Retell it
Write the film as a viewer would describe it to a friend, in three sentences at most: it opens on, then, it ends with. No shot numbers, no technique. If you cannot, there is no film yet, only a treatment.
2. The three answers
After one viewing, muted, could a stranger say:
- what it is,
- what it does for them,
- what to do next?
Write the answer the film gives to each, and the shot that gives it. An answer that only the caption or the end card gives does not count for the second one: what it does must be seen happening.
For the third, use the real next step. If no link, handle or date exists yet, do not invent one: design the place where it goes, say on delivery that it is missing, and still end on a plain instruction the viewer can act on.
Steps 3 to 5 are the slop test. Slop is work the viewer has already seen, that could be about any product, and that nobody chose.
3. Words off
Describe each shot with its words removed. If what is left is a plain ground, the idea in that shot is a caption, not a picture. More than about a third of the runtime as words on a plain ground is a slideshow. Rework the storyboard so the product, a person or an event carries the meaning and the words confirm it.
4. Defaults that have already failed
Each of these was built as a product launch, finished cleanly and rejected. Some are real story shapes with their own uses (formats.md); what failed was using them to introduce a product to people who might want it. An unattended run drifts toward them because they are safe and easy to make truthful. For an introduction, treat a storyboard that rests on one as a failed concept and choose again, unless the user asked for it by name.
| Default | What it looks like | Why it fails |
|---|---|---|
| The film about itself | its own prompt, storyboard or timeline on screen; "you're watching it"; the making-of as the payoff | Two independent runs landed on this same payoff. It is a wink at the maker. The viewer learns nothing they can use. |
| Samples doing the work | existing sample or portfolio footage as most of the product proof | The viewer sees a showreel assembled, not the product working. Hand-directed samples are not a first-run result, however they are labelled. |
| The process tour | the pipeline's steps as beats: storyboard, animation, music, review | It describes how the maker works. Viewers want the result. |
| The genre joke | mocking other launch films instead of showing this product | It spends the runtime on other people's work and still has to end on a name and a line. |
| The typed prompt on black | the opening is a request being typed on an empty ground | Every tool of this kind opens this way. It is also type on a plain ground (step 3). |
Result or process: decide it, do not leave it to the user
For a product film, decide how much of how the product works belongs on screen, and say so in the storyboard. The default: the result first and on screen throughout. What makes this product different from the others in its field is then shown as features, one per beat, each seen happening to the hero result itself (a shot sent back and returned better), never as cards, lists or captions. A tool with a real difference in how it works (a director's judgement, a review step, a team of roles) needs those beats, or its position is only a slogan. The moment the result leaves the screen to explain the process, it has become a tour, which has already failed (launch-film.md, Rules).
5. The one image
Name the single frame a viewer would describe to someone else. If it is a line of text, a logo or a grid of thumbnails, keep looking. This frame usually belongs in the first seconds, not at the end.
6. A cold read
This read protects the build, the expensive step. It must not delay the user's first look: when someone is waiting, show them the ways and the storyboard first and run this while they read or straight after they pick. One test run made a user wait half an hour for a first storyboard because two outside reads were placed in front of them.
Makers grade their own storyboards about two points above outside readers. So before building, give the storyboard to a fresh reader who has seen none of the discussion: the retelling and the shot table only, with no rationale and no grade of yours.
Use two fresh readers each round. Give each one line on who will watch it and where. Ask, in this order:
- What is this for, and what does it do?
- Would you watch to the end? Where would you stop?
- What here have you seen before?
- A score from 1 to 10 on each of the three answers (step 2), and overall. Give them the scale: 10, I would send this to someone; 7, good, I would watch it; 5, fine and forgettable.
- The one change that would raise your score most.
- What should not change.
The bar on paper is 8 overall from both readers. It was first set at 9, because an owner asked for a nine and a maker's own nine is not evidence. Then the same kind of reader was given three of the best-regarded launch films ever made, written up the same way with the names removed: the highest overall score was 8, and about 7 where the reader said they would not have known the film. So a 9 is not reachable on paper, and 8 is a high bar; the owner settled on it. The nine is judged later, on the picture: the decisive test shot, then the export. Readers hold points back for what the page does not prove, and their overall score follows "do I want the thing" more than "is this a great film". If a reader cannot answer the first question, or calls the idea generic, change the idea, not the wording. Keep what they say is strong. Revise and read again with two new readers, four rounds at most. If two rounds in a row end on the same overall grades, stop revising this idea: it has reached its ceiling. Widen the search instead: go back to the ways you scored and try a different one (see "Where a stronger idea usually is" below). And when the readers' doubt is whether the picture is as good as the words, no round of text will settle it: build the one decisive shot and judge that. Then show the person the best version with its real grades, round by round. Never report a grade the readers did not give. If it is still under the bar, say what is holding it down and do not build it unattended.
When no fresh reader is available, do the read yourself after a break from the idea, say that it was a self-read, and be harder on it.
7. Who says yes
- A person is there. First, within a few minutes: the goal as you understand it and the scored ways with the first choice. End with their turn: go, pick another, or say what to change. Wait for it. Then, for the way they chose: the retelling and the storyboard, and the cold reader's answers as soon as they are in. This is the default, because a person who can see the ideas judges them faster and better than any check here.
- They say it is close. Fold their note in and run. Do not come back with a new round of options.
- They delegated and left ("I trust you", "finish it while I'm away"). The cold read stands in for the yes. Record its answers in
DIRECTION.mdand continue.
Delegation moves who says yes. It never removes this check.
Where a stronger idea usually is
- The viewer's moment. When does this person need the product? Open inside that moment, not on the tool.
- The strongest real result, first. Make the best thing the product can make for this film, and build the film around showing it.
- Range. Several very different results, each made for this film and each on screen long enough to read as a real one.
- One change, seen. A single visible before and after, in one continuous shot, beats three explained features.