The AI Filmmaking Pipeline

The pipeline, end to end

Before you touch a single prompt, hold the whole pipeline in your head. Every project we make follows the same four moves. They are not four buttons; they are a chain of handoffs. A stage is finished only when it leaves a concrete input the next stage can trust. This first lesson is the map; the rest of the course walks each leg of it.

The four stages

Four moves, one fixed order — watch the strip play through them: each clip is the real screen work of that stage, and its caption names the deliverable you walk away with. If you can't point at that deliverable, the stage isn't done.

Think — work with Claude until the brief and prompt say what the shot must be.

The order is the lesson. Skipping straight to "generate" is the single most common reason a project stalls — you end up re-prompting the same thing ten times instead of thinking once.

Follow the handoffs

The pipeline works because each receiving stage gets something more useful than an idea:

HandoffWhat crosses itWhy the next stage needs it
Brief → setupAn agreed brief shaped with Claude: the shot, world, cast, props, and constraintsSetup can create the right homes instead of guessing what the production will need.
Setup → generationA project, useful folders, and one naming contractEvery candidate has an address; approved work can become a reusable @loc_, @char_, or @prop_ asset instead of an orphaned file.
Generation → SeedanceProofed, named location, character, and prop elementsSeedance combines the actual source pixels. A defect handed over here will be multiplied in motion.
Seedance → reviewA motion test and a diagnosis, not just a videoReview can approve the shot or send one specific source back for correction.

Setup defines the address; generation earns it. A file does not become production input merely because it exists. First inspect it, then approve it, name it, and save it as an element. Build the location before judging the character on it: the place is both the foundation of the shot and the test bed that lets you see whether the character really holds up.

Proof is the gate

A model description and a confident prompt are promises. The pixels are proof. Before anything reaches Seedance, inspect the source still against the same quality bar every time:

Review closes the loop

Test is not a finish line with only pass or fail. It tells you where to return. If the geography melts, fix the location. If identity drifts, fix the character sheet. If the assets are sound but the action is wrong, revise the Seedance direction. If the whole shot answers the wrong idea, reopen the brief with Claude. Correct the earliest broken handoff, approve it again, and retest; polishing a downstream symptom only hides the source.

That is also the map of the course. Working in Claude and Turning thoughts into a prompt build the brief. Name your assets and Set up your first project make work retrievable. Choosing your image model and Proof, not promises establish evidence. Generating locations builds the world before Generating characters adds the cast. Test in Seedance and Spot the slop teach the review loop. The section tests and final exam check those junctions, and the capstone runs the whole chain with your own location, character, prop, and final Seedance shot.

Home

Community

Generate

Library

Profile