One Asset In, Twenty Scenes Out
It was built on magazine covers, and nothing in it is specific to magazines. One flat asset goes in — a cover, a candle, a bottle — and a catalogue of staged scenes comes back. No designer places any of them by hand.
Input1 file
Before
Staging a cover into a lifestyle scene is a composite: mask it, place it, light it, match the grain. A launch needs a set of them in two aspect ratios, and another set the next time the cover changes.
Outcome
A half-day task now takes minutes.
- staged scenes from one flat asset
- 10–20
- in production since the month Runway shipped Workflows
- 2024
What Came Out
The same cover in every one of them, and none of them composited by hand. The empty half of each frame is where the headline goes.
Output7 shown
One Asset, Four Stages
01 Input
- Magazine coverJPEG only. The one asset.(Source)
02 Prompt generation
- Scene JSONSize · angle · placement · framing · negative space(Process)
03 Image generation
- RunwayOne staged scene per prompt(Model)
04 Output
- Staged scenesOne render per scene prompt(Output)
- 9:16 + 16:9Both formats, every scene(Output)
- Copy spaceHeld for the headline and the CTA(Output)
The leverage is the schema, not the model: the JSON decides how a scene is staged before anything is generated. Every scene shares it, which is why twenty of them read as one campaign instead of twenty lucky renders.
Inside the Canvas
Art direction as a text file
One JSON node fixes the camera angle, where the magazine sits in frame, and how tight the crop is — before Runway sees the cover at all. That node is the control surface: change a field in it and every scene above changes with it, in one pass, without anyone touching an image. I built it the month Runway shipped Workflows, and it has run since.
Ledger
- Input
- One flat asset. JPEG only.
- Output
- 9:16 and 16:9, every scene.
- Where it runs
- Inside Runway, as a workflow. There is nothing to click; what it hands over is the wall of scenes above.
- Role
- Solo. Independent build.
- Stack
- Runway ML · AI Image Generation · Workflow Automation