Scenario/ explorations Create yours

A technical demo with Scenario models

Green Screen to Any Scene

Film a subject once, on a green floor. Then place it in any world, and make it look like it was really filmed there.

2 min 29 s, with narration

The whole process, in one film

Two subjects, four steps, three video models side by side, and the winner in four more worlds.

The test

Two subjects, one green floor

A chrome car, the hardest surface for reflections. And an astronaut, with a face behind glass.

Chrome carthe hardest surface for reflections
Astronauta face behind glass

Step 1 of 4

The green shot

Two stills set where the shot starts and ends. A video model animates the move between them, on a real floor with a grid, so the ground and the shadows stay readable.

GPT Image 2.5 Sunburst

First frame: the empty green studio
First framethe empty green studio
Last frame: the final pose
Last framethe final pose

Seedance 2.5

The movefirst and last frame to video

Step 2 of 4

Remove the green

A background removal model keeps only the subject, on a transparent layer. Shown here on black, with nothing behind it yet.

Green shotwhat goes in
CutoutPixelcut Video Background Removal

Three tries, one kept

Three removal models on the same car clip, frozen at the moments that tell them apart.

  1. The green shot at one second: the front of the car is already in frame

    Source at 1.0 s

    The car is already in

  2. Try 1 at one second: an empty black frame, the car is missing

    Try 1 · Video Foreground Extractor, Matting

    Empty: the car only appears at 1.4 s

  3. Try 2: the alpha mask shows thin lines from the floor tape leaking into the cutout

    Try 2 · Video Foreground Extractor, Heavy

    Alpha mask: a line from the floor tape leaks in

  4. Try 3 at one second: the front of the car is cut out cleanly

    Try 3 · Pixelcut Video Background Removal, kept

    Clean from the first frame

Step 3 of 4

The composite

Video Studio lays the cutout over a new background, also made with GPT Image 2.5 Sunburst. On its own, it looks pasted.

Cutout

Background · GPT Image 2.5 Sunburst

The empty background plate

Video Studio · cutout layer over the background

Looks pasted: no shadow · studio reflections

Step 4 of 4

The integration

A video edit model has to make the subject belong. Three models got the same input and the same short instruction, one pass each.

The same instruction for every model

Integrate the car into the scene so it looks like it was filmed there. Keep everything else unchanged.

FLUX.3 Edit Video Input · composite

Input
Seedance 2.5
MiniMax H3
FLUX.3 Edit Video

The verdict

FLUX.3 Edit Video won on both subjects.

It was also the least expensive of the three. From most to least expensive: Seedance 2.5, MiniMax H3, FLUX.3.

The winner, in more worlds

One shot. Six places.

The same green shots, integrated by FLUX.3 Edit Video. The paint, the glass and the suit follow the light of each place.

Red hangarastronaut
Cold corridorastronaut
Alien planetastronaut
Desert sunsetcar
Frozen lakecar
Neon streetcar

The pipeline

One pipeline, all in Scenario

Every step ran on Scenario, driven by Claude Code through the Scenario MCP.

  1. GPT Image 2.5 Sunburstframes and backgrounds
  2. Seedance 2.5motion on green
  3. Pixelcut Background Removalthe cutout
  4. Video Studiothe composite
  5. FLUX.3 Edit Videothe light

Film it once. Place it anywhere.

Take the recipe with you: the illustrated making-of, and an agent skill that runs the same four steps on your own subject through the Scenario MCP.

Voice: Gemini 3.8 Flash TTS. Music: ElevenLabs Music v2.5. All on Scenario.