ScenarioExplorations Block or Build
Create yours

A Scenario exploration

Block or Build

Build the shapes. Block the motion.

Lay out the shot in Blender, and a video model paints the final picture over it. Six tests show which stand-in each thing needs.

ForGame, marketing and animation teams who lay out a shot in Blender and let a video model paint the final picture over the playblast.

What was madeSix A/B tests where only the stand-in changes, plus The Last Lantern and a coral reef film: six examples, a narrated explainer, a guide, an illustrated making-of and an agent skill.

What Scenario didConcepts and finished stills with GPT Image 2.5 Sunburst, 3D placeholders with Tripo P2, rigs with Tripo Rigging, every take with Seedance 2.5, narration and score with ElevenLabs.

01How it works

Blockout to render, in three steps.

Teams lay out the shot in Blender and let a video model paint over it. This page is about step 1: what you put in the blockout.

  1. What this page is about
    1

    The blockout

    Stand-ins in Blender set where everything is, how big it is and how it moves. Its video is the playblast.

  2. 2

    One finished still

    A picture of the first frame, made from the blockout, sets the look: materials, light, costumes.

  3. 3

    The video

    A video model, here Seedance 2.5, takes the layout and the motion from the playblast (step 1) and the look from the still (step 2).

02Step 1: the stand-ins

Each thing in the blockout gets a stand-in.

Two kinds, often mixed in one shot.

Block

A capsule, a box, a few primitives.Made in minutes. It marks where something is, its size and how it moves, and leaves the rest to the model.

Build

A 3D model, posed or rigged.Takes longer: a concept image, a 3D generation, sometimes a rig. It shows the real shape.

Hybrid

Both in one shot.The set and some props built, the characters and other elements blocked.

03The rule

The model copies the blockout: its shapes and its motion.

A copied shape is what you want. A copied motion is only as good as its keys: a walk sped up stays a walk. Give a character no legs and the model invents the gait from what it has seen.

Shapes

Build what must look exact

Products, props the art direction names, set pieces seen close. A block only gives the model a size to fill.

Temple steps with no braziers
As blocks: the braziers vanished.
Temple steps with two bronze braziers
Modelled: bronze, as designed.

The temple at 7.5 s. Far set pieces and filler can stay blocks: the still gives them their look. See the temple

Motion

Block what must move naturally

People and animals. The model copies a rig's motion as it is. A legless capsule leaves the gait to the model, which has seen wolves gallop.

A wolf running with a long reaching stride on a cliff path
From a capsule: it runs.
The same wolf walking with short steps
From a rigged walk: it walks.

The wolf at 2.0 s. One exception: a clean cycle of exactly the motion you want is copied well. See the run

Frozen legs: she slid.A rigged walk: her feet slid.A legless capsule: she walked.The Last Lantern, one shot, three stand-ins for the traveller.

Hybrid

Most shots have both: build the set and some props, block the characters and other elements.

04Step 1, done well

Six habits for a good blockout.

The video model sees only the playblast and the still, and both come from the blockout.

  1. Block characters and creatures.A capsule at true height, a head, two or three features. No legs. Rig one only with a clean cycle of the exact motion. See the run and the wolf
  2. Move it along its path, nothing more.Position and heading only. The model adds the steps. See The Last Lantern
  3. Build what must look exact.Products, named props, set pieces seen close or from behind. See the temple
  4. Use shapes and colors that cannot be misread.A disc under a turtle became a fifth flipper; a yellow lamp post became a second lantern. See the reef
  5. Keep real sizes.Meters in Blender, a capsule as tall as the character. The model keeps where things are and how big they are.
  6. Make the still from the blockout's first frame.Same framing, so the look lands on the right shapes.

05The evidence

Six examples, read the same way.

Each is a fair test: same camera, same finished still, same prompt. Only the stand-in changes, with the one sentence of the prompt that describes it.

  1. 1The shotWhat has to work.
  2. 2The stand-insOne card per option.
  3. 3The resultOver its blockout: drag the line.
  4. 4Pros and consWhat it gave, what it cost.
  5. 5TakeawayWhen to use which.

Group 1 of 4

Objects

Props and products. The finished still usually shows them already. Build when every side must match.

Example 1ObjectsMarketing, product film

The sneaker

A sneaker drops onto a plinth and the camera circles it. The shoe must look like the product from every side.

TakeawayBoth gave the shoe, because the still shows it. Build it when the back and the sole must match the real product.

Jump to

BlockThree boxes and a wedge, two colors

The shoe from the still, from every angle, heel included.

  • Made in minutes
  • The still carries the design; no box edge prints through
  • What the still does not show is the model's guess: here, the heel

BuildA generated 3D shoe, same two colors

The same shoe, the same heel. No visible gain over the block.

  • Every side fixed, heel included
  • Free if you have the CAD file or the game asset
  • A concept and a 3D generation first
Setup the still, the sentence that changed, the models
The finished still, @image1, shared by both versions.
The only sentence that changed

In @video1 the sneaker is a rough block stand-in, a few boxes and a wedge in the product's two colors: it only marks where the shoe is, its size, its angle and how it moves.

In @video1 the sneaker is a rough 3D placeholder model painted in the product's two flat colors: it only marks where the shoe is, its size, its angle and how it moves.

Made with
GPT Image 2.5 Sunburst concept and still, Tripo P2 model, Seedance 2.5 takes finished with Seedance 2.5 Draft Complete. One take per version, 6 s, the same real-time drop and orbit in both.

Group 2 of 4

Characters

People and animals. Their look comes from the still; the stand-in decides their motion. Block them by default: a rig's motion is copied as it is.

Example 2CharactersGames, companion

The run and the wolf

An explorer sprints along a cliff path with her wolf, and they leap a log. The wolf has to gallop. Only the wolf changes: the explorer is the same red capsule in both versions.

TakeawayBlock them. The legless wolf galloped; the rigged wolf, its walk sped up, came back as a trot. Rig a character only for a clean cycle of exactly the motion you want, like the run below.

Jump to

BlockA legless capsule wolf: head, ears, tail

A full gallop and a clean leap, in both shots.

  • Four legs at a full gallop, and the leap
  • Made and keyed in minutes
  • The model picks the gait

BuildA rigged wolf, its walk sped up

A trot. In the second shot it comes back upright: the walk cycle shows through.

  • The wolf's exact shape
  • The walk was copied: a trot, not a gallop
  • A concept, a 3D model and a rig first
Setup the still, the sentence that changed, the models
The finished still, @image1, shared by both versions.
The only sentence that changed

In @video1 the wolf is a blue-gray stand-in, a horizontal capsule body with a ball head, a snout, ears and a tail and no legs: it only marks where the wolf is, its size, its pace, its heading and its leap.

In @video1 the wolf is a rough blue-gray 3D model of the wolf: it only marks where the wolf is, its size, its pace, its heading and its leap.

Made with
GPT Image 2.5 Sunburst concept and still, Tripo P2 model, Tripo Rigging 2.5 rig and walk, Seedance 2.5 takes finished with Seedance 2.5 Draft Complete. One take per version, 6 s. The explorer is the same red featured capsule in both.

Close on the legs

The same cliff path with the camera at leg height for 5 seconds, the wolf and the explorer filmed one at a time. Two drafts per version, one shown.

The wolf

A rigged walk played faster stays a walk.

Jump to

BlockA legless capsule: head, ears, tail

A loping run: long reaching strides, dust from the paws.

  • Four clean legs, a natural run
  • Made in minutes
  • The model picks the gait: one of two drafts trotted

BuildA rigged wolf, its walk sped up

A fast walk, tail low, although the prompt asks for a gallop.

  • The wolf's exact shape
  • The walk was copied: short steps, tail low
  • No run cycle for the rig
The run: the exception

A clean run cycle is copied as it is, and here it looked more athletic.

Jump to

BlockA capsule at her height, no legs

She runs, clean legs, arms swinging bent: the model's own gait, more upright.

  • A natural run, boots plant, no sliding
  • Made in minutes
  • More upright, a shorter stride

BuildA rigged girl playing a run cycle

She runs, clean legs: the rig's cycle, copied as it is.

  • An athletic run: lean, bent arms, long stride
  • A concept, a 3D model and a rig first
  • Only as good as the cycle
Setup the stills, the sentences that changed, the models
The finished still, @image1, shared by both versions.
The wolf

In @video1 the wolf is a plain blue-gray stand-in shape with no legs: it only marks where the wolf is, its size and its direction; replace it with the real wolf from @image1.

In @video1 the wolf is a plain blue-gray 3D model of the same wolf; replace it with the real wolf from @image1.

The wolf's gait, in both prompts
It runs in a loose, natural gallop: front legs reaching far forward, hind legs gathering under the body, the back flexing with every stride, paws striking the dirt and never sliding.
Made with
GPT Image 2.5 Sunburst still, Tripo P2 model, Tripo Rigging 2.5 quadruped rig and walk (it has no run), Seedance 2.5 takes finished with Seedance 2.5 Draft Complete. Two drafts per version, one shown, 5 s. A wolf-shaped legless capsule ran no better.
The finished still, @image1, shared by both versions.
The run

In @video1 the girl is a plain red stand-in shape with no legs: it only marks where she is, her height and her direction; replace it with the real girl from @image1.

In @video1 the girl is a plain red 3D model of the same girl; replace it with the real girl from @image1.

The run's gait, in both prompts
She runs naturally at a steady pace: knees driving forward, feet striking the dirt under her hips and pushing off, never sliding, arms swinging bent at the elbow.
Made with
GPT Image 2.5 Sunburst still, Tripo P2 model, Tripo Rigging 1.0 biped rig and run, Seedance 2.5 takes finished with Seedance 2.5 Draft Complete. Two drafts per version, the same gait in both; one shown, 5 s.

Example 3CharactersAnimation, acting beat

The gesture

A girl brings her grandfather a toy boat; he kneels to take it. One hand-off, a locked camera.

TakeawayBlock it. The capsules gave the same kneel and hand-off as the posed mannequins.

Jump to

BlockTwo capsules with arms

One hand-off, he stays on one knee. The camera drifts closer at the end.

  • A natural kneel and hand-off
  • Set up in minutes
  • The camera drifts closer at the end

BuildTwo mannequins, posed on keys

One hand-off, he stays on one knee. The framing of the blockout holds to the end.

  • The wide framing holds
  • Every pose keyed by hand
  • No better gesture for the work
Setup the still, the sentence that changed, the models
The finished still, @image1, shared by both versions.
The only sentence that changed

In @video1 the girl and her grandfather are colored capsule stand-ins with ball heads and simple arms: red is the girl, blue is the grandfather. They only mark where each one is, their height, their timing and roughly where their arms go. The blue capsule is short because he kneels on one knee the whole time.

In @video1 the girl and her grandfather are gray jointed mannequins: the one with the red torso is the girl, the one with the blue torso is the grandfather, kneeling on one knee. They mark where each one is, their height, their timing and their poses.

Made with
GPT Image 2.5 Sunburst still, Seedance 2.5 takes finished with Seedance 2.5 Draft Complete. One take per version shown, 7 s; the block take is the second of two with identical inputs. The boat is the same four primitives in both, attached to the hands.

Group 3 of 4

Scenes

Sets and environments, big, mostly still and full of pieces. Build the layout the camera follows.

Example 4ScenesAnimation, nature film

The coral reef

A turtle swims down a reef channel; a manta crosses at the end. 12 s, in two looks from one blockout.

TakeawayBuild the set when its layout is the shot. One blockout, any look.

No A/B here: the whole scene was built. It shows what building a set gives, and what it costs.

BuildA reef placed piece by piece, a shaped turtle

Between the two looks only the three reference images and one sentence change. Both keep four flippers.

  • Channel and coral heads stay in place, in both looks
  • Four flippers in every frame
  • A big build: thousands of placed pieces
  • Every shape is read: an earlier disc under the shell became a fifth flipper

From the coral reef blockout, October 2026, finished for this page. GPT Image 2.5 Sunburst reference images, Seedance 2.5 takes finished with Seedance 2.5 Draft Complete. One take per look.

Group 4 of 4

Hybrid

Most shots: build the set and the props the art direction names, block the characters and other elements.

Example 5HybridGames, level reveal

The temple

The camera rises over a temple gate, then follows the hero to the shrine. The art direction names bronze braziers.

TakeawayGo hybrid: model the set and every prop the art direction names, keep the hero a capsule.

Jump to

BlockThe set in boxes, the hero a capsule

Right layout and light. The braziers vanished.

  • Layout and light hold
  • Built in minutes
  • The braziers vanished

HybridSet and props modelled, the hero a capsule

Every piece as designed, bronze braziers included.

  • Every piece as designed, bronze braziers included
  • The hero still walks freely
  • Seven models to make
  • 2 of 4 takes lost the hero's scale

The braziers at 7.5 s

Blockout: discs on posts
Blocks: gone
Modelled: bronze braziers
Braziers left as blocks: planters
Setup the still, the sentence that changed, the models
The finished still, @image1, shared by every version.
The only sentence that changed

In @video1 the temple is built from rough blocks that only mark where each piece is, how big it is and its color: the box arch with a teal slab on top is the carved stone gate, the long boxes are sandstone walls, the cones on brown posts are wind-sculpted pines, the box-and-ball pedestals are stone guardian lions, the teal pyramid roof on four columns is the shrine, the plain cylinders are broken stone pillars, the wide disks on posts are bronze braziers.

In @video1 the temple is built from rough 3D placeholder models of the gate, the walls, the pines, the guardian lions, the shrine, the broken pillars and the braziers, painted in flat colors: they mark where each piece is, how big it is and its shape.

Made with
GPT Image 2.5 Sunburst concepts and still, Tripo P2 models, Seedance 2.5 takes finished with Seedance 2.5 Draft Complete. One take per version shown, 8 s. The hero is the same red capsule in every version. Both takes shown are re-rolls with identical inputs: the first block take had a stiff cape, and the modelled version lost the hero's scale in two of its four takes.

Example 6HybridAnimation, short film

The Last Lantern

A 2-minute short in 40 cuts: a traveller carries a lantern through a forest.

TakeawayBuild the world the camera explores. Block whoever walks through it.

Hybrid43 modelled set pieces, the traveller a red capsule

12 s of moments from the film, each over the blockout it was made from.

  • Forest and lantern hold their design for 40 cuts
  • The capsule traveller walks naturally
  • 43 models to make and place
  • A yellow lamp post came back as a second lantern
Cut 21: the climb to the altar

Seen from the side at ground level, on the hill path up to the stone circle. The red capsule, with no arms and no legs, carries the modelled lantern at hand height.

HybridThe modelled lantern, the traveller a capsule

Seedance 2.5 gave her legs, boots and a steady climb; the lantern kept its design, in her hand.

  • A natural climb from a legless capsule
  • The lantern's design held
  • A lantern model to make first

One shot, three stand-ins for the traveller

From The Last Lantern, October 2026. GPT Image 2.5 Sunburst concepts and stills, Tripo 3.1 models, Seedance 2.5 takes. See the full exploration.

06Summary

Which stand-in, when.

BlockBuildHybrid
Use it forCharacters and creatures, filler, far setProducts, named props, set pieces seen closeMost shots
WhyNatural motion, made in minutesThe model copies the exact shapeAn exact world, natural motion
Watch outThe model invents the detailsMotion is copied as keyed: only as good as the cycleTwo kinds of stand-in to make
Seen inRun and wolf, gestureSneaker, reef, run (with a run cycle)Temple, The Last Lantern

07Your shot

Four questions, one stand-in.

Answer for each thing in your shot. The answer names the stand-in, how to set it up and the example it comes from.

1What is it?
2How does it move in the shot?
3Must one exact design appear?
4How close does the camera get?

Block

Capsule, no legs

A capsule at their height with two or three features that tell them apart: hair, a hood, a backpack. No legs: key the position and heading along the path, and describe the gait in one sentence of the prompt.

From example 2, the run and the wolf.

08Make yours

Block your own shot, then build what matters.

The skill gives your agent the stand-in rules, the prompt sentences and the A/B check. It runs through Scenario MCP.

Made with Scenario. Stills: GPT Image 2.5 Sunburst. 3D models: Tripo P2 (Tripo 3.1 in The Last Lantern). Rigs and motion: Tripo Rigging 2.5 (wolf), Tripo Rigging 1.0 (run). Video: Seedance 2.5, finished with Seedance 2.5 Draft Complete. Blockouts: Blender 5.2.

Create yours on Scenario