A SCENARIO EXPLORATION · ONE SHOT · 0:10
LAST
ORDER
2 a.m., rain, and one ramen stand still open. The camera glides from a puddle to a steaming bowl while a neon sign sputters on.
One prompt. Claude Opus 5.5 built the whole shot alone in Blender in 55 minutes, with the Scenario Blender Plugin making the signs, posters, menu, materials, sky and two 3D props.
01 / THE SHOT
One camera move.
No cuts.
It starts 16 cm above a rippling puddle, glides past a humming drinks machine and a parked delivery scooter, tilts up as the big sign sputters on, then cranes down to the bowl. Jump to a beat below. Music added in the edit.
02 / THE SIGN
Eighteen keys
and a flicker.
GPT Image 2 drew the YOMICHI sign (夜道, “night road”). The agent turned it into the sign's glowing face, then wrote the sputter as one power value with eighteen hard keys, frames 139 to 157, driving the sign and its lights. Press the switch: it plays those keys at 24 frames a second.
- Frame
- 138
- Power
- 0.00
- 1390.7
- 1400
- 1430.35
- 1440
- 1461
- 1470.15
- 1480.8
- 1490
- 1531
- 1540.3
- 1550.95
- 1560.6
- 1571
Keys as written in the build script, constant interpolation (no fades between values). Two more keys hold it off from frame 1 and on to frame 240.
03 / ONE PROMPT
Nobody to ask.
Every call its own.
The agent ran headless on one Mac with Blender 5.2 and the Scenario Blender Plugin, with a 900 CU budget and no follow-ups. It invented every name in the alley and chose where Scenario beat procedural work.
“LAST ORDER. 2 a.m. in a narrow back alley of an invented, rain-soaked future city. A tiny ramen stand is the only thing still open. One continuous camera move: start low over a puddle where neon reflections shiver under falling rain, glide forward and rise slightly past a humming drinks vending machine and a parked delivery scooter, under tangled cables and hanging signs, and end on the counter where a fresh bowl steams under a warm lamp. At about 6 seconds the big vertical neon sign above the stand sputters, then locks on, and its light washes over the wet ground.”
The shot, from the prompt the session received
- 55 min
- one session, render included
- 15
- Scenario generations, none refused or retried
- 581
- credits (CU) of the 900 budget
- 12 min
- to render 240 frames in EEVEE
04 / WHAT SCENARIO MADE
Every word in the alley
was generated.
BAR ORBIT, KARAOKE 99, CAPSULE INN 7, KISSA LUNA, FIZZLINE, KAZE: all invented, all legible. GPT Image 2 drew the signage and print; each image became a glowing or printed plane in the scene.
Four wet materials
PATINA generated each full PBR set; the plugin applied them. The agent then built puddles into the asphalt from its height map, with rain rings that spread and fade on every frame.
The sky over the roofs
Skybox GPT made a 360° night sky, rain clouds lit magenta and cyan over a far skyline. It lights the whole alley and shows as the sliver of sky above it.
Two props, from image to 3D
A concept image first, then a textured mesh placed from its bounds. The bowl arrived with 1.9 million faces; the agent cut it to 12% after it slowed every frame. Turn them here.
Drag to turn · scroll or pinch to zoom
The meshes as Scenario delivered them, reduced for the web (textures at 1024 px, about 55,000 triangles each).
05 / BLOCKING TO FINAL
Same camera,
121 steps apart.
Every wall, cable, stool and light was written in Python. The agent kept one fixed wide camera for its own record: step 30 in flat blocking colours, step 151 with materials, neon and rain. Drag to compare.
Drag the divider, or focus it and use the arrow keys.
06 / THE REVIEW LOOP
It opened its test frames.
Then fixed them.
The agent rendered six moments of the shot, looked at each image and changed the scene before the final render. Three of those catches, test frame on the left.
- 9 s a frame at half size.The 22,000 rain instances were casting shadows. Shadows now stay on the key lights only; the rain was also cut to 14,000 thinner, fainter drops.
- 8 to 17 s a frame at full size.The 1.9M-face bowl was redrawn into every shadow map. A 12% decimation brought frames to about 1 to 2 s.
- The scene read too magenta.A colder blue sky glow and street bounce, so the warm stand stands out.
- The camera never looked at the sign.It now tilts up while the sign sputters, then cranes down to the bowl.
07 / THE BUILD
55 minutes
in ten seconds.
The agent saved a frame after every meaningful change, 151 in all, each stamped with the step it had just finished: “063 / Signage: GPT Image 2 neon Luna”.
Its own list of weak points
No rain splashes on the ground or the counter. A stand and a bowl light on detail. No sound in the shot itself: the neon buzz at 6 seconds, the rain and the machine's hum are the next pass.
08 / MAKE YOUR OWN
Take the pipeline.
Pick your street.
The skill packages this workflow for an agent in Blender with the Scenario Blender Plugin: plan one camera move, generate signage, materials, sky and props, record the timelapse, animate a light, fix what the test frames show, render.
scenario-blender-single-shot
SKILL.md, references for the shot plan, the wet ground and the test-frame fixes, and the tested scripts: timelapse stamping, the sputter, test frames, a resumable render and the encode.
Download the skill ↓ PDF · ILLUSTRATEDThe making-of
The brief, what Scenario made, the blocking, the review loop and what it changed, the render, and the receipt.
Read the making-of ↗ VIDEO · 2:21How it was made
Narrated: the brief, the stack, what Scenario made, the timelapse, the sign's eighteen keys, the review loop, then the shot.
Download the video ↓Your alley, your sign, your last order.
Make yours