Features / 3D Greybox Previz
EN 中
Open workspace
SEELE feature · 3D Greybox Previz

3D Greybox Previz — one prompt for camera-controlled CG films

Seedance 2.5 Available now
SEELE-02-Pro 3D Greybox Previz · frame-by-frame restyle
Try
Just one sentence Fastest

Just describe the shot. SEELE auto-builds the 3D greybox previz (camera and animation logic included), then restyles it into a film.

Upload a video

Already have a screen capture or footage? Restyle it into a new look — camera and composition fully preserved.

Upload your model

Build the CG animation yourself, control every shot precisely, then let us paint it.

One prompt vs eight manual steps

They run eight steps; you just say one sentence

“Build a 3D greybox previz to lock the camera first, then restyle the look” is the industry-accepted way to make controllable CG. The catch: it chains several paid tools and endless file shuffling. SEELE automates every step.

Pro pipeline8 steps · 3–4 tools
  1. 01Code a 3D scene in Claude / three.js
  2. 02Hand-author the camera motion
  3. 03Screen-record and export the greybox animation
  4. 04Upload to Runway to generate
  5. 05Extract the first frame
  6. 06Restyle the first frame with Magnific structure reference
  7. 07Drop it into Runway Restyle
  8. 08Tweak params and regenerate, over and over
SEELE1 step · one sentence
01
Say one sentence

SEELE auto-builds the 3D greybox previz, locks the camera, then restyles it frame by frame into a cinematic film.

  • No modeling
  • No screen recording
  • No tool chaining
  • No frame extraction or param tweaking
Before · After

Same shot, only the look changes

Drag the slider: on the left is SEELE’s auto-built 3D greybox previz (camera and space locked); on the right is the restyled film. Composition and camera are identical — only material, light and style change.

Final film · frame-by-frame restyle · Seedance
SEELE 3D greybox previz · camera & timing locked
3D greybox previz · auto-generated
Final FILM
One prompt: cyber station, rainy night → auto 3D greybox previz → 4s · 1080p film
Workflow

Four steps, from one sentence to a finished film

You just spell out what you want; the Agent handles both “building the skeleton” and “painting it.” No render-pipeline knowledge, no modeling required.

01

Describe the shot you want

Spell out subject · style · light · camera in one sentence. Or upload a video or your own model.

02

SEELE builds the 3D greybox previz

It auto-generates a 3D greybox previz that pins down camera, timing and spatial relations first — that’s the secret to control.

03

Restyle frame by frame

Locking the greybox’s composition and motion, it restyles frame by frame. Temporally consistent — no flicker, no drift.

04

Preview and export

Preview in the workspace, swap styles and rerun, then export the film — or move on to editing and scoring.

The greybox is a method, not a barrier. Camera and spatial relations are the hardest things to get right by rolling the dice. SEELE auto-builds a 3D greybox previz to pin them down, and the restyle only handles the look — you don’t need to build the greybox yourself, and you don’t need to know modeling.

Good fit

What it’s great at, and what it isn’t

Good fit
  • You want CG video where camera and composition are truly controllable, not left to chance
  • You have just a one-line idea and want SEELE to auto-build the greybox and produce a film
  • You already have a video or your own model and want to switch it to another style
  • You want to try multiple art styles on the same shot (cyber / hand-drawn / realistic…)
  • Fast output for game CG, trailers and level-mood previz
Not a fit
  • You just want a static poster with no camera motion (image-to-image fits better)
  • You upload a video but want to heavily change the camera motion or re-block shots during restyle (edit the greybox first)
  • You need frame-accurate lip-sync / dialogue timing
  • You require pixel-perfect reproduction of a real branded IP
Prompt assistant

Fill in four elements, and it assembles a solid prompt

Not sure how to write it? Fill in or pick subject / style / light / camera, and a usable prompt assembles live on the right — one click takes it to generate.

01Subject / sceneWhat’s in the shot
02StylePick or customize
03Light / mood
04CameraJust spell it out; the greybox locks it
Assembled · promptlive
Write a good prompt

A reliable prompt structure

Break the sentence into fixed parts and the output stays stable. The first four come from your intent; the last two are fidelity locks — keep them unchanged.

  1. 1Subject the core objects and scene in frame (people, buildings, vehicles, space)
  2. 2Style art direction (cyber / hand-drawn / realistic / low-poly…)
  3. 3Light time, weather, mood (neon rainy night / dusk backlight)
  4. 4Camera the camera motion you want (the greybox stage locks it)
  5. 5Fidelity lock keep composition & camera from source
  6. 6Quality 1080p · cinematic · temporal-consistent
station_platform · restyle.prompt
/* subject */ Train-station platform at dusk, a person by the tracks
/* style */ Cyberpunk, wet streets, neon-sign reflections
/* light */ Rainy night, cool-blue key + magenta neon accents
/* camera */ Slow dolly, pushing deep into the platform
--keep composition, camera-motion from source
--quality 1080p cinematic temporal-consistent
Style library

Click to append into your prompt

Not sure what words to use? Pick from below; a click appends to the top input — try several directions on the same shot in one go.

Art style

Light / mood

Camera

Example templates

Apply in one click, straight to generate

Pick the template closest to what you need; clicking jumps to the workspace with the full prompt, then just change it to your own shot.

CG mood

Cyber rainy-night platform

One prompt → neon rainy night · dolly · cinematic

Hand-drawn narrative

Ghibli morning village

One prompt → hand-drawn warm sun · mist · cozy

Trailer

Wasteland dusk highway

One prompt → realistic wasteland · backlight · tracking

Model comparison

Runway, Seedance and SEELE — where they differ

On the road to “controllable CG video,” Runway and Seedance are two major pieces. SEELE builds on their capabilities and automates the hardest step — building the greybox.

Dimension Runway Gen-4 Seedance 2.5 SEELE 3D Greybox Previz
Camera control Motion Brush + camera presets, tuned shot by shot Multimodal references + greybox previz, bring your own references One sentence auto-builds the greybox previz; camera locks automatically
3D greybox previz Not native; build it yourself in three.js / UE and screen-record Supported, but you supply the 3D structure + style reference Auto-generates the greybox, nothing to prepare
Temporal consistency Strong; mature v2v restyle Strong; more natural motion and physics Locked to the greybox; frame-by-frame with no drift
Learning curve Need to know Motion Brush / keyframes Need to prepare the greybox and references One sentence, zero barrier
Inputs Text / image / video Text / image / video / multi-reference One sentence / upload video / upload model

Comparison based on each vendor’s public materials (Runway Gen-4 and Seedance 2.5 released in 2026); capabilities keep evolving, so defer to official sources. SEELE’s goal isn’t to replace them, but to turn the controllable “greybox first, then restyle” approach into a single sentence.

FAQ

FAQ

3D greybox previz means first using a material-free, grey-block 3D animation to pin down camera motion, timing and spatial relations; the AI then only restyles the look. It’s the common industry method for controllable CG video, and tools like Seedance 2.5 have adopted the concept. SEELE builds that greybox for you automatically — no modeling skills needed.
Ordinary text-to-video does it in one shot, leaving camera and composition mostly to chance. SEELE first pins down composition, camera motion and spatial relations with a 3D greybox previz, then only restyles material, light and style. So camera and layout are fully controllable rather than rolled from scratch — that’s the essential difference.
It’s not either-or. “Build a 3D greybox previz to lock the camera, then restyle the look” is the industry-accepted controllable approach; with Runway you hand-brush the camera shot by shot, and with Seedance you prepare the greybox and references yourself. SEELE automates the whole flow into one sentence — it chases the same cinematic temporal consistency underneath, but you build nothing yourself.
No. The greybox is SEELE’s internal method, not a barrier. By default you only need one sentence and the system auto-builds a 3D greybox previz before producing the film. You can also upload a video, or upload your own CG animation / model for precise control.
Three: just one sentence, with SEELE auto-building the 3D greybox previz before producing the film; upload a video (engine capture, live footage or existing shots, .mp4 supported) to restyle directly; or upload your own CG animation / model for precise control before painting.
No limit. The 3D greybox previz is a reusable baseline; swap styles and rerun as often as you like, compare cyber / hand-drawn / realistic versions side by side, then deepen the best direction.

The greybox is ready — let’s paint it

One sentence turns a grey-box scene into the film in your head. Start here and take it all the way in one workspace.

Open the workspace and start restyling