TAVOA AI 3D Jutsu: Build and Animate 3D Scenes from a Prompt

3D Jutsu is TAVOA AI's AI 3D workspace: a prompt becomes an editable scene with objects, lighting, cameras, and animation, then renders to video from the same chat.

3D Jutsu is TAVOA AI's AI-native 3D workspace. It turns a prompt or a reference into an editable 3D scene with objects, layout, lighting, cameras, and animation, and then into video. An AI agent running on Supercomputer assembles the scene, you refine real scene objects without regenerating everything, and the finished shot renders from the same chat.

What is 3D Jutsu?

Blocking is the stage where a scene takes shape before any final rendering: objects get placed, characters get positioned, cameras get framed. Previz turns that rough scene into a moving reference for the final shot. In classic 3D pipelines both stages are done in desktop 3D software; 3D Jutsu moves them into the browser.

3D Jutsu is how TAVOA AI handles scene blocking and previz. It is a standalone workspace within TAVOA AI's AI-native creative suite, connecting early scene planning with final video generation. Scenes live in your TAVOA AI account under My Scenes.

How do you start a scene?

A scene starts from a text prompt. You describe what you want to build, attach references if you have them (your uploads or past TAVOA AI generations), and set the parameters:

Model: the LLM that drives the agent, with Auto as the free default.

Duration: from 1 to 60 seconds of scene time, 15 seconds by default.

Aspect ratio: 16:9, 9:16, 1:1, 4:3, 3:4, or 21:9.

Press Generate and the agent assembles a draft of the scene in an interactive 3D viewport: drag to orbit, scroll to zoom, right-drag to pan.

What can you build and edit?

Everything the agent generates lands in the viewport as editable scene data, so you refine real scene objects instead of regenerating the whole scene. The editor covers the same ground a 3D artist expects from a blocking pass:

Tool

What it does

Scene generation

A prompt or a reference becomes a full scene: objects, layout, and light placed as editable geometry

Add object

Primitives (sphere, cube, pyramid), three light types (sun, point, spot), and cameras added manually

Asset library

Curated GLB assets and Mixamo characters drop directly into the active scene

Scene properties

Background color, scene resolution, and light intensity controlled from a side panel, with the full object hierarchy visible

Animation

A timeline with a playhead; motion comes from a camera trajectory, an animated character, or a GLB file imported with its animation

Camera work

Camera presets and trajectory editing frame the shot; Esc leaves a camera view back to the free viewport

Revision history

Every editing session saves as a revision; preview a checkpoint or restore it as a new head

Export

A finished scene renders as an animated MP4, a static frame, or a GLB file with dimensions and file size set before render

After generation, the scene stays editable by hand: objects move, rotate, scale, and duplicate, lights swap, cameras reframe. Undo, redo, and revision history apply to every change.

How does the AI agent work?

The chat panel inside the workspace takes instructions in plain language, for example "move the car closer to the streetlamp, add a second character across the street". Two behavior modes control how much autonomy it gets: in "generate without asking" the agent runs without confirmation, in "ask before generating" it checks with you before each generation. A separate ask mode answers questions without touching the scene.

The agent runs on Supercomputer, so you choose which LLM drives it. Auto is the default and free. The selector also carries specific models, including GPT 6 Astra, Claude Fable 5.1, Claude Opus 5, GPT 5.6 Sol, and GLM-5.3 Flash, switchable per chat.

The same chat sends the result to video generation: when the scene is ready, you request the render directly from the conversation.

Scenes are built for teamwork. Share invites people by email as an admin, a collaborator, or a viewer, and team access ranges from invited people only to anyone with the link.

How does the full workflow look?

Here's how a finished shot comes together. The example: a rapper in a moving subway car, filmed by a robotic-arm camera move.

Step 1: Block the previz. A prompt to the agent builds the scene in grey proxies: the subway car, the character, and the camera path with its framing and timing.

Step 2: Render the previz. Press Export and the camera timeline renders as a video: the grey blockout becomes a reference clip for the intended camera move and the character's behavior.

Step 4: Combine in the same chat. The character and the location drop into the same 3D Jutsu conversation next to the previz render, and one message asks the agent to put them together. The previz provides a reference for the intended camera path, framing, and character behavior, and grey proxies become the real location, wardrobe, and performance.

What does it cost?

3D Jutsu prices agent messages by model. On Auto, the free default, messages cost nothing; picking a specific LLM charges credits per message. The selector carries 30+ LLMs across price ranges, the estimated cost shows before you send, longer chats use more, and the credits are the same balance you use everywhere on TAVOA AI.

Typical message cost by model:

Model

Approximate cost per message

Auto

Free

DeepSeek V4 Flash

~0.4 credits

Seed 2.1 Turbo

~1 credit

Gemini 3.5 Flash

~2 credits

Gemini 3.1 Pro

~3 credits

GPT 5.6 Sol

~7 credits

Claude Opus 5

~8 credits

Claude Fable 5.1

~11 credits

GPT 6 Astra

~17 credits

The cost of the engine depends on the model and its version: the selector carries models across price ranges, and any of them can drive the agent.

When can creators use it?

A blockout settles composition, framing, and lighting before final generations. Three situations where this fits:

Previz before committing: compare several camera angles and lighting setups on grey boxes in minutes, then spend credits on the final render of the version that works.

A pitch due tomorrow: a client review needs something to look at, and a rough blocked scene with a camera move communicates the idea without a production day.

Planning multi-shot sequences: block the whole scene once, then frame each shot from the same geometry, keeping spatial continuity across the sequence.

Where does a finished scene go?

A finished scene has three exits. Video generation runs from the chat, so the scene you block is the scene you render. Export saves it as an animated MP4, a static frame, or a GLB file. And the same scene layer runs in the TAVOA AI Blender plugin: Scene Builder assembles blockouts in an open .blend file, on the same account and the same credits.

TAVOA AI 3D Jutsu: Build and Animate 3D Scenes from a Prompt

Got any questions left?

No. The agent assembles the scene from a text description, and manual editing is limited to moving objects, adding primitives and lights, and framing cameras. Knowing 3D software helps but is not required.

No, it is a standalone browser workspace. TAVOA AI's Blender plugin is a separate product that works inside Blender; 3D Jutsu requires no installation.

Yes. GLB files import into the scene with their animation intact, and finished scenes export back as GLB, animated MP4, or a static frame.

Messages on Auto, the default model, are free. Picking a specific LLM charges your regular TAVOA AI credits per message, the estimated cost shows before you send, and the price depends on the model.

Yes. Invite people by email as an admin, a collaborator, or a viewer, or open access to anyone with the link. Revision history keeps every version restorable.

by TAVOA AI

Discover more

TAVOA AI for Blender: Features, Installation, and MCP Bridge Setup

How to Make a Game Trailer with AI in 2026: Full Workflow + Prompts

TAVOA AI Meets DaVinci Resolve: Generate B-Roll Right in Your Edit