topview-3d-builder README

项目的 README,来自 github.com/topviewai/topview-3d-builder。合并为一个 markdown 文件:README.md。

Topview 3D Builder

Topview 3D Builder

Let AI build your 3D scenes

From an idea, to a 3D scene built by your agent, to a real animated video. Free, runs locally, no sign-up.

License: Apache 2.0 Python 3.11+ Node.js 20.6+ Runs 100% local

Quick Start | Examples | Tutorial | 3D Builder | Topview.ai | Issues

https://github.com/user-attachments/assets/ae8923ce-e03e-473c-8e75-941bcb945f9c

Build the scene, place the cameras, plan the camera move in Studio, then get the shot.

Quick Start

Recommended model: Qwen 3.8 Flash. It costs about 1/150 of GPT 6 Astra and reaches a similar result. We recommend using it.

Works with ChatGPT / Codex, Qoder, Claude Code, Cursor and any agent with skills support. You need Python 3.11+, Node.js 20.6+, and either uv or pipx.

1. Install the plugin

ChatGPT / Codex: open the plugin page in the app and install it from there.

Install Topview 3D Builder in ChatGPT / Codex

Other agents (Qoder, Claude Code, Cursor, or any agent with skills support): send this to your agent.

Install the Topview 3D Builder plugin from https://github.com/topviewai/topview-3d-builder for the agent I'm using (Qoder, Claude Code, Cursor or another agent with skills support). Follow the install steps in the README, then tell me when to restart.
What your agent will run

Claude Code (restart Claude Code afterwards)

claude plugin marketplace add https://github.com/topviewai/topview-3d-builder.git
claude plugin install topview-3d-builder@topview-3d-builder

Cursor (restart Cursor afterwards)

git clone https://github.com/topviewai/topview-3d-builder.git ~/.cursor/plugins/local/topview-3d-builder

Any agent with skills support, including Qoder

npx skills add https://github.com/topviewai/topview-3d-builder.git

Then pick topview-3d-cli. The first render downloads a browser component once; after that it is cached. For the CLI and project format, see the CLI reference.

2. Start a new session and describe the scene you want

Build a scene: a character sits next to a box. Render a front view.

3. Review the scene and keep refining

Keep talking to your agent, or open Studio and adjust by hand. When you're happy, turn the 3D scene into a real animated video.

What is Topview 3D Builder?

Topview 3D Builder is a 3D director's stage for AI video. You describe the scene, and your agent builds it in 3D: who stands where, who faces what, what they hold, and where the camera sits. That 3D scene then guides the video model, so the staging comes from you instead of from luck.

It covers the whole path from an idea, to a 3D scene, to a real animated video. The 3D scene fixes blocking, scale and camera angles. Texture, lighting and performance come from the video model.

An idea  →  Agent builds the 3D scene  →  Auto-check and refine  →  Multi-camera render  →  Real animated video

Everything before the video runs on your machine, free, with no account and no upload.

What problem does it solve?

If you make AI videos, you've probably hit this: text prompts can't describe space precisely, so you can't control where people stand or where the camera goes. Directing AI video ends up feeling like gambling. 3D Builder gives you control back.

The problem How 3D Builder solves it
😵 Blocking is a guess. "Two people facing each other across a table" comes back side by side, clipping, or without the table. 🎬 Direct, don't gamble. You place characters, poses, props and cameras, so blocking and facing are decided, not guessed.
🎰 Camera angles are a lottery. You write a paragraph and get one random angle. 🎥 Design the camera move in advance. Plan push-ins, pull-backs, rises, orbits and cuts on a timeline.
🔁 One tweak, full redo. Move a character half a step and everything changes on regenerate. ✏️ Edit one thing, keep the rest. The scene is a saved project. Move a character, change a pose or add a camera without starting over.
💸 Retries burn credits. A usable shot can take a dozen attempts. ✅ Checked before you generate. Floating or clipping characters, spacing and framing are measured automatically, so problems are caught early.
🧩 Shots don't match. Character positions and space contradict each other from shot to shot. 🎬 One scene, many cameras. Render several cameras from the same scene and characters and space stay consistent in every shot.
📐 Not everyone can storyboard. No art skills means no way to show the staging in your head. 💬 Just describe it. No 3D software to learn. Ask in plain language and your agent builds the scene.

How it works

  1. Describe the shot. Tell your agent the scene you want, for example: "Two people sit across a table, a woman is talking. Give me a front, an over-the-shoulder and a top-down shot."
  2. The agent builds the 3D scene. It uses built-in characters, poses and props, and places the cameras.
  3. Check and refine. The scene is measured automatically. Keep talking to your agent, or open Studio and drag things around.
  4. Render and generate. Render the cameras, plan the camera moves, then turn the 3D scene into a real animated video.

Example

Each row shows the prompt, the 3D scene built from it, and the video generated from that scene: the way you stage it in 3D is the way it gets shot.

Input 3D scene
generated by 3D Builder
Generated video
rendered by Topview.ai
Reference image:
Reference image: roadside car repair
Prompt: Build the 3D scene from the reference image.
0–3s: Pull back from the leaning man to show both men and the car.
3–5s: Over-the-shoulder push-in to the engine.
5–7s: Fixed first-person shot of the engine.
Reference image:
Reference image: man reading a tablet in a café
Prompt: Build the 3D scene from the reference image.
0–2.5s: Wide shot, slow push-in.
2.5–5s: Medium shot, tracking sideways.
5–7s: Close-up of him reading.
Prompt: A person stands beside a sports car in a showroom. In one continuous ten-second shot, the camera moves from a low angle by the front wheel to a frontal view of the person, then rises and moves around to reveal the car from above.
Prompt: A runner crosses a rooftop, leaps across a gap with a forward flip, lands on the next roof and keeps running. The camera follows from behind, keeping the jump and landing in frame.
Prompt: A player jumps and shoots at a hoop. The camera starts low beside them, then rises and pushes in toward the basket, finishing above the rim and looking down over eight seconds.
More examples
Input 3D scene
generated by 3D Builder
Generated video
rendered by Topview.ai
Prompt: A rider flies a dragon over a castle. The camera starts just behind and above them, looking forward over the wings at the towers. Over eight seconds it follows the flight as the dragon clears the castle and heads out over the hills.
Prompt: One person pours from a pitcher into a cup on a café counter. The camera starts wide across the counter, then pushes in and rises to a top-down close-up of the pour over eight seconds.
Prompt: A person rides a bicycle along a city street. The camera starts level and tracks alongside, then rises and pulls back to a high angle looking down the street over eight seconds.
Prompt: A rider on horseback rides through city traffic toward a red light. Over ten seconds the camera starts high behind the rider among the cars, then cuts to the horse's point of view, ears in the foreground, as it reaches the crosswalk.

More prompts to try

Build a small café. Two people sit across a table from each other. Put the camera outside the window and slowly push in over four seconds. Render the first, middle and last frame.

A woman stands in a white studio holding a pose from the "walking" set. Three cameras: front, three-quarter, and a low angle. Render all three.

Put a car in the middle of an empty street. Orbit the camera 90 degrees around it over five seconds.

Take my current scene, add a second camera behind the main character, and cut to it halfway through.

Why Topview 3D Builder?

Built-in assets

Ready to use out of the box.

Asset Count Notes
👤 Characters 4 Child, youth, woman, man
🧍 Poses 121 Sitting, standing, walking, fighting, crawling, reaching and more
📦 Props Primitives Boxes, cylinders and more, for quickly building tables, walls and objects
🎥 Cameras As many as you need Story cameras plus a bird's-eye overview to see the whole staging at a glance

Built-in assets are made by Topview and licensed under CC BY 4.0.

Who it's for

Directors, storyboard artists and anyone previsualizing a scene can use it too.

FAQ

Do I need any 3D modeling skills? No. Describe what you want in plain language. Your agent builds, checks and renders the scene.

Is it free? Do I need an account? Building, editing and rendering are free and need no account.

Is my content uploaded? No. Every project stays in a .topview-3d/ folder on your machine.

Can it generate video? Yes. It covers the whole path from an idea, to a 3D scene, to a real animated video: you lock the staging first, then generate the video from that 3D scene.

Model Rating Why
Qwen 3.8 Flash ⭐️⭐️⭐️⭐️⭐️ Good results at a very low price. The same task costs about 1/150 of Astra, so it has the best overall value.
Opus 5.5 ⭐️⭐️⭐️ Strong aesthetics, but expensive.
Astra ⭐️⭐️⭐️ Strong 3D results and fast generation, but the most expensive.

Tutorial

Build scenes, stage action and move the camera

Learn more and join in

🌐 Topview.ai www.topview.ai
🎬 3D Builder home www.topview.ai/3d-builder
🛠 CLI reference CLI and project format
🐞 Feedback and bugs GitHub Issues

If Topview 3D Builder helps your work, give the repo a ⭐ and share what you make. We'd love to see it.

License

Acknowledgements

Thanks to the Qwen team for their strong support of this project.

topview-3d-builder (topviewai/topview-3d-builder)

正在搜索数据库…

这是什么?Topview 3D Builder 能让你的 AI 智能体根据一段白话描述搭建 3D 场景,包括角色、姿势、道具和摄像机。在把场景变成真正的 AI 生成视频之前,你可以先确定走位、取景和运镜。
这是命令行工具、智能体技能,还是完整应用?它可以作为技能或插件安装到 ChatGPT/Codex、Claude Code、Cursor 或 Qoder 中,命令行名称是“topview-3d-cli”。此外,它还附带一个完整的独立 3D 编辑器界面 Studio,智能体能做的事情,Studio 里都能做。
我该怎么把它跑起来?在 ChatGPT/Codex 中,直接从插件页面安装插件;在其他智能体(Claude Code、Cursor、Qoder)中,发送安装命令并选择“topview-3d-cli”,然后新开一个会话,直接描述你想要的场景即可。
试用前需要先装什么?你需要 Python 3.11+、Node.js 20.6+,以及 uv 或 pipx 其中之一,还需要以下支持的智能体之一:ChatGPT/Codex、Qoder、Claude Code 或 Cursor。
安装时有哪些需要注意的地方?首次渲染时会下载一个浏览器组件,之后会缓存起来。另外请注意,3D 场景只锁定站位和运镜,材质、光影和表演仍由视频模型决定。
这东西到底是给谁用的?主要面向需要跨镜头保持角色位置一致的 AI 影视创作者,以及希望在消耗视频生成额度前先锁定调度的广告、品牌和电商团队;导演和分镜师也是目标用户。
我的内容会被上传到别处吗?不会。搭建、编辑和渲染全部在你的电脑本地完成,无需账号,也不会上传任何内容;每个项目都保存在本地的 .topview-3d/ 文件夹中。
内置了哪些角色、姿势和道具?内置 4 个基础角色(儿童、青年、女性、男性),121 种姿势,涵盖坐、站、走、打斗等动作,还有用于快速搭建场景的基础道具,摄像机数量不限。
生成视频应该用哪个 AI 模型?README 推荐使用 Qwen 3.8 Flash,其成本约为“Astra”的 1/150,画质却相近。Opus 5.5 和 Astra 则是价格更高的备选,分别在美感或速度方面表现更突出。
它能直接生成成片视频,还是只做 3D 场景?它覆盖完整流程:你先搭建并锁定 3D 场景,然后用这个场景驱动真正的 AI 视频模型,生成最终的动画片段。
为什么不直接把镜头描述给视频模型?仅靠文字提示无法精确控制走位、朝向或摄像机角度,直接提示更像是在碰运气。3D Builder 可以让你精确摆放角色和摄像机,并在消耗渲染额度之前自动检查取景和碰撞。
作者
topviewai
大小
100.1 MB
星标
285
更新
提交Oct 16
时长
1个月
提交 · 90 天