Charakuru README

项目的 README,来自 github.com/nanocle/Charakuru。合并为一个 markdown 文件:README.md。

Charakuru (きゃらくる): AI Character Modeling Kit, sponsored by Tripo

Charakuru / きゃらくる

A kit for turning one character illustration into a VRM avatar, sharing the work with an AI agent.

日本語 · 한국어 · 简体中文 · Tutorial · Download (Releases) · Terms of Use · Discord

version Node.js output license

💬 Join the Discord community!

Tripo has kindly set up a dedicated Charakuru category on its official Discord server!

Ask questions, share feedback, and show off the avatars and other creations you've made with Charakuru. You're also welcome to chat with fellow creators and Nano (ナノ), the creator of Charakuru!

We'll also use this space to announce upcoming events, including contests using Charakuru. Whether you're curious about trying the kit or already working on a project, come say hello!

👉 Join the official Tripo Discord

Once you've joined, head to the dedicated Charakuru category!

What is Charakuru?

A character modeling support kit powered by Tripo's Smart Mesh.

Creating high-quality character models with Tripo alone has been difficult, making workflow research and Blender knowledge essential. This kit solves that problem.

It packages an original workflow devised by Nano into a local app and agent Skills, distributing it in a form users can reproduce. You can leave Tripo and Blender operations, rigging, and other complex setup work to GPT-6.1 Sol or GPT-6 Astra.

Some steps still need a human, but you don't need to operate Blender yourself. Everything can be done through simple controls in the local app. This practical kit aims for a good balance: more realistic than handing everything off to an AI agent, and easier than doing everything by hand!

The learning curve might be a little steep. If anything feels difficult or unclear, we'd appreciate your feedback in an issue!

All you need is a single full-body illustration to create a VRM model.

[!WARNING] This kit is still in development.

Some documentation hasn't caught up yet, and there are bugs and missing features. Thank you for your understanding.

These docs were written by AI.

They can be a little hard for people to read. We recommend opening this repository (or the extracted release folder) in a coding agent such as Codex or Claude Code, and learning about the kit by talking with the agent.

For example: "Read README.md and AGENTS.md and tell me what this kit can do."

From the illustration to 8 images, a 3D model and motion review

Example: 海月 みる (Miru Umitsuki). Left to right: illustration → 8 images painted by the agent → 3D model → in-app motion review.


How it feels to use

You go back and forth between reviewing in the app and asking in the chat, five times.

Workflow: you review, the agent works

  1. When you're happy with a step, press the app's continue button and copy the request it shows.
  2. Paste it into your chat with the agent (Codex) and send it.
  3. When the agent finishes, the app switches to the new state on its own and stops at the next checkpoint.

There's no exporting, importing or re-saving files by hand. The agent never moves past a checkpoint on its own, and you are the one who picks the candidates.

Compare model candidates in a dialog and reopen it with Choose model candidates in the model editor. The editor focuses on proportions; all three reference views share character height and foot level. Selecting a different candidate after texturing creates a separate project copy and preserves the original. Save the finished VRM at the end of Motion check.

Reviewing the 8 images Choosing the head and body candidates Checking the motion with a walk cycle
Review images
Head (with hair) and body, four views each. You can ask to fix a single image.
Pick 3D candidates
Rotate and compare four candidates for the head and four for the body.
Check the motion
Watch shoulders and knees bend in a walk cycle.

Design ideas

How the app, project.charakit and the agent fit together

What you need

Used for
Node.js 22 or later Running the app.
A desktop Chromium-based browser (Chrome, etc.) Opening the app.
Codex with image generation and browser control The agent that does the work. It paints images, operates Tripo in the browser and drives Blender.
A Tripo account (signed in in the browser) 3D candidates, retexturing and rigging.
Blender 5.2, VRM Add-on for Blender, Blender MCP Weights, physics and VRM export.
An illustration you're allowed to use PNG / JPEG / WebP, up to 20 MB. A full-body character sheet works well.

Accounts, terms and fees for external services are your own. The agent checks what's missing at the start and tells you.

Getting started

  1. Download charakuru-1.1.0-….zip from Releases and extract it to any folder you can write to.
    GitHub's automatic "Source code" archive contains only these documents, not the kit.

  2. Open the extracted folder in Codex and ask:

    Read the root AGENTS.md and help me start creating a character with this kit.
    
  3. The agent checks your setup and starts the app. Open http://127.0.0.1:8766 in your browser and you're ready.

To start it yourself, run npm start in the extracted folder.

Then follow the tutorial, which walks through your first avatar screen by screen.

New agent requests use the begin command to avoid Codex on Windows mistaking the old start argument and a URL for a URL launch. If an old request is rejected with blocked by policy, see the recovery steps and copy a new request from the updated kit.

What's in the kit

AGENTS.md The agent's starting point: requirements, stages and saving rules.
public/, server.mjs The local review app.
skills/ Nine Skills for the production workflow.
scripts/ Scripts the agent uses to update the project.
assets/donor/weighted-body.blend The base body used for weight transfer.
docs/ The tutorial, the motion review guide and plain-language terms in English, Japanese, Korean and Simplified Chinese.
The nine Skills
Skill Used for
charakit-project Checking, reading and updating the project file
character-parts-images Painting the head (with hair) and body in four views: 8 images
character-three-views Full-body reference views for proportion adjustment
character-material-brushup Finishing the painting and materials without changing the design
tripo-agent-workflow Tripo candidates, part sorting, UV and retexturing, rigging
tripo-studio-download Receiving exported Tripo Studio models and checking the saved files
avatar-weight-transfer Separating swinging parts and transferring weights from the base body
avatar-spring-setup Hair and outfit physics, VRM export and checking
tripo-character-setup Detailed rigging from joint markers you place (only when you choose it)

Current limitations

Terms of Use

You may freely use the kit for non-commercial and commercial purposes within the following conditions.

Contact Nano in advance before redistributing the kit or integrating it into a product or service that provides kit code to users.

Private modification and copying are OK.

You must not extract bundled assets alone to take them outside the kit or redistribute them separately.

To the extent permitted by law, we accept no liability for damage arising from use of this kit.

If your entire business's annual revenue and cumulative funding are both below ¥100 million, you may use the Kit within the standard License conditions without a separate agreement or registration. If either is ¥100 million or more, contact Nano and obtain a separate written license agreement with the Licensor or an authorized representative before starting or continuing commercial or business use. Revenue and funding are not added together.

Assess annual revenue using the most recently completed fiscal year. If there is no defined fiscal year, use the preceding twelve months; if no fiscal year has yet ended, use actual revenue since the business began, limited to the preceding twelve months. Do not annualize revenue for a business operating for less than twelve months or for a short initial fiscal year. Reassess cumulative funding when funds are received, regardless of the fiscal year. See Terms of Use explained for calculation details and agreement timing.

Read the formal license here.

Reporting issues

Please report problems and ideas through Issues, following the reporting guide. Pull requests are not accepted.


Made by Nano (ナノ) · sponsored by Tripo

Expression authoring

The Expressions tab follows Rigging. Its top button copies an agent request. Import VRM 0.x/1.0, GLB, or reopen a character kit; select expressions, combine their strengths, and adjust minimum/maximum values. Controls are disabled until expressions exist. The source model, selection and adjustments are saved in the kit. Custom names are retained. These preview settings do not rewrite the source VRM.

Codex uses bundled knowledge and tools to fit a skin-only eye mask, transfer upper/lower lip motions separately, and create independent brow poses. The agent chooses correspondences by looking at the model, then refines attractive expressions with visual comparison and Blender editing. The library provides 8 eye, 10 mouth and 5 brow poses. Fixed multi-view captures, intermediate strengths and overlays support finishing; successful transfer or saving does not establish visual quality. Original geometry, UVs, materials and existing expressions are retained. Results return as optional Blend/GLB and comparison data to the same project; existing VRM exports are unchanged. See the guide. Requires Blender 5.2+ and the VRM Add-on for MToon preview. Computer Use can be used for local finishing. The expression assets were created by the kit creator and follow the same restrictions on extraction and redistribution as the other bundled assets.

Charakuru (nanocle/Charakuru)

正在搜索数据库…

这是什么?Charakuru 是一套工具包,可以把一张全身角色插画变成 VRM 虚拟形象:它将一个本地审核应用与九个 AI 智能体 Skill 配对,这些 Skill 驱动 Tripo 和 Blender 绘制参考图、生成 3D 模型、绑定骨骼并导出完成的 VRM。
这是独立应用,还是我需要一个 AI 编程智能体来驱动它?它是一个本地网页应用(在 Chromium 内核浏览器中打开),配合九个 Codex 智能体 Skill 使用:你在应用中逐步审核,复制生成的请求,然后粘贴到你的 Codex 对话中,由它替你操作 Tripo 和 Blender。
我该怎么让它运行起来?从 Releases 下载 charakuru 压缩包,解压到任意一个你有写入权限的位置,然后在 Codex 中打开该文件夹,Codex 会检查你的环境并启动应用(你也可以自己运行 npm start);内置教程会带你完成第一个虚拟形象。
开始之前我需要安装哪些东西?你需要 Node.js 22+、基于 Chromium 的桌面浏览器、一个具备图像生成和浏览器控制功能的 Codex 智能体、一个已登录的 Tripo 账号,以及安装了 VRM Add-on 和 Blender MCP 的 Blender 5.2。
目前有哪些限制?该工具包仍在开发中,部分文档过时、功能尚不完整;复制请求并不会启动智能体(你必须自己粘贴到对话中),新建的角色默认为中性面部表情且没有视线动画,并且不接受 Pull Request(仅接受 Issue)。
我做的模型可以商用吗?可以,但需遵守随附的使用条款,前提是你的企业年营收和累计融资额均低于 1 亿元人民币;超过该门槛,你必须获得创作者 Nano 单独出具的书面许可。
为什么工作流总是在检查点停下,而不是让智能体一次性全部完成?这样设计是有意为之:像挑选模型候选这类视觉判断交由人工完成,而文件拆分、整理和导出则交给智能体,目的是在完全手动和完全自动化之间取得平衡。
这个应用支持哪些语言?界面以及标准的智能体请求和报告支持英语、日语、韩语和简体中文,即使随附的参考文档使用其他语言撰写。
我可以为我的虚拟形象定制面部表情吗?可以,有一个可选的「表情」标签页,让你导入 VRM/GLB 文件,并从包含 8 种眼部、10 种嘴部和 5 种眉部姿势的库中构建姿势;Codex 会通过视觉对比和 Blender 编辑来进一步打磨这些姿势。
开始时需要什么格式的图片?一张全身角色插画,格式为 PNG、JPEG 或 WebP,大小不超过 20 MB;全身角色设定图效果尤其好。
作者
nanocle
大小
9.2 MB
星标
1.7K
时长
1个月
提交 · 90 天