[AI of Interest #05] From images to 3D models, and videos to character movement! I checked out 8 trending GitHub projects

“Create 3D models from a single image”

“Animate 3D characters using ordinary videos”

“AI that directly operates Blender”

When you hear things like that, don’t you want to give them a try?

This time, I found an interesting post on X.

It introduces 8 GitHub repositories related to AI video, AI-based 3D production, Three.js, and more.

In the previous “AI of Interest #04,” I introduced how to create videos from program code using Claude Opus 5.5.

This time, the theme goes even further into AI-based 3D model production, character motion generation, and automatic Blender operation and more.

However, there are some things I’m curious about.

Can these really be used for free?

Can beginners handle them?

And is it safe to let AI operate my PC?

This time, I investigated mainly based on the official GitHub pages of these 8 projects.

*This article is based on public information as of October 23, 2026. While I have checked the GitHub READMEs and similar documentation, I have not performed actual installations, operational tests, or security audits of the entire code.

1. An era where you can create 3D models with AI?

First, to understand this topic, I will briefly explain the difference between “images” and “3D models.”

Illustrations created with image generation AI are basically flat images.

For example, suppose you generate a cute robot viewed from the front.

You cannot look at that image from the right or from behind.

On the other hand, 3D models have height, width, and depth.

You can rotate the robot you created 360 degrees, change how the light hits it, and set up animations.

Until now, creating 3D models required specialized software operation and knowledge of modeling.

However, recently AI has started to estimate 3D shapes based on images and write code to create 3D models.

Here are two technologies you should keep in mind.

Blender

This is a high-performance 3D production software that can be used for free.

It is equipped with many functions, such as “modeling” to create the shape of a 3D model, texture settings, animation, and video rendering.

Three.js

This is a JavaScript library for creating 3D expressions on a web browser.

For example, you can create a mechanism where a product rotates when you move the mouse on a website, or a 3D character walks around the screen.

The GitHub projects introduced this time combine such technologies with AI.

2. 8 Trending GitHub Projects

1. img2threejs: Creating 3D objects from images

GitHub: img2threejs

The first one I will introduce is “img2threejs,” which I was particularly interested in this time.

This is a project that reconstructs objects shown in reference images in 3D using Three.js code.

The official README introduces a mechanism that assembles 3D shapes using only code and improves them while checking the quality.

The feature is that it assembles the shape of the object programmatically, rather than simply stretching the image in the depth direction.

The completed model is designed to be rendered in Three.js and can also be used for animation.

For example, you can think of use cases such as creating a 3D object that can be rotated and displayed in a browser based on a reference image of a robot or furniture with a design you like.

However, the shape of parts not visible in a single image cannot be determined.

Therefore, AI inference and correction are necessary, and it does not necessarily result in the exact same shape as the image.

This is an interesting technology for those who want to turn illustrations into 3D representations that can be moved on a website.

② Motion2MixamoRig: Creating character movement from ordinary videos

GitHub: Motion2MixamoRig

This is a project that creates motions for 3D characters from videos of people moving.

It aims to apply human movement to 3D characters with skeletal structures compatible with Adobe Mixamo.

For example, you film yourself dancing with your smartphone.

The way it is used is to estimate body movement from that video and have a 3D character replicate it.

Originally, there are ways to use dedicated motion capture equipment to acquire such movements.

However, this project uses regular videos as input.

Of course, the same accuracy as dedicated equipment is not guaranteed.

Challenges remain, such as pose estimation when arms and legs overlap, the issue of feet appearing to slide on the ground, and the reproduction of complex movements.

Even so, it is a technology worth noting as an option for individuals to animate their original characters.

③ dsh-blender-plugin: AI directly operates Blender

GitHub: dsh-blender-plugin

This is what I paid particular attention to in terms of both technology and security this time.

It is a plugin for AI to operate Blender.

The public README explains that 15 types of major tools can be used.

It has functions to acquire the state of the screen, manipulate objects, and execute rendering.

For example,

“Create a model of a red sports car and light it from the front-diagonal,”

you could instruct the AI, and it would call Blender functions to carry out the work.

The mechanism involves linking DeepSeek Harness with Blender and operating Blender through a local communication channel.

There is also a feature that allows you to execute arbitrary Python code within Blender.

This is extremely powerful, but it requires caution when handling.

I will explain the safety aspects in detail later.

④ dioramas: Creating cinematic 3D websites

GitHub: dioramas

Next, I will introduce a framework for building interactive 3D websites using AI-generated assets.

The official GitHub showcases 3D expressions using Three.js and WebGL.

By combining AI-generated assets, you can create a small world within the screen and build a website where you can look around that world with a camera.

For example, you could prepare a three-dimensional room on a product introduction site, allowing users to view products with mouse controls.

Or, you could create a portfolio site that represents a fictional city in 3D.

It seems useful for such purposes.

Unlike regular websites, it requires control over 3D models, cameras, and lighting, but the visual impact is significant.

However, even if the framework itself is free, there may be additional costs if you use external services to generate the AI assets.

⑤ video-shotcraft: AI produces product introduction videos

GitHub: video-shotcraft

This is a skill for AI agents that works with Claude Code or Codex to produce product introduction videos.

The feature is that it uses a video production framework called Remotion.

Remotion is a system for creating videos as programs using React.

In video-shotcraft, you use pre-prepared shot recipes to assemble the composition and movement of the video.

The public README explains that it was organized into 124 cards and 150 shots in the October 2026 update.

Although this differs from the “152” figure in the original X post, the structure has changed due to project updates.

What is interesting is that you can use actual product webpages and the like as source material.

For example, you could use the screen of a web service you created yourself to make a promotional video that looks like a movie trailer.

This is a project that seems to have potential uses for individual developers as well.

However, the actual quality, required production time, and costs have not been verified.

⑥ lemo-opuscar: Choose a style and create videos with code

GitHub: lemo-opuscar

This is a project I also introduced in the previous “AI of Interest #04”.

It is a collection of styles for creating short videos using code, utilizing tools like Claude Opus 5.5.

The October 22, 2026 update introduces 59 styles, as well as features for modifying existing styles and creating new ones.

The approach of rendering video with code, without using image or video generation models, is its defining feature.

Since this technology is related to the 3D production in this article, I think reading it alongside the previous article will deepen your understanding.

⑦ jianying-headless: Automating video editing with programs

GitHub: jianying-headless

This is a project for automating the editing process of the video editing software “Jianying”.

Jianying is a ByteDance-affiliated video editing software related to CapCut.

jianying-headless provides functionality to create editable projects from structured editing instructions, edit independent copies, and export videos.

The public README explains that an Apple Silicon-equipped Mac and a compatible version of Jianying are required.

What is important here is that this project is not an official SDK for Jianying.

Furthermore, the license for the original parts is for personal study and non-commercial use; commercial use requires written permission from the developer.

Also, you need to separately check the terms of use for the Jianying application itself and its assets.

The idea of automating video editing is interesting, but it does not seem to be a tool that general users can easily implement.

⑧ mg-styles-15: 15 types of motion graphics

GitHub: mg-styles-15

Finally, this is a collection of motion graphics works that I also introduced in the 4th installment.

For the 15 types of video styles, the finished videos, prompts, and source code are made public.

There are various types, such as hand-drawn style, pixel art, 3D expression, and cyberpunk-style UI.

It is also related to the theme of 3D production this time, and in some cases, videos using Blender are also introduced.

Even just looking at the published works before installing anything seems like it would be a good learning experience.

3. Which one looks the most interesting?

If we organize these 8 items by purpose, it looks like this:

Purpose | Featured Project | Want to recreate images in 3D | img2threejs | Want to transfer human movement to 3D characters | Motion2MixamoRig | Want to have AI operate Blender | dsh-blender-plugin | Want to create 3D websites | dioramas | Want to produce product introduction videos | video-shotcraft | Want to create short videos with code | lemo-opuscar | Want to automate video editing | jianying-headless | Want to learn animation | mg-styles-15

Personally, I am particularly interested in “img2threejs” and “dsh-blender-plugin”.

img2threejs has the potential to represent objects I imagine as 3D models on the web.

dsh-blender-plugin is attractive because AI performs the difficult operations of Blender on your behalf.

However, since the latter executes code within your PC, you need to sufficiently verify the content before trying it.

4. Can you really use them for free?

You need to be a little careful here.

The fact that source code is published on GitHub and the ability to complete videos or 3D models for free are two different things.

For example, the following costs might be involved:

AI model usage fees

When using AI agents like Claude, subscription plans or API fees come into play.

3D asset generation services

If you use external image/3D generation services like dioramas, the pricing structure of each service will apply.

PC processing power

For 3D rendering or human motion estimation, your PC’s CPU, GPU, and memory may become important.

Other software licenses

For example, Remotion, which video-shotcraft uses, also has its own terms of use, and depending on the company’s usage, a paid license may be required.

Furthermore, there are projects like jianying-headless where, even if the source code is public, there are restrictions on commercial use.

“Available for free download” and “free to use commercially without additional costs” are different things.

Since I did not actually run each tool this time, the specific production costs remain unconfirmed.

5. [Safety Verification] Is it okay to leave 3D production to AI?

I will check the security this time as well.

In conclusion, for all 8 items, I have not performed actual machine verification or a full audit of the source code, so a comprehensive safety judgment is withheld.

I will introduce points that require special attention.

Caution Point 1: You can execute arbitrary Python code in Blender

The dsh-blender-plugin README explicitly states that it has the ability to execute arbitrary Python code within Blender.

Python is a programming language capable of not only manipulating Blender objects but also reading and writing local files.

In other words, a design that executes AI-generated code as-is could potentially lead to unintended file operations.

The official README also warns that the code execution feature should only be used in a trusted environment.

Additionally, it advises using 127.0.0.1 for local communication connections and not exposing them externally.

While this is a certain level of security measure, it cannot prevent all risks.

It should be noted that the mere presence of such powerful features does not mean the program is malicious.

Caution Point 2: AI agents reading external instructions

Tools that use Claude Code or Codex may have the AI reference files such as README or SKILL.md within GitHub.

If malicious instructions are mixed into those files, the AI could potentially perform operations different from its intended purpose.

This is called prompt injection.

While I did not actually discover any malicious instructions in the repositories this time, it is an issue to be aware of when using AI agents.

It is important not to trust instructions obtained from external sources blindly and to verify the content to be executed.

Caution Point 3: Corruption of video editing data

jianying-headless manipulates editing projects programmatically.

The official documentation explains designs that use copies separate from the original project and procedures for data verification.

Even so, issues with editing results may occur due to unsupported software versions or unexpected errors.

Therefore, backing up original data and confirming supported versions is important.

Safety Assessment

In this investigation, I checked the public README, licenses, and design considerations.

However, I have not audited all dependency packages, external communications, or code execution paths.

Therefore, I cannot guarantee that individual software is safe.

It is best to start by checking the README and demos, and if you actually install them, it is recommended to verify permissions and execution details in a dedicated testing environment.

6. Be careful with commercial use

Just as important as security are licenses and copyrights.

For example, even if licenses like MIT or Apache-2.0 are displayed on GitHub, it does not necessarily mean that external images, music, fonts, and 3D assets can be used under the same conditions.

When using Motion2MixamoRig, you also need to check the usage rights of the original video and the terms of use for Mixamo.

Also, when extracting motion from videos filmed by others, you need to be mindful of the rights associated with those videos.

You should be especially careful with jianying-headless.

The original parts of this project have restrictions on non-commercial use.

If you intend to use it for business, you need prior permission from the developer.

You should always distinguish between what is technically possible and what is permitted for commercial use.

7. Can individual users use these?

Among the technologies introduced this time, there were several that individuals could try out.

For example, displaying a 3D object created with Three.js on your own website.

Using a video you filmed yourself to animate an original character.

Or, having AI create a product introduction video.

These methods of utilization can lead to individual development and content creation.

However, many projects require initial setup and programming knowledge, so they are different from apps that beginners can use with a single click.

Personally, I recommend starting by looking at the publicly available demos and samples to understand what is possible, rather than installing high-functionality tools from the start.

Furthermore, I believe that verifying the ones you are interested in one by one will deepen your understanding of the technology.

8. Conclusion: Will AI make 3D production more accessible?

After researching these eight GitHub projects, I felt that the scope of creative production using AI is expanding.

Until now, the focus has been on having AI generate images and videos.

However, mechanisms are now emerging that allow AI to write code for 3D models, operate 3D software, and extract character movements from videos.

What is particularly interesting is that AI is moving not only toward generating the work itself, but also toward operating production software and controlling the production process.

If this develops further, it may become easier for people without specialized knowledge in 3D production to bring their ideas to life.

On the other hand, there are risks in giving AI agents significant operational authority.

You also need to check API usage fees, PC performance, licenses, and copyrights.

The conclusion this time is, “The possibilities for individuals to take on challenges are expanding. However, you should understand the mechanisms and safety before using them.”

Personally, I would like to look into “img2threejs,” which creates Three.js 3D objects from reference images, in a bit more detail in the future.

From AI-based image production to 3D production, and then to automated operation of production software.

Once again, I found many technologies that caught my interest.

Reference Materials

*This article is based on information publicly available as of October 23, 2026. I have not performed hands-on verification of the actual operation, costs, or security of each project. Specifications and licenses are subject to change.