I spent a month building a dedicated 3D director's station for AI short film creators. Today, I’m officially open-sourcing it—free for everyone to use. I’ll walk you through each feature one by one.

Article author and source: Jiguang

Guys, I've been holding onto this article for a long time.

It’s not because it’s hard to write, but because I wanted to refine this even more, and even more, before sharing it with everyone.

Anyone creating AI short films likely shares a common pain point: scene consistency and character positioning. When you use prompts to describe “the male lead sitting on the left chair, the female lead standing on the right, three meters apart, with a café in the background,” the AI-generated image might show the male lead standing upright, the female lead correctly placed, and the café transformed into something unrelated to your reference image.

You changed the prompt, added details, added constraints, added negative prompts, went through all sorts of adjustments, but the output was still random.

This feeling is too unbearable.

You clearly have a very vivid image in your mind—you know where the camera should be, how the characters should stand, and what the spatial relationships in the scene should be—but you just can’t precisely convey it to the AI.

Language has its limits; my previous solution was to use Blender—first setting up the scene in 3D, positioning the characters, adjusting the camera angle, taking a screenshot, and then using that image as a reference for the AI. The results were significantly better, maintaining scene consistency and character positioning.

But the problem is that Blender is too heavy—I’m not saying Blender is bad; it’s an incredible tool. However, for the simple need of “I just want to quickly set up a storyboard,” it’s overkill. It takes time to launch, the models require adjustment, the workflow has a learning curve, and even just moving a character’s position can take forever.

I need a lightweight tool specifically designed for character positioning.

I spent a month building a 3D director's platform specifically designed for AI short film creators. Today, I’m officially open-sourcing it—free for everyone to use.

I’ll explain each feature one by one.

Director’s view / Camera angle switch

The logic of this feature works as follows: you can add any number of camera positions within the scene. After adding them, you can freely switch between “Director’s View” and “Camera View.” The Director’s View offers a bird’s-eye perspective, allowing you to see the entire scene and the position of all camera setups. The Camera View shows exactly what that specific camera captures.

This switch is real-time—when you adjust the character’s position, you can instantly switch to the camera angle to see “what it looks like from the camera’s perspective.” Anyone who has worked on storyboards will understand how crucial this is.

Includes 8 different characters and 20 different character poses.

Next is the character system, which includes eight different character models and twenty different poses.

You might think eight character types are too few—but for storyboarding, they’re more than enough. You don’t need detailed character models; you just need placeholders that represent “a person is standing here.” What matters most are the poses. I’ve included all the most common ones: standing, T-pose, walking, running, sitting, crouching, one-knee kneeling, two-knee kneeling, hands on hips, leaning, bowing, thinking, fighting, kicking a ball, throwing, pushing, waving, reaching out, hugging the hips, and looking at a phone.

And these 20 poses have been carefully selected—they are the most commonly used movements in short dramas.

One-click addition of basic geometric shapes

Why am I emphasizing the "basic geometric shapes" feature?

Because often you don’t need a detailed table model—just a placeholder saying “there’s a table.” A simple rectangular prism is enough. Storyboarding isn’t about creating CG; it’s about speed, efficiency, and being able to visualize your ideas on screen within five minutes.

Import local FBX/OBJ models with a customizable model library

If you need a more detailed model, you can import your own—support for local FBX and OBJ file formats is available.

I also added a model library feature—you can store your commonly used models there, and next time you need them, just drag them directly from the library instead of searching for files each time.

This feature should be very useful for those with some 3D experience. You can organize your favorite models into a personal storyboard asset library.

Crowd array—choose as many people as you like.

The next feature is my personal favorite: the Crowd Array.

Those who’ve handled big productions know that extras can be a hassle. If you want to depict a scene with “a crowd in the square,” do you really need to place each person individually? That would be exhausting.

The crowd array logic allows you to specify an area, set the number of rows and columns, and generate the layout with one click. Want 10 people? Generate 10. Want 100 people? Generate 100. After generation, you can move or rotate the entire array as a whole.

Moreover, each user's pose can be customized, eliminating the stiff, identical appearance that would occur if everyone looked the same.

After implementing this feature, I found it incredibly satisfying to use—I set up a "concert audience" scenario and had it ready in two seconds.

Panorama Import and Background Adjustment

Again, regarding the background, support for importing panoramic images.

You can import a 360-degree panoramic image as a background, then adjust its rotation angle and brightness. This way, your scene won’t appear in a dark void but instead will have a realistic environmental backdrop.

During my testing, I used some AI-generated panoramic images, and the results were excellent. Simply generate a panoramic image using Nano Banana Pro or GPT Image 2, import it, and the entire scene instantly gains atmosphere.

On-site filming, screenshot documentation, and basic shot management

I've also put in effort into seat management.

Each camera position can take a screenshot individually, and the screenshots will be automatically saved to the screenshot history.

What is this used for?

Set up a scene, take a shot from Camera 1, adjust the subject’s position or change the camera angle, then take another shot—repeat this process of adjusting and shooting. In the end, you’ll have a series of screenshots that form your storyboard sequence.

Viewport aspect ratio box, 3x3 grid, pan/rotate/zoom controls

You can also use different aspect ratios—landscape, portrait, or 1:1. The aspect ratio frame will be displayed in real time on the viewport, so you can clearly see the boundaries of the final image.

There are also grid assist lines—a feature I definitely won’t miss, as it helps with composition.

In terms of control, basic operations such as panning, rotating, and scaling are all intuitive and smooth.

After selecting an object, you can precisely adjust its position and angle. I’ve added some shortcuts so common transformations don’t require navigating menus each time. Nothing elaborate here—just everything you’d expect, designed to feel intuitive and smooth.

That covers the features.

You might be wondering, how big is this? Do I need to install it?

No installation required.

This is a web-based tool that you can use by simply opening your browser. I built it with Three.js, and all calculations run locally—no internet connection or server is required, and your data is never uploaded.

Just provide the GitHub address to Codex or any AI tool, and tell it: Help me clone this project and open the preview.

To be honest, building this tool was quite interesting.

Over the course of a month, I started from scratch and gradually built up the features one by one. After adding each feature, I used it myself to see if it felt intuitive—if it didn’t, I tweaked it. The final version is one I genuinely use myself.

The open-source address is provided here:

https://github.com/jiguang132/storyai-3d-director-desk.git

Free to use, no restrictions.

If you find it helpful, giving a star is the best support.

I’ve always believed that perfecting the toolchain is a crucial part of creating AI-generated short films. While AI’s ability to generate images and videos has become quite strong, there’s still significant room for improvement in how precisely we can communicate what we want to the AI.

Language has its limits, and 3D previsualization is an excellent complement.

First, set up the scene in 3D to transform it from a "fuzzy idea" into a "concrete visual," then use this visual to guide the AI in generation—this will make the entire process much smoother.

We hope this tool is helpful to everyone.