Tesseract lets AI agents edit 3D scenes while preserving objects and layers

The video is almost approved, but the client asks to slow down the camera and give the device housing a matte finish. At that point, what matters is being able to adjust two parameters while preserving the rest of the scene. Mirage offers Tesseract for tasks like these: an engine controlled by an AI agent. The practical idea is straightforward: delegate assembly and revisions to the agent, and keep the result as an editable project.

This deserves the attention of developers and product teams that need device demos, animated explainers and short presentation videos. However, the original post's claim that the tool “buries” diffusion-based video is too bold. The available material demonstrates the possibilities of a different production workflow, but does not prove an advantage in cost, speed or quality.

What the agent actually gets

According to the repository documentation, Tesseract consists of the tsrct CLI tool and skills that explain to the agent how to work with video, animation and 3D. The user describes the task and supplies source assets if needed. The agent then creates a .tsrct project, checks the result and makes changes within that same project.

Imagine a demo of a desktop device. The brief specifies the housing shape, material, lighting and camera movement. After the first review, it becomes clear that the camera moves past a port too quickly. The next request concerns a specific movement rather than the entire image: linger at the port while preserving the other decisions. This is an example of a task for evaluating the tool, not a result from our own testing.

For a team, the interesting part is how the work is organized. The project has a state that can be revisited after each round of feedback. Approved elements become the foundation for the next revision. A person still needs to check whether the agent understood the request correctly and actually preserved the required details.

Where the image comes from

In its description of the 3D capabilities, Mirage says the engine works with geometry, materials, lighting and cameras. It provides a fast rasterized preview and path-traced rendering. The latter calculates reflections and indirect lighting within the scene. That explains the appeal of this approach for product shots, where shape and surface behavior matter.

But having this kind of renderer does not, by itself, guarantee a good result. A poor model remains a poor model even when the lighting is calculated convincingly. Incorrect proportions, intersecting objects and awkward camera angles still have to be spotted during review. A beautiful individual frame also says nothing about how smoothly the camera moves throughout the video.

There are significant limitations, too. Tesseract does not include a built-in model for generating arbitrary meshes from text: a complex object may require an existing 3D model. Blender import is partial. Rendering speed depends on the runtime environment; without a suitable GPU, it may drop considerably. Promising an instant photorealistic result from a single prompt would therefore be misleading.

The handoff to a designer also needs to be checked separately. The product page lists Adobe project import and export, but support is partial, and After Effects export is experimental. The engine being free does not eliminate the cost of the agent itself, either.

How to assess the value for your team

I would start with a short product video and three predefined revisions: change the material, adjust the camera movement and edit an on-screen title. This is a small experiment that makes it easy to see which elements were preserved, where human intervention was needed and how long approval took.

Measure the entire cycle: preparing source assets, reaching the first acceptable version, making revisions, rendering and handing over the project. If the agent produces a draft quickly but the team then spends a long time restoring materials after export, the initial savings may disappear. It helps to define in advance exactly what counts as a successful handoff.

Tesseract offers a reason to test agent-driven management of a visual project on a real task. For a product team, the decisive result will be a predictable next revision: how long it takes, what it preserves and whether the team can continue working in its usual way afterward.