Synthesis AI turns text prompts into cinematic 3D humans

Creating a cinematic 3D digital human usually involves an agonizing pipeline: building a photogrammetry rig with dozens of cameras, casting real actors, securing waivers, and spending weeks manually cleaning up meshes and textures. It routinely costs thousands of dollars per character.

San Francisco-based synthetic data startup Synthesis AI wants to replace all of that with a text prompt. The company just showcased a high-resolution text-to-3D capability inside its online platform, synthesis labs, letting creators generate cinematic-quality 3D digital humans directly from plain text descriptions.

Splitting mesh from texture

Users can type out details like age, gender, ethnicity, hairstyle, and clothing to generate a complete 3D model. Once generated, the character isn't locked down. Creators can modify prompts or use on-screen sliders to tweak facial expressions and lighting on the fly.

The real technical magic lies under the hood. Building 3D assets is notoriously difficult because geometry and surface textures have very different constraints. Rather than attempting a messy end-to-end generation, Synthesis AI built proprietary diffusion-based models specifically to handle 3D meshes governed by demographic parameters. It then uses a separate generative model to build high-fidelity texture layers before fusing them together.

Synthesis AI CEO and founder Yashar Behzadi told VentureBeat that synthetic generation is orders of magnitude faster and cheaper than traditional scanning pipelines, saving thousands of dollars and weeks of production time per asset. The release builds on the company's existing synthesis humans and synthesis scenarios offerings.

While anyone who can use a search engine can generate models with the no-code interface, Behzadi expects early adopters to be technical artists in VFX and gaming, as well as product managers populating industrial simulations who need realistic digital humans without writing ML code.

Why it matters

We think Synthesis AI's smartest move here is decoupling mesh generation from texture layers. Most early text-to-3D models produce warped, low-resolution blobs that look decent from a distance but fall apart in production pipelines. Splitting the structural geometry from the surface material gives creators the fidelity required for serious gaming and VFX work.

There is also a much bigger play beyond entertainment: training data. Because Synthesis AI focuses on synthetic datasets, these 3D humans are fully labeled, highly customizable, and completely free of privacy or licensing headaches. That makes them ideal for training computer vision systems in smart cities, automotive cabin sensing, and virtual try-on platforms.

Generative AI is rapidly graduating from flat 2D pixels to spatial 3D assets. By lowering the technical barrier from high-end photogrammetry rigs down to a simple text prompt and a few sliders, Synthesis AI is putting cinematic digital human production into the hands of teams that could never before afford it.