Lux3D generates 3D assets from text or images, with Turbo Mode producing models in as little as 20 seconds

HANGZHOU, China, Aug. 27, 2026 /PRNewswire/ -- Manycore Tech Inc. ("Manycore") (HKEX: 00068) today announced the launch of Lux3D, its proprietary 3D generative model. The company also outlined the next phase of its roadmap for developing world models built on explicit 3D representations.

The launch follows strong growth across Manycore's AI portfolio. In the first half of 2026, revenue from new AI applications and products increased by 177% year over year. Adjusted net profit, a non-IFRS measure, rose by 211% to RMB55.42 million, approaching the company's adjusted net profit for the full year of 2025, according to its 2026 interim results announcement.

Supporting Multiple 3D Representations, with an Emphasis on Material Fidelity

Lux3D generates 3D assets from text prompts or reference images. It offers two modes: Standard Mode for high-precision geometry and material generation, and Turbo Mode for generation in as little as 20 seconds and cost-efficient batch creation through Harness Mode.

Material generation is a core focus of Lux3D. The model reproduces how materials such as glass, metal, ceramic and plastic respond to light, helping assets remain visually consistent across different viewpoints and rendering conditions.

Lux3D was trained on a combination of generated and reconstructed physical-world data to improve geometric and visual consistency. The model delivered competitive results in internal evaluations, with detailed benchmarks to be released separately.

Addressing the Cost and Speed of 3D Content Production

The time and computing cost required to generate individual 3D assets have limited the adoption of generative models in workflows such as game development and embodied AI training. Lux3D's Turbo Mode addresses this challenge with faster, more cost-efficient generation for large-scale asset creation.

"3D Gaussian technology has been an important area of exploration for Manycore," said Tianze Long, Senior Director of Manycore's AI business. "Its combination of rendering fidelity and computational efficiency makes it particularly relevant to applications that require rapid previews or large-scale asset generation."

Manycore has also opened API access to Lux3D. Through Harness Mode, developers can generate assets in batches using tools such as OpenAI Codex and refine them in Blender, reducing repetitive manual work across e-commerce, industrial design, game development, XR, film production and simulation.

From Generated Assets to Interactive Worlds

Manycore said Lux3D is part of its broader world model roadmap. The company is testing a next-generation model that combines SpatialGen with Lux3D to generate interactive 3D environments with greater consistency, editability and dynamic interaction. Potential applications include AI-generated video, game development, embodied AI training and digital twins.

Manycore also plans to add dynamic asset generation to Lux3D, enabling generated objects to include component relationships, motion logic and interactive properties. The goal is to move beyond static models toward assets that can act and respond within simulations and virtual environments.

"World models are beginning to follow different technical paths. Manycore has consistently pursued an explicit 3D approach—one that generates not just pixels, but three-dimensional worlds," said Dr. Zihan Zhou, Chief Scientist at Manycore Tech.

"Our starting point is to align digital and physical worlds across more dimensions, enabling digital environments to incorporate more signals from the physical world," Zhou added. "Ultimately, this can help AI systems trained in digital environments act more effectively in the physical world."

To learn more and explore Lux3D, visit the Lux3D product page:
https://lux3d.aholo3d.com/