Home
Hyper3D Rodin Gen-2.5 Generates Million Faces in 4 Seconds, First 10M-Face 3D Model with Production-Level Detail
The 3D generation AI space has reached another milestone. Hyper3D (Deemos Technology) has unveiled Rodin Gen-2.5, positioning it as the most advanced 3D generation model available today. It can produce a million-polygon model in just 4 seconds and is the world's first 3D generation system to reach 10 million polygons, capturing details as fine as pores and skin microstructures. This marks a significant shift for AI 3D from merely 'viewable' to truly 'usable'.

Speed and Scale Reimagined: Adaptive Thinking Takes Center Stage
Rodin Gen-2.5 features an **Adaptive Thinking Effort** mechanism, inspired by large language models, that dynamically assigns computing power based on task complexity. It offers several generation modes:
Very Low Mode: Generates in 4 seconds, ideal for rapid prototyping and batch testing of simple assetsLow Mode: Generates in 9 seconds, suited for minimalist models and small hard-surface objectsMiddle Mode: Generates in 20 seconds, offering a balanced mix of structure and detailHigh Mode: Generates in 40 seconds, delivering high-quality assets with rich structural layers and smooth finishesVery High Mode: Generates in 80 seconds, producing professional-grade assets with microscopic precisionAIbase notes that this tiered approach greatly enhances real-world production efficiency, letting users select the right mode for each task—whether iterating quickly on concepts or generating high-fidelity final assets.
Native 3D Textures: Full 360° Coverage with No Blurry Spots
Unlike the prevalent 2D-to-3D approach in the industry, Rodin Gen-2.5 uses a native 3D generation architecture. A single reference image is enough to produce a fully textured model with PBR materials:
Textures are created directly in 3D space, delivering complete 360° coverage with no distortion or blurring on any surface. Tightly aligned with geometry, fine details such as fabric texture and seams are rendered naturally. Offers Faithful Mode (closely follows the reference) and Creative Mode (intelligently refines structure, e.g., correcting tires to perfect circles). At maximum precision, users can toggle between Micro (pore-level microscopic detail) and Clean (smooth geometry for stylized or animation work).The model also supports generating up to 10 outputs simultaneously, making it easy to explore multiple creative directions in one batch.
Homegrown Team Committed to Native 3D: SIGGRAPH 2025 Best Paper Award
The team behind Rodin, **Deemos Technology**, is a homegrown group dedicated to 3D generation. Since 2016, they have been deeply immersed in this field. While most of the industry took the 2D-to-3D shortcut, they opted for a more demanding yet comprehensive native 3D approach, successfully addressing persistent problems like broken faces and messy topology.
This year, the team received the SIGGRAPH 2025 Best Paper Award—among commercial companies, only Google and Meta were also recognized—underscoring their technical expertise.
AIbase observes that with Rodin Gen-2.5, AI 3D generation is moving rapidly toward production-ready quality. Its potential spans game development, e-commerce, AR/VR, and 3D printing. Going forward, the barrier to 3D content creation is expected to drop substantially, giving professional creators and developers more efficient tools for their workflows.
Related article
5 Days Left to Exhibit at TechCrunch Disrupt 2026
Time is running out. Secure your exhibit table for TechCrunch Disrupt 2026 before the deadline: September 18 at 11:59 p.m. PT.What is Disrupt?TechCrunch Disrupt 2026 is scheduled for October 13–15 at Moscone West in San Francisco, uniting over 10,000
CodeBuddy and WorkBuddy Extend Free Trial of Hy3 Model Until August 31st
The limited-time free trial for the Hy3 model, offered by CodeBuddy and WorkBuddy, has been extended until August 31, 2026. Responding to overwhelming user interest, the platform has decided to maintain free access, giving more users the chance to ex
Tongyi Lab Unveils Qwen3.7-Max with Orthogonal Decoupling Tech, Tops Multiple Domestic Benchmarks
Tongyi Lab has officially launched the next-generation AI Agent foundation model, Qwen3.7-Max, which tops multiple authoritative benchmarks and addresses key industry challenges like the fragility and instability of intelligent agents in long-chain p
Related Special Topic Recommendations
Comments (0)
0/500
The 3D generation AI space has reached another milestone. Hyper3D (Deemos Technology) has unveiled Rodin Gen-2.5, positioning it as the most advanced 3D generation model available today. It can produce a million-polygon model in just 4 seconds and is the world's first 3D generation system to reach 10 million polygons, capturing details as fine as pores and skin microstructures. This marks a significant shift for AI 3D from merely 'viewable' to truly 'usable'.

Speed and Scale Reimagined: Adaptive Thinking Takes Center Stage
Rodin Gen-2.5 features an **Adaptive Thinking Effort** mechanism, inspired by large language models, that dynamically assigns computing power based on task complexity. It offers several generation modes:
Very Low Mode: Generates in 4 seconds, ideal for rapid prototyping and batch testing of simple assetsLow Mode: Generates in 9 seconds, suited for minimalist models and small hard-surface objectsMiddle Mode: Generates in 20 seconds, offering a balanced mix of structure and detailHigh Mode: Generates in 40 seconds, delivering high-quality assets with rich structural layers and smooth finishesVery High Mode: Generates in 80 seconds, producing professional-grade assets with microscopic precisionAIbase notes that this tiered approach greatly enhances real-world production efficiency, letting users select the right mode for each task—whether iterating quickly on concepts or generating high-fidelity final assets.
Native 3D Textures: Full 360° Coverage with No Blurry Spots
Unlike the prevalent 2D-to-3D approach in the industry, Rodin Gen-2.5 uses a native 3D generation architecture. A single reference image is enough to produce a fully textured model with PBR materials:
Textures are created directly in 3D space, delivering complete 360° coverage with no distortion or blurring on any surface. Tightly aligned with geometry, fine details such as fabric texture and seams are rendered naturally. Offers Faithful Mode (closely follows the reference) and Creative Mode (intelligently refines structure, e.g., correcting tires to perfect circles). At maximum precision, users can toggle between Micro (pore-level microscopic detail) and Clean (smooth geometry for stylized or animation work).The model also supports generating up to 10 outputs simultaneously, making it easy to explore multiple creative directions in one batch.
Homegrown Team Committed to Native 3D: SIGGRAPH 2025 Best Paper Award
The team behind Rodin, **Deemos Technology**, is a homegrown group dedicated to 3D generation. Since 2016, they have been deeply immersed in this field. While most of the industry took the 2D-to-3D shortcut, they opted for a more demanding yet comprehensive native 3D approach, successfully addressing persistent problems like broken faces and messy topology.
This year, the team received the SIGGRAPH 2025 Best Paper Award—among commercial companies, only Google and Meta were also recognized—underscoring their technical expertise.
AIbase observes that with Rodin Gen-2.5, AI 3D generation is moving rapidly toward production-ready quality. Its potential spans game development, e-commerce, AR/VR, and 3D printing. Going forward, the barrier to 3D content creation is expected to drop substantially, giving professional creators and developers more efficient tools for their workflows.
5 Days Left to Exhibit at TechCrunch Disrupt 2026
Time is running out. Secure your exhibit table for TechCrunch Disrupt 2026 before the deadline: September 18 at 11:59 p.m. PT.What is Disrupt?TechCrunch Disrupt 2026 is scheduled for October 13–15 at Moscone West in San Francisco, uniting over 10,000
CodeBuddy and WorkBuddy Extend Free Trial of Hy3 Model Until August 31st
The limited-time free trial for the Hy3 model, offered by CodeBuddy and WorkBuddy, has been extended until August 31, 2026. Responding to overwhelming user interest, the platform has decided to maintain free access, giving more users the chance to ex
Tongyi Lab Unveils Qwen3.7-Max with Orthogonal Decoupling Tech, Tops Multiple Domestic Benchmarks
Tongyi Lab has officially launched the next-generation AI Agent foundation model, Qwen3.7-Max, which tops multiple authoritative benchmarks and addresses key industry challenges like the fragility and instability of intelligent agents in long-chain p











