Workflow
3D生成大模型
icon
Search documents
字节跳动推出3D生成大模型Seed3D 1.0 实现一张图生成仿真级3D模型
智通财经网· 2025-10-23 08:32
Core Insights - ByteDance's Seed team launched Seed3D 1.0, a 3D generative model that creates high-quality simulation-level 3D models from single images, utilizing an innovative Diffusion Transformer architecture [1][2] - Seed3D 1.0 demonstrates superior performance in texture and material generation compared to existing models, achieving industry-leading capabilities [1][2] Model Performance - Comprehensive evaluations showed that Seed3D 1.0 outperforms other methods in detail retention and structural integrity, with 1.5 billion parameters surpassing industry models with 3 billion parameters [2] - The model's stepwise generation strategy allows for the expansion from single object generation to creating complete, coherent 3D scenes [2] Future Developments - The Seed team plans to integrate multimodal large language models (MLLM) to enhance the quality and robustness of 3D generation, aiming for large-scale applications in world simulators [2]
字节跳动推出3D生成大模型Seed3D 1.0
Core Insights - ByteDance's Seed team announced the launch of Seed3D 1.0, a 3D generative model that enables end-to-end generation of high-quality simulation-level 3D models from a single image [1] Group 1: Product Features - Seed3D 1.0 is based on an innovative Diffusion Transformer architecture [1] - The model has been trained on large-scale data, allowing it to generate complete 3D models with fine geometry, realistic textures, and physically-based rendering (PBR) materials [1]
字节跳动Seed3D 1.0 发布:一张图生成高精度3D模型
Feng Huang Wang· 2025-10-23 05:21
Core Viewpoint - ByteDance's Seed team has launched a 3D generative model, Seed3D 1.0, which enables end-to-end generation of high-quality simulation-level 3D models from a single image [1] Group 1: Product Features - Seed3D 1.0 demonstrates superior texture and material generation capabilities compared to existing open-source and closed-source models [1] - The geometric generation performance of Seed3D 1.0 surpasses that of larger parameter-scale models in the industry, indicating a significant advancement in 3D modeling technology [1] - Overall, Seed3D 1.0 achieves a leading position in the industry in terms of comprehensive capabilities [1]
腾讯开源混元3D 2.1大模型 开发门槛大大降低
Guang Zhou Ri Bao· 2025-06-14 14:17
Core Viewpoint - Tencent's Mix Yuan 3D model has undergone a significant upgrade with the release of version 2.1, which is the first fully open-source industrial-grade 3D generation model, achieving industry-leading standards [2]. Group 1: Model Features and Performance - The Mix Yuan 3D model has surpassed 1.8 million downloads on Hugging Face, indicating its popularity and utility in the open-source community [2]. - The new version 2.1 has improved geometric generation quality and introduced PBR (Physically Based Rendering) material generation, enhancing the texture and lighting effects of 3D assets, moving away from a "plastic feel" [6][8]. - In user blind tests, the PBR textures of the Mix Yuan 3D 2.1 model outperformed RGB textures with a quality superiority rate of 78% [8]. Group 2: Accessibility and Customization - The Mix Yuan 3D 2.1 model has lowered development barriers by providing a fully open-source solution, including model weights, training code, and data processing workflows, allowing developers to fine-tune, retrain, or optimize the model for various customization needs [8]. - The model is compatible with consumer-grade graphics cards, enabling it to run on personal computers, thus broadening its accessibility for developers and researchers [8]. Group 3: Technical Enhancements - The new model optimizes detail modeling, resulting in higher mesh precision and better topological consistency, which lays the groundwork for subsequent texture mapping [8]. - It supports the generation of base color, metallicity, normal, and roughness maps, allowing for high-quality rendering of various complex materials such as leather, wood, metal, and ceramics [8].
影眸科技获美团龙珠、字节领投数千万美元A轮融资,专注3D生成大模型
IPO早知道· 2025-01-13 03:55
45天破百万美金ARR。 本文为IPO早知道原创 作者|Stone Jin 据IPO早知道消息,3D生成大模型公司影眸科技完成数千万美元 A 轮融资。本轮融资由美团龙珠、 字节跳动领投,老股东红杉中国种子基金及奇绩创坛持续跟投,光源资本担任独家财务顾问。 此次融资将用于支持影眸科技进一步推进3D大模型的前沿探索,保持公司在3D大模型领域的全球领 先位置。同时推动以3D生成大模型Rodin为核心的Hyper3D系列产品在全球市场的商业化落地。 值得注意的是, Rodin上线45天后即实现100万美金年经常性收入(ARR),并持续高速增长。此 外,影眸已与 Amazon、字节跳动、Unity、拓竹等全球知名客户建立深度合作,截至目前已获得数 百万元B端商业化订单,覆盖游戏、电商、具身智能、空间计算、3D打印、影视等多个行业 。 影眸科技孵化于上海科技大学,由吴迪、张启煊、张龙文、曾初啸等人于2020年创立,团队平均年 龄 24 岁 。 其 团 队 与 上 科 大 共 同 提 出 的 可 控 3D 原 生 DiT 生 成 框 架 CLAY 与 3D 服 装 生 成 框 架 DressCode,均获计算机图形学顶会A ...