Bohan Zeng (@bohan_zeng) header
Bohan Zeng (@bohan_zeng) profile picture

Ph.D. student @PKU1898, Data Centric AI, World Model.

Joined October 2022
Photo by Bohan Zeng (@bohan_zeng)Photo by Bohan Zeng (@bohan_zeng)Photo by Bohan Zeng (@bohan_zeng)Photo by Bohan Zeng (@bohan_zeng)Photo by Bohan Zeng (@bohan_zeng)Photo by Bohan Zeng (@bohan_zeng)Photo by Bohan Zeng (@bohan_zeng)Photo by Bohan Zeng (@bohan_zeng)Photo by Bohan Zeng (@bohan_zeng)
Introducing 3AGameFactory: github.com/OpenDCAI/GameFact… — a skill and reference code framework designed for 3A game development. It covers image, 3D asset, motion, audio, and CG generation tools, along with reference code for UE5, Blender, Unity3D, and three.js. We welcome your stars!
1
29
1
167
9,858
Bohan Zeng retweeted
这下可好了 独立开发者也能做3A大作了 OpenDCAI开源 了3A 游戏生成 Skill 与资产框架 github.com/OpenDCAI/GameFact… 覆盖了图片、3D 资产、动作、音频与 CG 视频生成,并支持使用 UE5、Blender、Unity、Godot 4 和 three.js 构建游戏。 赶紧立刻马上拿下
如果 你做独立游戏或演示 需要快速搭出一套看起来统一的 3D 场景 现在不用自己从零建模了 有一个开源github项目 github.com/lucas-martinic/po… 提供三千多个低多边形模型和 32 套主题场景包 能直接用在 three.js、Godot、Unity、Blender 里。个人和商用都可以
23
92
3
744
44,411
Bohan Zeng retweeted
Kling shares the secret to controllable video generation A new technical report argues the bottleneck is no longer rendering but semantic control. It outlines Kling-MLLM, a unified multimodal foundation model, and details lessons on captioning, prompt enhancement, reward modeling, and evaluation.
2
9
34
2,934
Bohan Zeng retweeted
👋Hy4 is here, new with vision. Firmly among the leading open models. More productive everywhere: finance, frontend, game, officework, research… Built responsibly, with cyber/biosecurity in mind. Try it out and let us know! 🔗:hy.tencent.ai/research/hy4?l…
279
343
311
3,016
684,434
Bohan Zeng retweeted
4 Real Productivity
186
165
184
2,599
499,112
Bohan Zeng retweeted
New imagination. New stories. New ways to create. ✨ See what creators are making with Kling 4.0 and Kling 4.0 Flash.
38
27
8
275
501,161
Congratulations to @Moonwines21! latentOmni has been accepted to NeurIPS 2026. This paper leverages latent reasoning to effectively improve omni reasoning performance. Everyone is welcome to check it out.
🚀Our new work, LatentOmni (NeurIPS 2026), moves beyond text-only CoT by interleaving text with AV latent states. Feature-level supervision anchors reasoning to AV evidence, keeping attention on native sensory inputs. Paper: huggingface.co/papers/2605.2… Code: github.com/yfanDai/LatentOmn…
5
227
Bohan Zeng retweeted
🚀Our new work, LatentOmni (NeurIPS 2026), moves beyond text-only CoT by interleaving text with AV latent states. Feature-level supervision anchors reasoning to AV evidence, keeping attention on native sensory inputs. Paper: huggingface.co/papers/2605.2… Code: github.com/yfanDai/LatentOmn…
4
1
7
474
Bohan Zeng retweeted
RayOrch: lineage-controlled dataflows for foundation-model data prep A distributed pipeline engine that tracks parent-child lineage, batches ready work across inputs, and gathers results in order. Speeds up MinerU by 13.1% vs Ray Data and 29.0% vs Daft.
2
4
1
22
1,845
We introduce Physical Coding: agents represent physical-world state and execution policies as code, verify and recover explicitly, and turn validated traces into reusable artifacts for self-evolving physical intelligence. Our paper: github.com/HexaFuture/Physic…
11
14
4
33
4,299
Claude Ops5 vs GPT6, In fact, there is no difference between the two in terms of the overall lighting, overall layout, and gameplay. However, GPT-6 can leverage its 3D object generation capability to build higher-quality scenes. Our repo: github.com/OpenDCAI/GameFact…
2
323
Bohan Zeng retweeted
Top AI papers this week on @huggingface (April 6-12) - GrandCode: First AI to beat all humans in live competitive programming contests, surpassing Google's Gemini 3 Deep Think - Adam's Law: New framework showing LLMs prefer frequent textual data for prompting and fine-tuning - Video-MME-v2: Next-gen video benchmark exposing huge gaps between AI and human experts - ClawBench: Testing AI agents on real-world tasks like booking flights (Claude 4.6 scores just 33%) - SkillClaw: Collective skill evolution for agentic systems - HY-Embodied-0.5: Embodied foundation models from Tencent for real-world robot agents - InCoder-32B-Thinking: Industrial code world model for chip design and GPU kernels - OpenWorldLib: Unified codebase and definition for advanced world models - Plus: Self-Distilled RLVR and rethinking generalization in reasoning SFT
5
11
2
62
11,508
Bohan Zeng retweeted
Flux-OPD makes the case that open-ended distillation needs contexts that evolve with the student, not a fixed target frozen at the start. Flux-OPD: On-Policy Distillation with Evolving Contexts Paper: arxiv.org/abs/2607.28022
1
9
71
4,158
Bohan Zeng retweeted
🚀 Introducing Flux-OPD, an OPD paradigm that uses evolving contexts as in-training supervision to capture task preferences in open-ended domains. 🧩 It improves video prompt optimization and medical question answering. 📄 arxiv.org/abs/2607.28022 🤗 huggingface.co/papers/2607.2…
1
4
1
6
2,067
Bohan Zeng retweeted
NVIDIA releases Lyra 2.0 on Hugging Face A framework for generating persistent, explorable 3D worlds at scale by solving spatial forgetting and temporal drifting in long-horizon video generation.
3
21
159
23,296
Bohan Zeng retweeted
--Towards Scalable Sidewalk Autonomy (2/4)-- Introducing UrbanVerse — turning city-tour videos into physics-aware, interactive simulations for scalable robot learning in NVIDIA Omniverse. 🌆 Real-world scene layouts 🧱100K+ 3D assets ICLR’26 paper and **Open-source** assets at 🔗 urbanverseproject.github.io/ We thank @CocoRobotics for robot support and NVIDIA Academic Grant support @NVIDIARobotics
3
34
211
15,147
Bohan Zeng retweeted
Genie3 generates videos. We generate 𝟯𝗗 𝘄𝗼𝗿𝗹𝗱𝘀 you can actually use. Launching tomorrow — Tencent #HYWorld 2.0, an engine-ready World Model🚀 This isn't a video. It's a real 3D scene, all generated & editable. One image in. A whole 3D world out. 🔥Open-source tomorrow
198
460
92
4,159
553,660
🌧️❄️🌨️Procedural rain and snow as Gaussian splats — infinite particles that move with the camera, fully rendered on the GPU. Add fog and it starts to feel like actual weather @playcanvas
16
60
5
647
56,548