What's this about?
Tencent's Hunyuan3D-Buffalo is one multimodal model spanning 3D generation, understanding, and editing — tackling 3D's core bottleneck, the scarcity of large-scale, geometrically consistent editing data (40↑).
What exactly is "unified" multimodal modeling?
One model that both understands and generates within a domain, instead of separate specialized models bolted together. For 3D that means the same system can create an asset, reason about it, and edit it — mirroring the unified image models that already merged generation and understanding.
Why does it matter?
3D has lagged image and video because good training data is scarce and geometry is unforgiving. A unified model that also edits is a step toward practical 3D content pipelines — relevant to the same game/asset-generation demos (otter mech, Van Gogh builder) trending this week.