Lucas

NVIDIA built an AI model that generates 3D character animations from text, and you can now use it through Scenario. 🎮 It's called Kimodo, a motion-generation model developed by @NVIDIA researchers using more than 700 hours of optical motion-capture data. Instead of generating video frames, Kimodo produces actual skeleton motion that can be transferred to 3D characters. The model was designed for both character animation and humanoid robotics. 🧠 What makes it interesting? Kimodo doesn't just turn a prompt into a random sequence of poses. You can also control how the character moves. ▪️ Text-to-motion: Describe walking, running, dancing, jumping, gestures, or object interactions. ▪️ Keyframe constraints: Specify particular poses at certain moments and let the model generate motion between them. ▪️ Hand and foot constraints: Control where the character's hands or feet should be positioned. ▪️ Movement paths: Draw waypoints or a continuous route that the character should follow. ▪️ Prompt sequences: Combine multiple actions into a longer animation. For example, you could describe: A character walks toward a table, picks up an object, turns around, and runs away. Then provide positional constraints to control the character's route. ⚙️ How does it work? Kimodo uses a motion diffusion architecture with two prediction stages. The first handles the character's global movement through space. The second generates the corresponding body poses. Separating those problems helps reduce common animation artifacts such as sliding feet and unnatural root movement. 🎬 Now available in Scenario Scenario added Kimodo to its model catalog on November 6. The integration supports: ▪️ Motion clips up to approximately 10 seconds per prompt. ▪️ 30 FPS skeletal animation. ▪️ Multiple prompts within a sequence. ▪️ BVH exports using the 77-joint SOMA skeleton. ▪️ NPZ exports for additional processing. But there's an important limitation. Kimodo generates motion, not a finished 3D character. You'll still need a character mesh, textures, rigging, and retargeting to your target skeleton. The model also works with a simplified internal skeleton, so fine finger animation isn't automatically guaranteed. 🔓 Can you run it yourself? NVIDIA has published Kimodo's research code, model weights, and a local motion-authoring demonstration. The Scenario integration offers a more convenient hosted workflow, with usage subject to its credit system. Why this matters Character animation is one of the more time-consuming stages of 3D production. Kimodo points toward a workflow where developers describe an action, generate a motion draft, constrain the movement, and refine it in their usual animation tools. And the same technology could help generate training demonstrations for humanoid robots. 🔗 Research & demos: research.nvidia.com/labs/sil… 🔗 Scenario: scenario.com/ #NVIDIA #GameDevelopment #AIAnimation #Robotics #3DArt