--- title: Nvidia | Scenario Docs --- > This page is auto-generated from model configurations. Last updated: 2026-11-09. This reference lists all available **Nvidia** 3d generation models and their parameters. Use these parameter names when calling the [Generation API](/api/postgeneratecustom/index.md). - [NVIDIA Kimodo Text to Motion](#nvidia-kimodo-text-to-motion) --- ## NVIDIA Kimodo Text to Motion Generate a 3D skeleton animation from one or more text prompts with NVIDIA Kimodo. Outputs a BVH animation on the SOMA skeleton, plus the raw NPZ data. **Model ID:** `model_nvidia-kimodo-text-to-motion` **Capabilities:** `txt23d` **LLM Markdown:** | Parameter | Type | Required | Default | Min | Max | Allowed Values | Description | | ------------------------- | ------------ | -------- | ----------- | ------- | ---------- | ------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `prompts` | string_array | Yes | - | - | - | - | The motions to generate, played in order with smooth transitions. Start each one with “A person…” and keep it to one or two actions. For example: “A person walks forward”, then “A person sits down”. | | `durations` | number_array | No | `5` | 1 | 10 | - | How long each prompt’s motion lasts, in seconds. Give a single value to use for all prompts, or one value per prompt. | | `constraints` | file | No | - | - | - | - | Optional JSON file of Kimodo constraints: full-body keyframes, hand or foot positions, or a ground path for the character to follow. Set each constraint at a frame number, counted from the start of the whole sequence at 30 frames per second. Positions are Y-up, in meters, relative to where the motion starts. | | `numOutputs` | number | No | `1` | 1 | 4 | - | How many alternative motions to generate for the same prompts. | | `numInferenceSteps` | number | No | `100` | 10 | 200 | - | Number of denoising steps. Fewer steps are faster; more steps give higher quality. | | `guidanceScale` | number | No | `2` | 0 | 10 | - | How closely the motion follows your prompts. | | `guidanceType` | string | No | `separated` | - | - | `nocfg`, `regular`, `separated` | None (`nocfg`) turns guidance off, Regular (`regular`) applies the guidance scale, and Separated (`separated`) applies the guidance scale to the prompts and the constraint guidance scale to the constraints. | | `constraintGuidanceScale` | number | No | `2` | 0 | 10 | - | How closely the motion follows your constraints. Used in Separated guidance mode only. | | `numTransitionFrames` | number | No | `5` | 0 | 30 | - | Number of frames used to blend one prompt’s motion into the next. | | `firstHeadingAngle` | number | No | `0` | -6.2832 | 6.2832 | - | The direction the character faces at the start, in radians around the vertical (Y) axis. 0 faces +Z. | | `postProcessing` | boolean | No | `true` | - | - | - | Reduces foot sliding and snaps the motion onto your constraints. | | `seed` | number | No | - | 0 | 2147483647 | - | Used to reproduce previous results | | `bvhStandardTPose` | boolean | No | `false` | - | - | - | Writes the BVH with a standard T-pose as its rest pose, instead of the skeleton’s default rest pose. |