# AI motion capture and animation for indie devs (October 2026): Rokoko Vision 3.0, Move One, MetaHuman Animator, Cascadeur, Kimodo and Viggle

> You no longer need a suit or a capture stage to animate a character. What each AI mocap and animation tool costs, what it needs (camera, phone, GPU), and how it gets into Unreal, Unity or Blender, checked October 18, 2026.

Published 2026-10-19T04:24:46-07:00 · AI for 3D · Canonical: https://news.littlewolflabs.com/posts/2026-10-18-ai-mocap-compared

Animation is the step builders keep naming as the bottleneck once AI can make the meshes. In 2026 that started to change: you can film yourself with a phone or webcam and get a usable skeleton animation, or type a prompt and get motion back. The tools split into three kinds: **video-to-motion** (film a person, get animation), **AI-assisted keyframing** (you pose, the software fills in), and **text-to-motion** (describe it, get it). We checked prices and requirements on official pages on October 18, 2026. We have not run our own capture test.

## At a glance

| | Kind | Price (USD) | What you need | Runs | Exports |
|---|---|---|---|---|---|
| [Rokoko Vision 3.0](https://www.rokoko.com/products/vision) | Video → motion | Free: 30 s a month. Basic $10/mo annual ($12 monthly): 600 s ([pricing](https://www.rokoko.com/pricing)) | Any single camera; video file upload | Cloud | FBX (all plans); BVH on paid plans |
| [Move One](https://docs.move.ai/knowledge/move-one-pricing) (Move AI) | Video → motion | Free: 30 one-time credits. Starter $18/mo + VAT for 60 credits; up to $490/mo | iPhone app | Cloud | FBX, USDC, USDZ, .blend (Blender 4+) ([Move docs](https://docs.move.ai/knowledge/which-file-formats-are-available-for-exporting)) |
| [MetaHuman Animator markerless plugin](https://www.fab.com/listings/4095b8e0-3eff-44f1-acb4-cb40b99228b9) (Epic) | Video → motion (body, hands, face) | **Free** on Fab | UE 5.8+, Windows, DirectX 12 GPU; webcam or phone video | Your PC | Unreal animation sequence |
| [Cascadeur 2026.2](https://cascadeur.com/plans) | AI-assisted keyframing, AI inbetweening, motion generation | Free (non-commercial, no FBX). Indie $19/mo or $8/mo annual (under $100k revenue). Pro $49/mo or $33/mo annual | Windows, macOS or Linux | Your PC | CASC only on Free; FBX, DAE, USD, glTF on paid |
| [NVIDIA Kimodo](https://github.com/nv-tlabs/kimodo) | Text/constraints → motion | **Free**: code Apache-2.0; most models under the NVIDIA Open Model License | NVIDIA GPU: ~17 GB VRAM fully on GPU, under 3 GB with text encoding on the CPU | Your PC | Custom NPZ (SOMA skeleton); AMASS NPZ (SMPL-X); MuJoCo CSV (G1 robot) |
| [Viggle PINOC](https://viggle.ai/pinoc/mcp) | Photo → rigged character; video → motion; agent MCP | From $10.39/mo ([pricing](https://viggle.ai/pricing)) | Browser | Cloud | GLB, FBX, USDZ ([PINOC MCP page](https://viggle.ai/pinoc/mcp)) |
| [FreeMoCap](https://github.com/freemocap/freemocap) | Video → motion | **Free**, AGPL-3.0 | Single or multiple webcams, GoPros or phones; 2–3+ cameras plus a printed ChArUco board for 3D calibration | Your PC | Blender scene (.blend), CSV, NumPy; site also lists FBX |

## Video-to-motion: film yourself

**Rokoko Vision 3.0.** Rokoko rebuilt its single-camera solver and moved it into its web app, Rokoko Create, on [July 16, 2026](https://www.rokoko.com/insights/introducing-rokoko-vision-3-0). You record with any camera, upload the file, and it processes in the cloud, so no strong GPU is needed. Rokoko [retired](https://support.rokoko.com/hc/en-us/articles/48834414923793-Rokoko-Studio-2-6-2-Vision-3-0-Support) dual-camera and live webcam capture on August 16 and says the new single-camera solver beats the old dual-camera one. There's no finger tracking yet ([product FAQ](https://www.rokoko.com/products/vision)). You then edit in Rokoko Studio and export FBX with skeleton presets for Mixamo, HumanIK and Unreal's Manny; Studio has live plugins for Unreal, Unity and Blender. Free gets 30 seconds of video processing a month. Basic is $10 a month billed annually for 600 seconds, Plus $20 for 3,000 seconds, Pro $50 for 15,000 seconds ([pricing](https://www.rokoko.com/pricing)).

**Move One (Move AI).** An iPhone app that turns single-camera video into 3D motion. Plans: Starter $18 a month for 60 credits, Standard $48 for 180, Plus $225 for 240, Advanced $490 for 700, all plus VAT ([Move docs](https://docs.move.ai/knowledge/move-one-pricing)). One credit covers one person for one second on the Gen 1 model; Gen 2 costs 2 credits per second. So Starter covers 60 seconds of Gen 1 or 30 seconds of Gen 2 a month. Free users get 30 one-time credits and 30-second recordings; paid users can record 60 seconds. Move's docs list exports as FBX, USDC, USDZ and .blend (Blender 4 or newer) ([Move docs](https://docs.move.ai/knowledge/which-file-formats-are-available-for-exporting)), and Move publishes guides for retargeting in Unreal (including MetaHuman rigs), Unity, Blender, Maya and C4D ([Move docs](https://docs.move.ai/knowledge/move-one)). Overage rates per credit were not verified as of Oct 18, 2026 (move.ai/pricing loads them with JavaScript we couldn't read).

**MetaHuman Animator markerless plugin (Epic).** Since [UE 5.8](https://www.metahuman.com/news/metahuman-5-8-is-now-available) (June 23), MetaHuman Animator can capture face, body, or both from one off-actor camera. Epic says "a webcam is enough". The tech comes from Meshcapade. The plugin is a **free** download from [Fab](https://www.fab.com/listings/4095b8e0-3eff-44f1-acb4-cb40b99228b9), runs entirely on your PC ("no cloud, no credits"), and captures hands too. Limits: it is **Experimental**, **Windows only**, and needs UE 5.8+ with a DirectX 12 GPU ([Epic docs](https://dev.epicgames.com/documentation/en-us/metahuman/getting-started-with-metahuman-animator); [Digital Production](https://digitalproduction.com/2026/06/23/metahuman-gets-markerless-mocap/)). Epic's own [hardware page](https://dev.epicgames.com/documentation/metahuman/metahuman-hardware-requirements-in-unreal-engine?lang=en-US) recommends, for MetaHuman Animator, "at least 16 physical, high performance cores with 32GB RAM" and "At least nVIDIA RTX 3070, AMD RX 6800 XT, or Apple M2 Ultra, with 8GB VRAM", and lists the markerless plugin as Windows-only. A third-party guide by [CG Dive](https://addons.cgdive.com/tools/unreal-engine-markerless-mocap) reported about 20–25 minutes of processing for 30 seconds of footage and 15–20 GB of disk for the engine data and plugin. One forum user saw an extra 3 GB of VRAM used on a 10-second clip ([Epic forums](https://forums.unrealengine.com/t/epic-games-metahuman-animator-markerless-motion-capture-plugin/2729293?page=2)). Output lands as an Unreal animation sequence on the MetaHuman skeleton, which you can retarget. Getting it into Unity or Blender means exporting from Unreal (we did not test that path).

**FreeMoCap (free, open source).** A community project, AGPL-3.0, about 10,500 GitHub stars, last pushed October 17 ([repo](https://github.com/freemocap/freemocap)). Its docs say it works from single cameras, multiple cameras or imported videos, tested mostly on cheap webcams. For 3D capture it recommends two or more (ideally three) cameras 40–60° apart, calibrated with a printed ChArUco board ([calibration guide](https://docs.freemocap.org/documentation/multi-camera-calibration.html)). It outputs NumPy arrays, CSVs and a Blender scene ([docs](https://docs.freemocap.org/documentation/about-us.html)); the [project site](https://freemocap.org/) also lists FBX. BVH export is not documented. Some builders also use research models such as GVHMR to pull motion from phone video ([@AlecSaysThings, Sep 14](https://x.com/AlecSaysThings/status/2099369531152183755)); GVHMR's licence is not a standard open-source one (GitHub reports no recognised licence), so check before any commercial use.

## AI-assisted keyframing: Cascadeur 2026.2

[Cascadeur](https://cascadeur.com/plans) is animation software where you set a few key poses and AI tools fill in the rest: AutoPosing, AutoPhysics, AI inbetweening and Motion Generation. Version 2026.2 ([CG Channel, Aug 6](https://www.cgchannel.com/2026/08/nekki-releases-cascadeur-2026-2-with-animation-layers/)) added animation layers (alpha), easing controls, BVH import for cleaning up mocap, retrained AI models, and a "simple" MCP server so an assistant like Claude can drive it.

Pricing from the [plans page](https://cascadeur.com/plans):
- **Free:** full animation toolset, but saves only in Cascadeur's own .casc format and is **non-commercial**. You can't get animation out to Unreal, Unity or Blender.
- **Indie:** $19 a month, or $8 a month billed annually, for revenue under $100,000 a year. Exports CASC, FBX, DAE, USD and glTF.
- **Pro:** $49 a month, or $33 a month billed annually. Unlimited commercial use, plus retargeting and quadruped autoposing.
- Yearly Indie and Pro plans include perpetual-licence builds.

It runs on Windows 10+, Ubuntu 20.04+ and macOS 13.3+ (CG Channel).

## Text-to-motion: Kimodo and others

**NVIDIA Kimodo.** An open motion-generation model trained on 700 hours of commercially licensed optical mocap ([repo](https://github.com/nv-tlabs/kimodo)). You control it with text prompts plus constraints: full-body keyframes, hand or foot positions, and 2D paths or waypoints. The code is Apache-2.0. The recommended models use the [NVIDIA Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/); the SMPL-X variant uses a research-only licence. The newest models (v1.2) shipped October 15, 2026. NVIDIA's README says Kimodo "requires ~17GB of VRAM to generate locally entirely on GPU", mainly for the text model; with `TEXT_ENCODER_DEVICE=cpu` that drops to "<3 GB", a bit slower. It was tested most on RTX 3090, RTX 4090 and A100. Output is a custom NPZ file for the default SOMA skeleton, AMASS NPZ for SMPL-X, and MuJoCo CSV for the G1 robot; there is no FBX or BVH exporter in the README, so getting motion into an engine means a converter or a plugin. A community Unreal plugin, DDS Motion (**Unofficial**), claims Kimodo runs locally "with a minimum of only 6GB VRAM" ([@DarkDojoStudios, Aug 21](https://x.com/DarkDojoStudios/status/2090894641814020490)); not verified as of Oct 18, 2026. NVIDIA's [June 16 post](https://developer.nvidia.com/blog/build-on-device-ai-companions-with-the-nvidia-ace-game-agent-sdk-and-unreal-engine-5-plugins/) also points to an Animotive Kimodo plugin for Unreal.

**Rokoko text-to-motion.** Rokoko Create also generates clips from text. Generating is unlimited on every plan; importing them into Studio is capped (5 a month free, 100 on Basic) ([pricing](https://www.rokoko.com/pricing)).

**Viggle PINOC.** Viggle is best known for swapping characters into video (Viggle AI plans from $7.99 a month). Its separate PINOC product makes rigged 3D characters from photos and exports motion as FBX or GLB, from $10.39 a month ([pricing](https://viggle.ai/pricing)). Its [MCP page](https://viggle.ai/pinoc/mcp) lets agents generate rigged characters and motion. The API is pay as you go at $0.01 per credit, with 100 free credits for new accounts.

## Quality: what we can and can't say

No independent lab has published a side-by-side quality test of these tools that we could find, and we have not run one. Vendors make their own claims: Rokoko says Vision 3.0 has "drastically reduced joint noise" ([Rokoko support](https://support.rokoko.com/hc/en-us/articles/43395990943377-Introducing-Rokoko-Create)). NVIDIA publishes a [Kimodo motion-generation benchmark](https://huggingface.co/datasets/nvidia/Kimodo-Motion-Gen-Benchmark) for comparing text-to-motion models, but it doesn't cover video capture tools.

The closest to a comparison is one developer's informal series on X:
- "Comparison of 4 alternative AI (humanoid) animation pipelines": Opus 5.5 driving Blender, Kimodo, Cascadeur, and a local "neural video mocap" — [@nazimadakli, Oct 5](https://x.com/nazimadakli/status/2107159129420964343)
- Follow-up: "This time in cascadeur I enabled AI in-betweening. It did not work very well. Without in-betweening cascadeur is nearly equal to blender" — [@nazimadakli, Oct 16](https://x.com/nazimadakli/status/2111160333217800466)

Other reports (individual experiences, not tests):
- On MetaHuman markerless: "you don't even need two cameras nor calibration … Tried it, it is amazingly good." — [@fhajji68, Aug 25](https://x.com/fhajji68/status/2092297713492938838)
- Cascadeur's mocap cleanup: "So they now have a button to just clean raw MoCap data clipping" — [@TheRoyalSkies, Sep 9](https://x.com/TheRoyalSkies/status/2097700468260515890) (the poster also shares a promo code, so treat it as promotional)
- A mixed pipeline: "Bones-seed, Kimodo, video-model/mocap -> gvhmr -> retargeting, and a little bit of paying talented animators." — [@AlecSaysThings, Oct 9](https://x.com/AlecSaysThings/status/2108623892956827903)

## Getting it into your engine

| | Unreal | Unity | Blender |
|---|---|---|---|
| Rokoko | FBX with Manny preset; live plugin | FBX; live plugin | FBX/BVH; live plugin |
| Move One | FBX; retarget guides incl. MetaHuman | FBX; retarget guide | FBX or .blend; Move One Blender plugin |
| MetaHuman markerless | Native (runs inside Unreal) | Via export from Unreal (not tested) | Via export from Unreal (not tested) |
| Cascadeur (paid) | FBX/USD/glTF | FBX/USD/glTF | FBX/USD/glTF |
| Kimodo | Community plugins (**Unofficial**): DDS Motion, Animotive | No exporter in the README | No exporter in the README |
| Viggle PINOC | FBX/GLB | FBX/GLB | FBX/GLB |

## What to weigh

- **How you'll make motion.** Acting it out (Rokoko, Move One, MetaHuman, FreeMoCap), posing it by hand with AI help (Cascadeur), or describing it (Kimodo, Rokoko text-to-motion, PINOC).
- **Your engine.** MetaHuman markerless is free and local but lives inside Unreal on Windows. Rokoko and Cascadeur export to anything.
- **Seconds per dollar.** Cloud capture is metered: Rokoko's free 30 seconds a month vs 600 seconds for $10 a month; Move One's Starter covers 30–60 seconds a month depending on the model.
- **Hardware.** Cloud tools need only a camera and a browser. MetaHuman wants a strong Windows PC (Epic recommends 16 cores, 32 GB RAM and an 8 GB GPU). Kimodo needs an NVIDIA GPU (about 17 GB, or under 3 GB with CPU text encoding).
- **Commercial rights.** Cascadeur Free can't export and is non-commercial. Kimodo's recommended models use NVIDIA's Open Model License, but one variant is research-only. FreeMoCap is AGPL. Read the licence for the exact model you use.
- **Fingers and faces.** Rokoko Vision 3.0 has no finger tracking. MetaHuman captures hands and face. Finger and face tracking in the other tools not verified as of Oct 18, 2026.
- **Maturity.** MetaHuman's plugin is Experimental, Cascadeur's layers are alpha, and Kimodo is research code. Test on a short clip first.