# Kapnetix, full reference Last generated from the site source. Canonical pages are linked from each section. Every figure here carries the condition that makes it true; figures without their conditions are not usable statements about this product. ## Kapnetix Post https://www.kapnetix.ai/ai-automated-mocap-cleanup Kapnetix Post is the AI front end for motion-capture post-production. Upload a shoot day and it diagnoses every take, predicts how cleanly each one can be repaired, cleans the ones worth cleaning, and hands back a verified result in your own file format. Your raw capture is never modified: every cleanup lands as a new version you can inspect, compare and revert. ### The passage this page has always led with Motion capture is the process of recording the movement of complex objects, usually people. While most people would associate it with filmmaking or animations in video games, it also has its military, sports, and medical applications. Capturing movement is hard. While most of the market uses optical tracking, relying mostly on its high accuracy, there are also inertial systems that use sensors or solutions that are looking at tracking motion just from 2D images. Each system has its advantages, but most require heavy post-processing and corrections. While it's not possible to eliminate the creation of these errors, Kapnetix aims to automate the correction tasks so that people don't have to do it. Our approach dramatically speeds up the overall time to deliver great quality MoCap. ### Capabilities - No need to change your existing pipeline. Kapnetix is a plug & play cloud service. It accepts common formats like FBX or C3D and returns cleaned mocap to the same place it originated. It also runs on-prem or air-gapped, inside your own network, when policy says nothing leaves the building. - Hours of boring cleanup completed in minutes. Our goal is to help animators focus on the creative process and not imperfections of the optical mocap process. - We can personalize through re-training. Creativity has no boundaries and our solution has to be ready for motion that can be very unique. - We can fix complex multicharacter scenes. Multiple characters and de-noising are supported today. Props are not. ### The run, moment by moment 1. Upload. C3D or FBX, from shared storage or a watched folder, a take or a whole shoot day. Format, capture system, actor count and duration are read from the files. Raw stays immutable and every run writes a new version. 2. Triage. Every take is diagnosed on arrival and sorted worst-first in four states: clean, high confidence, low confidence, recommended manual cleanup. A per-frame heat timeline and a 3D viewer show where the trouble is, per actor, with flagged regions for gaps, marker swaps and jitter. 3. Clean. Auto runs unattended; Assisted, the default, lets you scope which takes, which frame range, which error types, which markers, and which solve constraints are held. The governing rule is that the AI only touches what you scope. Repairs run in a fixed order so a fix never lands on top of a mislabelled marker. 4. Review. The same map and taxonomy, now grading the result. Per region you see what changed, which model changed it and at what confidence, and you can revert any region. Verification is a neutral fact; quality is a separate signal. One take in the published example came back at 59, down 40 points, flagged for review. 5. Export. Cleans round-trip in the original format, including Vicon Shogun VDF and OptiTrack Motive TAK, byte-preserving everywhere it was not corrected. ### Questions and answers Q: What does Kapnetix Post actually do? A: Kapnetix Post is the AI front end for motion-capture post-production. You upload a shoot day and it diagnoses every take, predicts how cleanly each one can be repaired, cleans the ones worth cleaning, and hands back a verified result in your own file format. Your raw capture is never modified: every cleanup lands as a new version you can inspect, compare and revert. Q: How long does motion capture cleanup take with Kapnetix Post? A: The industry baseline is roughly one hour of footage to five working days of professional manual cleanup, about 40 artist-hours. On our own measurements, the machine work for that hour is about two hours of compute on a single laptop core, and the artist reviews the confidence map rather than opening every take by hand. The honest framing is artist-days, not realtime: cleaning runs at roughly 40 times realtime per performer, but diagnosis runs at about realtime today and is the bottleneck we are working on. Q: What file formats does Kapnetix Post support? A: It reads FBX, C3D, BVH, TRC, TAK from OptiTrack Motive and VDF from Vicon Shogun. Exports keep the original format, including TAK and VDF, using template-based writing so the file that comes back is your own file with the animation patched in rather than a lossy re-export. Native FBX round-trip is validated on our demo build and is not yet promoted to production. Q: Does it fit into Maya, MotionBuilder and Shogun? A: Yes. Output is validated in MotionBuilder, Maya and Shogun, and there are Maya and MotionBuilder plugins so an artist can diagnose and send from inside the tool they already have open. Zero pipeline disruption is a constraint we design to, not a feature we added: nothing downstream has to change. Q: What is a confidence score, and what does it measure? A: Confidence is a measure of fixability, not of error. Any system can flag that something is broken. A Kapnetix confidence score predicts how cleanly a take or a region can be repaired, which is the question that changes your schedule. You see the same map twice: predicted before the run so you can plan the review, and measured after it so you can verify what came back. Q: What happens when a cleanup makes a take worse? A: We show you. The needs-review queue ranks takes by before-and-after confidence and delta, including negative ones. Per region you see what changed, which model changed it and at what confidence, and you can revert any region, which excludes it from that clean version. When a result cannot be measured honestly it is labelled unverified rather than given a fabricated number. Q: Does Kapnetix clean every take automatically? A: No, and it should not. Takes that arrive clean stay raw instead of being cleaned for no reason. On easy data, roughly one percent missing with gaps of one to five frames, our router sends the work to interpolation, because that is the correct engineering answer. The rule that governs a run is that the AI only touches what you scope: which takes, which frame range, which error types, which markers, and which solve constraints are held. Q: Does Kapnetix support props? A: No. Kapnetix Post handles optical capture of performers, single and multiple, and covers occlusion gaps, marker swaps and jitter. Props are not supported. We would rather tell you that before you send a shoot than after. Q: Do you train your models on our capture data? A: No. Base models are trained entirely on data we captured or purchased to our own specification, with full IP rights. Customer motion is never used to train shared models without an explicit agreement. If you want a model tuned to your own data, we train a private one for your use only. Q: Can Kapnetix run on-premise or air-gapped? A: Yes. Kapnetix Post deploys as a cloud service with EU processing, on-premise with SSO and SCIM, or fully air-gapped with an offline licence bundle. Unreleased performance is the most sensitive data in a studio, so if policy says nothing leaves the stage, we run on the stage. Q: How is Kapnetix Post priced? A: One annual licence sized to the hours of motion capture you plan to shoot that year. No per-seat fees and no per-file charges, so the price scales with the capture and you know it before the year starts. We onboard a small number of studios at a time, so every licence starts with a pilot rather than a checkout. Q: How do I try it on my own data? A: Start on our broken sample takes today, with no approvals and no NDA needed. When your own data is cleared, send a handful of takes under NDA and we triage them with you, you scope the run, and you get the measured pass and your own files back. ## Company https://www.kapnetix.ai/our-story Kapnetix started in 2020 as a research project for one of the world's largest game publishers. Based in Brno, Czechia. Ten years of applied AI, more than 50 international projects, five innovation awards, and seven clients above $1B revenue. That track record belongs to Kapnetix the company, not to any single product claim. Base models are trained entirely on data Kapnetix captured or purchased to its own specification, with full IP rights, including commissioned shoots at OptiTrack studios where errors were introduced deliberately. Customer motion is never used to train shared models without an explicit agreement. ## Kapnetix Creator Studio https://www.kapnetix.ai/ai-3d-pose-estimation Markerless generative capture from a single camera, for creators without a volume or an IMU suit. In development. Not released and no date announced. Do not describe it as available. ## Articles - The Kapnetix team visits eNStudios in Poland (2025-02-22): https://www.kapnetix.ai/post/enstudios We asked an OptiTrack Prime X studio for something unusual: hours of animation with deliberate errors in every shoot. - Kapnetix at Game Industry Conference 2024 (2024-11-04): https://www.kapnetix.ai/post/gic2024 Our team went to Poznan to meet some of the top motion capture companies in Central Europe. - Kapnetix at Forbes Next Big Thing 2024 (2024-09-26): https://www.kapnetix.ai/post/forbes-next-big-thing-2024 We showed what we have been building at the Forbes event in Prague, and turned a few visitors into their own 3D avatars. - Product update: September 2024 (2024-09-16): https://www.kapnetix.ai/post/product-update-september-2024 Affordable capture solutions put 3D animation in reach of independent creators. The cleanup that follows did not come with them. - You no longer need so many cameras (2024-09-05): https://www.kapnetix.ai/post/reduced-hardware-costs New studios spend tens of thousands on redundant cameras, largely to avoid the cost of fixing a blocked marker in post. - Learnings from creating AI for the game industry (2023-03-30): https://www.kapnetix.ai/post/learnings-from-creating-ai-for-the-game-industry The game industry is unique, and presenting AI tools to it requires a unique approach. A summary of what the last two years taught us. - How generative AI is changing creative workflows (2022-12-17): https://www.kapnetix.ai/post/how-generative-ai-is-changing-creative-workflows Generative AI has the potential to be disruptive in highly creative industries like gaming. As with any technology, it is important to consider the limitations alongside the potential. - The Kapnetix story (2022-11-19): https://www.kapnetix.ai/post/the-kapnetix-story Kapnetix started as a research project for one of the world's largest game publishers in 2020. Since then we have been blown away by how AI can improve the game development process. ## Contact https://www.kapnetix.ai/contact Kapnetix onboards a small number of studios at a time. A pilot starts on broken sample takes with no approvals needed, then moves to a handful of the studio's own takes under NDA.