Postshot vs Cloud Gaussian Splatting: Local GPU or Hosted Pipeline

An RTX 4090 sitting idle on a desk is either a resource or a reminder. Postshot vs cloud Gaussian splatting comes down to what you do with that card: train on it directly, or send the job to someone else's cloud instead. Postshot, from the German studio Jawset, turns a folder of photos or a video into a Gaussian splat scene without the file ever leaving the machine. Captures Studio does the same job in a browser, on hardware you never see. Which one is right depends less on taste than on what happens to the scene once training finishes.

Should you train Gaussian splats locally with Postshot, or in the cloud?

Postshot is the better choice if you already own a capable NVIDIA GPU, work mainly from straight photo or video capture, and want the whole pipeline running on a machine you control, with no recurring cloud step. Captures Studio is the better choice the moment sharing matters: the scene has to open on a client's phone with nothing installed, the source is 360 footage or a LiDAR scan, or nobody on the team owns an RTX card capable of the job.

How we are comparing

Four questions decide most of this, stated before any verdict so the reasoning can be checked rather than taken on faith.

• Who owns the hardware. One approach needs an RTX card sitting in a specific box. The other needs a browser tab.
• What goes in. Photos and video only, or photos, video, 360 footage and LiDAR as well.
• Where the scene ends up. A local export file, or a link, embed and QR code that opens on any device.
• What happens after training. A viewer built for a technical operator, or an editor built for hotspots, tours and a client who has never touched 3D software.
• Who can open the result. Someone with the same plugin or viewer installed, or anyone with a phone.
• What the cost actually is. A GPU bought once plus a licence, against no hardware at all and a cloud bill instead.

Postshot, in short

Postshot is local gaussian splatting software in the most literal sense: structure from motion, training and export all run on Windows, on the machine that owns the GPU, with nothing sent anywhere else. It requires Windows 10 or later and an NVIDIA GPU of compute capability 7.5 or higher, with a GeForce RTX 2060 or a Quadro T400/RTX 4000 named as the entry point (as of Radiance Fields' Postshot platform page, updated 2026). Accepted input is JPG, PNG, TIF, EXR, DNG and RAW photos, plus MOV, MP4 and MKV video, with no 360 or LiDAR format listed (per Jawset's official product page, 2026).

The free tier trains at unlimited image count and model size, but it is non-commercial and watermarks the rendered output. Indie, at 17 euros a month or 204 euros billed yearly, removes the watermark and adds commercial use plus PLY and SPZ export. Studio, at 39 euros a month or 468 euros yearly, adds RAW/EXR/DNG input handling, HDR training and a command-line interface for batch jobs (Radiance Fields, 2026). A user watches the splat densify in real time during training, which makes Postshot a genuinely good tool for learning how the optimisation actually works, not just for producing a finished file.

Jawset moved Postshot out of beta after launching it in December 2023, and the current build, 1.1.43, writes SPZ version 4 and adds an HTML export target alongside PLY (Radiance Fields, 2026). Training runs on the 3DGS MCMC and Splat3 profiles, with a legacy Splat ADC option kept for older projects. The Unreal Engine plugin tracks versions 5.4 through 5.8, and an After Effects integration covers compositing work, so a studio already built around either tool gets a short path from raw footage to a usable asset inside software it already runs.

Captures Studio, in short

Captures Studio moves the same job off the desk. Photos, ordinary video, 360 or equirectangular footage, and LiDAR point clouds in E57, LAS, LAZ or PLY all go in through image-to-3d, video-to-3d, 360-to-3d or the LiDAR pipeline. Processing runs in the cloud, typically 10 to 120 minutes, with no local GPU and nothing to install. The result opens as a link, an embed or a QR code, on a phone, tablet, desktop or VR headset, with no app required.

Editing happens in the browser too: guided camera paths, hotspots carrying text, image, video and links, and objects placed directly in the scene, through the Splat Editor. Two free tools ship alongside the main pipeline: Multi Splat, which aligns two captures of the same space and renders the difference as a 4D timelapse in MP4, and the editor itself. An API at /docs/reconstruction-api covers automated jobs, and finished scenes can be published to the community for anyone to view or remix.

Enterprise plans go further than Postshot's local install ever needs to: a white-labelled viewer and editor under a client's own brand, colours and domain, plus an offline viewer that opens a scene with no connection at all, which matters for a site record that has to travel to a location without reliable internet. None of this asks the person opening the link to have installed anything first, which is the structural difference from a desktop tool: the compute and the delivery both live in the same place, and that place isn't the viewer's machine.

What training actually looks like

The two pipelines diverge earliest at the moment training starts. In Postshot, footage loads into the app, structure from motion runs on the GPU already in the machine, and the splat appears on screen while it is still forming, gaining detail pass by pass until it settles. That's the postshot vs cloud gaussian splatting trade in its most literal form: local wait, local feedback, and a result you can interrupt or restart the instant something looks wrong, because the whole loop lives on hardware you can see.

Captures Studio removes that loop from view entirely. Footage uploads, the job queues in the cloud, and the finished scene is ready in 10 to 120 minutes depending on size, with no GPU fan spinning up on the desk and no training progress to watch. That's a fair trade for someone who would rather send a job and come back to a link than babysit a render, but it does mean the granular control Postshot offers mid-training doesn't exist on the Captures Studio side.

At a glance

Where Postshot wins

Once a scene has trained, Postshot owes nothing to a network connection. There is no queue, no upload, and no dependency on a remote service staying up, which matters for a studio that has already built a pipeline around it. Watching the Gaussians densify during training is a real advantage for anyone learning or troubleshooting the optimisation itself, not something most cloud tools expose at all. Postshot also plugs directly into Unreal Engine and After Effects, which shortens the path for a team already working in those tools, and the Studio tier's RAW, EXR and HDR handling suits colour-critical VFX work that a JPEG-only pipeline cannot match. A studio that has already paid for the GPU sees no marginal cost per job either, which matters at high volume in a way a per-project cloud bill does not.

Where Captures Studio wins

Nobody has to buy or maintain a GPU to use it, which removes the single largest barrier to getting started. The input list reaches further too: Postshot's own documented formats stop at photos and video, while Captures Studio also takes 360 footage and LiDAR point clouds, so a firm scanning a site with a survey-grade laser scanner has a route in that Postshot simply does not list. Delivery is the bigger gap, though. A splat trained in Postshot has to be exported and opened in a separate viewer or plugin; a scene finished in Captures Studio is already a link a client opens on a phone at midnight, in another city, with nothing installed. For a real estate listing, a hotel walkthrough or a site record meant for people who were never going to install a Windows GPU app, that difference is the whole product. Anyone searching for a postshot alternative specifically because they don't own that GPU is really asking for this: the same category of result, minus the hardware requirement. Listings carrying a 3D tour were viewed 43% more than those without one, per Zillow's 2022 figures, and that number only applies to a scene a buyer could actually open without friction.

Which to choose: Postshot vs cloud Gaussian splatting

If an RTX 3080 or better already sits under the desk, the source material is photos or straight video, and the output only ever needs to reach a teammate with similar hardware, Postshot's Indie or Studio tier keeps the entire pipeline local and gives up nothing.

A 360 camera or a LiDAR scanner as the source points the other way. Take Captures Studio; Postshot's published format list has no place for either.

Sharing to a client's phone, in a browser, with no software installed on their side, is also a Captures Studio job. That's not a delivery model Postshot was built around.

Where the work already lives inside an Unreal Engine or After Effects pipeline, and the team owns the GPUs to run it, Postshot's plugin integration is worth more than where the compute physically sits.

And when nobody on the team wants to think about GPU compute capability, driver versions or which Windows box has enough VRAM for a 2,000-photo job, that entire category of problem disappears with a cloud pipeline. It's a smaller consideration than formats or sharing, but for a small team without a dedicated technical operator, it's often the one that decides things first.

FAQ

Does Postshot support 360 video or LiDAR input?

No. Jawset's documented input formats cover JPG, PNG, TIF, EXR, DNG and RAW photos plus MOV, MP4 and MKV video. 360 footage and LiDAR point clouds are not on that list, which is where a cloud pipeline built around both becomes the more direct route in.

Can a Postshot scene be shared the way a Captures Studio one is?

Postshot exports a PLY or SPZ file that then needs a separate viewer, plugin or hosting step to reach anyone else. Captures Studio produces a link, embed or QR code as the default output, opening on a phone, tablet or desktop with no app.

Does either tool need a subscription?

Postshot's free tier is real and unlimited in size, but it is non-commercial and watermarks output; commercial use starts at Indie, 17 euros a month. Captures Studio's plans are listed on its pricing page.

What GPU does Postshot actually need?

Windows 10 or later and an NVIDIA card of compute capability 7.5 or higher, with a GeForce RTX 2060 or Quadro T400/RTX 4000 named as the floor (Radiance Fields, 2026). A 2,000-plus photo reconstruction is a heavier job than that entry-level card is built for, so the practical minimum for serious work runs closer to an RTX 3080 or better.

All trademarks belong to their respective owners. No affiliation, partnership or endorsement is implied.

A GPU sitting idle under a desk is a sunk cost either way. What decides this one is not which tool trains a cleaner splat, since both can, but what has to happen to the scene the moment training stops: stay on the machine that made it, or reach someone who was never going to install anything to see it.