3D · Beta

Spatial Video to Gaussian Splat

Reconstruct a static scene from an original iPhone spatial video. Upload both camera views, follow the processing queue and download a Gaussian PLY.

Shared processing queue

Processor unavailable · submissions stay queued

Beta

Two views. One scene you can explore.

Turn an original iPhone spatial recording into a trained Gaussian splat. Both camera views contribute to the reconstruction.

Keep the scene still

Choose a textured object or space in even light. Avoid people moving, mirrors and transparent surfaces.

Move around it

Walk slowly through overlapping views. A fixed camera or a pan from one spot gives too little coverage.

Keep both views

Upload the original spatial MOV. Regular video exports and messaging apps can remove the second view.

Choose a spatial videoiPhone photo library or Files · original MOV or MP45–60 seconds · 1080p per view · 30 fps · up to 200 MB

Uploading from your iPhone

Open this page in Safari and choose your original spatial recording from the photo library. If the check reports only one view, save the unmodified original to Files and choose it from Browse. Keep Safari open during upload; once the job is accepted, you can return in the same browser to check progress.

The original video, including its metadata and any audio, is uploaded securely to the private processing queue. Audio and location metadata are omitted from the output. HDR footage is converted to SDR for training. This beta creates a static scene at relative scale; moving objects can leave ghosts, and unseen areas remain incomplete.

I have the right to use this recording and agree to upload it to MakePip’s private processor. Save the result within 24 hours after completion.

One shared processor runs one job at a time. Requests that cannot start expire after 12 hours. Email delivery is unavailable. Return in the same browser and keep its site data to access the result.

How to use Spatial Video to Gaussian Splat

1Record a stationary, textured scene while moving slowly around it. Use 5–60 seconds of original 1080p spatial video at up to 30 fps.
2Choose the original MOV from your iPhone photo library or Files. Confirm the upload; the tool checks that both camera views are present.
3Follow camera alignment and training in the queue, then download the Gaussian PLY and open it in a compatible splat viewer.

Before you start

Is this a separate tool from video to 3D?

Yes. This tool trains a radiance representation made of coloured Gaussians. It exports a PLY splat, not a GLB triangle mesh. It is intended for scene viewing, not 3D printing.

Can I upload directly from an iPhone?

Yes. Open the page in Safari and select an original spatial recording from Photos or Files. The server checks for both MV-HEVC views. If Photos supplies a single-view copy, save the unmodified original to Files and select it there. Keep Safari open until upload is accepted.

Will any spatial clip work?

No. Spatial video adds a second camera view, but useful reconstruction still needs a still scene and a moving camera with overlap. People moving, blur, blank walls, reflections and unseen areas can cause failed alignment or ghosting. A fixed-camera recording is not a complete 3D capture.

What are the beta limits?

Up to 200 MB, 5–60 seconds, 1080p per view and 30 fps. Both views are sampled at up to 960 pixels, camera poses are estimated, and the local GPU trains up to 250,000 Gaussians for 4,000 steps. Output is limited to 100 MB. Processing shares one queue.

How are my recording and colours handled?

The original recording is uploaded over HTTPS to temporary private storage. Audio and location metadata are not included in the splat. HDR is converted to SDR for training, so brightness and colours can differ. Results have relative scale and incomplete areas outside the camera path. Save the result within 24 hours.

Read the guides →File processing & privacy →