Creating a 3D Item

Create new 3D items on the Create page. Every item is built on Gaussian Splatting, and there are four ways to make one.

• Import a 3DGS file — open a 3D file you already have (PLY etc.) directly in the advanced editor
• Create 3D from 1 to 4 Images: upload one to four product photos and AI generates the 3D on the server
• Train in the browser — train a COLMAP-preprocessed folder or ZIP on a WebGPU-capable device
• Create from video · images (recommended) — upload your footage and a 3D item is generated automatically on the server

Not sure which to pick? Use Create from video · images. It only needs your footage and gives the most reliable quality. The others are for when you already have a 3D file (import), have only a few photos (1 to 4 images), or want to train on your own device without uploading the data (browser training).

Import a 3DGS file

Register a Gaussian Splatting file made with another tool. The card opens an empty advanced editor.

• Drag the file onto the editor or pick it with the import button — supported formats: PLY, compressed PLY, .splat, .ksplat, SOG, SPZ
• Large PLY files are converted as a stream, so scenes with millions of points open fine
• Right after import you can use every editing tool without signing in. Sign in when you press Save at the top and the file is registered as a new item in your account
• Until you save, your edits live only in the browser, so save before refreshing

For editing, see the Getting started and Import · Export chapters in the → Editor section.

Create 3D from 1 to 4 Images

Build a 3D item from one to four photos, with no footage. This is not a capture-based reconstruction: AI estimates the unseen sides, so if exact geometry matters, prefer the video · images workflow.

• Enter an item title and drag or click to upload the image — JPG, PNG, WEBP, up to 25 MB
• A sharp, well-lit front-facing photo gives the best result. Avoid shots with several products or a busy background
• You can add photos from other sides (up to 4 in total). The first is the front; photos of the other sides give a more accurate multi-view result
• Quality Level: Preview and Balanced cost 1 EvoCredit; High (Recommended) costs 5 EvoCredit. High adds AI texture refinement that keeps text and logos true to your photos, and takes longer
• Press Generate 3D Model to submit the job to the server. Sign-in is required, and generation continues on the server even if you close the page
• When it finishes, Open Completed 3D opens it in the advanced editor, and the item is also added to your library

Train in the browser

Train 3DGS on your own computer's GPU instead of the server. The folder or ZIP you select is never uploaded, which helps when the source material must stay on your machine.

• Requirements — desktop Chrome/Edge with WebGPU and a current GPU driver. The page checks GPU capabilities first and points you to server training if they are missing
• Input data — not raw photos or video, but a COLMAP-preprocessed dataset: a folder or ZIP containing images/ and sparse/0/ (cameras · images · points3D, as bin or txt). At least 3 trainable images are needed, and since a ZIP is fully unpacked in browser memory, selecting a folder is safer for large sets
• Training quality — Preview (512px · 5,000 iterations) / Balanced (928px · 12,000) / High (1280px · 24,000). If you run out of memory, retry at a lower quality
• While training — once the first Gaussians exist, progress is shown as a live 3D preview with the remaining time. You can pause, resume and cancel, but do not close the tab while training
• When done — the result is converted to PLY and uploaded to your account; after post-processing it is ready to edit from your library. Download PLY lets you keep the file directly as well

If you do not have a COLMAP dataset, prepare one in EVOVA Desktop or use server training below. Server training is also recommended when you need more accurate camera poses.

Create from video · images (recommended)

Upload your footage or photos and the server handles both camera solving and 3DGS training. This is the simplest method with the most consistent quality.

Shooting tips

90% of a good 3D result is decided at capture time.

• Shoot a static subject — moving objects cannot be reconstructed
• Move the camera slowly to minimize motion blur
• Cover the subject evenly from many angles
• Products and spaces are processed the same way. Circle a product all the way around; for a space, move slowly so walls and floor overlap between frames

Creation steps

1. Basic info — enter the 3D item name. There is no type selector any more; products and spaces are reconstructed the same way
2. File upload — choose Video or Images as the input
   • Video: one file (MP4 · MOV · AVI · MKV · WebM). Pick it on this computer, or use Upload from phone (QR) below the box to send a video straight from your phone
   • Images: 3–300 photos from many angles (JPG · PNG, 250 MB total) or a ZIP of them
3. Create — when uploading from this computer, keep the tab open until the upload finishes

Advanced settings (optional)

• Quality — Standard (15+ minutes, higher quality · 300 frames, 30,000 training iterations) / Fast (5+ minutes, for quick previews · 120 frames, 15,000 iterations)
• Spherical Harmonics level — level 0 (small size, fast rendering, default) / level 3 (high quality, preserves lighting). Either way the finished item can be downloaded as PLY

Upload from phone (QR)

Send a video that is on your phone without copying it to the computer first. The button appears under the upload box when the input is Video.

1. Enter the 3D item name first; the button stays disabled without it
2. Press Upload from phone (QR) and a QR code appears. Scan it with the phone camera within 10 minutes. The phone page needs no login
3. On the phone, tap Choose video and pick one video (up to 500 MB). Keep the phone screen open until it finishes; if the connection drops, choose the same video again and it resumes
4. When the computer shows "Your video is ready", press Create. You can close the QR dialog, and the uploaded video stays available for 2 hours

If the code expires, press Prepare a new QR code; to stop, press Cancel phone upload. Phone upload and picking a file on this computer are exclusive: choosing one clears the other.

Tracking progress

After the upload you are taken to the item page, where processing status updates live — Waiting → Processing (with a % progress bar) → Done. While it processes, you can leave the page — the server keeps working, and once finished the item is ready to edit from your library.

If an error occurs it is shown in the status. If it persists, please contact support.