Pixal3D multi-view: prepare the input images

Cinevva's 3D timeline says Pixal3D gained multi-view inference in September 2026, added to its May release, and that it is open source under the MIT license. Pixal3D is not part of Sume, but multi-view 3D models need clean, consistent input views, and Sume's image edit with reference images can help prepare them.

What the timeline reports

The facts below come from one third-party timeline page read on the date shown. Check the project's own repository for installation, hardware needs and the exact input format before you build on it.

What Sume does and does not do

Sume does not carry Pixal3D or any image-to-3D model, and nothing in its docs produces a 3D mesh. What it does offer is image generation and reference-guided editing. The Image API accepts input_references; the catalog entries for the reference-capable models advertise a range of 0 to 10, and models with a range of 0 to 0 are text-to-image only and reject references.

Reference URLs must be public HTTPS. Check GET /v1/images/models for the model you intend to use before sending references.

Preparing views for a multi-view model

A multi-view model reads several images of the same object. The inputs work best when the object looks the same across them: one object, one lighting, one scale, a plain background and the same framing. Mismatched colour or proportions between views are the usual problem.

With a reference-capable image model you can start from one product photo and ask for the object from another angle on a plain background, keeping the first image as the reference. Review every view next to the others before passing them on, and discard any view in which the shape changed.

• One object, one lighting setup, plain background.
• Same framing and scale across views.
• Generate each view from the same reference.
• Compare views side by side and drop any that drifted.

{
  "model": "your-reference-capable-model-id",
  "prompt": "The same product seen from the left side on a plain white background",
  "input_references": [
    { "type": "image_url", "image_url": { "url": "https://example.com/front.png" } }
  ]
}