Image to 3D: Build a Walkable Scene from One Photo
Upload a single reference image and WorldGen extends it into a full 360 degree environment you can explore from any angle.
Why start from an image
A photograph or a render fixes the things that are hardest to describe: palette, materials, lens, time of day and the exact framing. Image to 3D takes that fixed look and grows a world around it, so the result matches your reference instead of merely echoing it.
It is the fastest route when the composition already exists: a concept render, a storefront interior, a floor plan visual, or a landscape you want to step into.
Image or text: choosing the right input
Both inputs run through the same two-stage pipeline, so the question is only what you already have.
- Use an image when the look matters more than the layout: the palette, lens and lighting of the reference are preserved.
- Use text when the place does not exist yet, or when you need a set of variations quickly.
- Use both in sequence: generate from text, pick the frame that reads best, then re-run it as an image to lock that composition in.
How the pipeline completes your photo
Your reference anchors one frame of the 360 degree map. The model keeps what it can see, including camera height, field of view and the materials in shot, and invents the parts a single photo can never show while keeping light and materials continuous as the camera turns.
The completed panorama then goes through the same second stage used by text input and becomes Gaussian splats or a mesh, so you move through the space instead of looking at a flat backdrop.
Choosing a reference image that works
The reference is the strongest signal you give the pipeline, so it is worth a minute of choosing.
- Prefer wide, level shots. Extreme wide-angle distortion or a tilted camera makes the generated surroundings bend with it.
- Keep the horizon visible if there is one; the panorama stage uses it to keep the world level.
- Avoid watermarks, logos and dense text, which tend to reappear on the walls the model invents.
- Use a source you hold the rights to. You are responsible for the reference image you upload.
- Interiors work especially well: the model extends a room into a complete space without inventing a new building.
What comes out
The scene stays in your workspace, opens in the browser and can be exported as 3D Gaussian splats or as a mesh. Credits are shared with text to 3D, so iterating from either input costs the same.
From photo to scene in four steps
The same pipeline, anchored on a different input.
- 01
Upload the reference
Pick the image that carries the look you want, ideally a wide and level shot.
- 02
Review the extended panorama
Check how the world continues behind the camera. Re-run with a different reference if the surroundings drift.
- 03
Convert to 3D
The panorama becomes Gaussian splats or a mesh and keeps the materials of your photo.
- 04
Explore and export
Walk through the space, then export splats or a mesh, or share the interactive result.
What the second stage keeps from your photo
A single image fixes some properties and leaves others to the model.
Frequently asked questions
Can I use a photo of a real place?
Yes, as long as you hold the rights to it. You are responsible for the reference image you upload and for how the resulting scene is used.
How many images do I need?
One. A single well-framed photo anchors the scene; the pipeline does not require extra views.
Will the result look exactly like my photo?
The visible part does: palette, materials and framing carry over. Anything the photo never showed is generated, so treat the surroundings as an interpretation rather than a reconstruction.
Related guides
Step inside the picture you already have
Upload one reference image and explore the world around it.