From raw frames to video, in one command.
Scenes of the Mundo project page, section 1.1, indoors and outdoors, each taken once through the deployment pipeline: the raw frames are reconstructed into a proxy scene, a camera path is designed from the scene, the proxy is rendered along it, and the video is generated. The number of chunks (one to three) and both seeds were drawn at random before the runs, and no result was picked from several: where a scene was run again, the list under the tables says why.
Stage by stage
Each scene shows what the stages wrote, in order: the input frames, the reconstruction as a point cloud with the boxes of the exported scene (drag to orbit, scroll to zoom, pick an object to highlight its box), the proxy video next to the generated video on the same camera path, and for every chunk the context photographs chosen and the description written for the prompt.
Input frames

Reconstruction: point cloud and boxes
Camera path: proxy and generated video
Context photographs and written prompt per chunk
The same office space with a blue door, light wood desks, whiteboards, and a window with blinds. the walls are white, the floor is blue carpet, and the lighting is bright and even. furniture includes chairs, monitors, and a shelving unit. The whole path: a smooth camera move in which the camera pulls back 0.6 m, then slides left 1.1 m and glides forward 1.0 m and turns right 30 degrees.








The camera pulls back, revealing more of the office space, including the blue high-back office chair, the light wood desks with black monitors, and the whiteboard on the wall, all under bright, even lighting. The scene is brightly lit with even, natural light, highlighting the white walls, blue carpet, and light wood desks.







The camera slides left, glides forward, and turns right, passing a black monitor on a light wood desk, revealing a blue chair and a whiteboard on the wall, with a window with blinds visible on the left side of the frame. The scene is brightly lit with even lighting, and the atmosphere is calm and still.
Input frames

Reconstruction: point cloud and boxes
Camera path: proxy and generated video
Context photographs and written prompt per chunk
The same grand, circular hall with marble columns, wooden benches, and a central mosaic floor, illuminated by warm ceiling lights and stained-glass windows, featuring an upper gallery with ornate railings and a main staircase. The whole path: a smooth camera move in which the camera slides right 1.3 m and rises 0.6 m and turns right 15 degrees.








The camera slides right, rises slightly, and turns right, passing the southeast wooden bench on the left, revealing the main staircase's lower flight on the right, then centering on the staircase as it moves toward the west hall pier and the central information sign, all framed by marble columns with gold capitals and a mosaic floor. Warm ceiling lights illuminate the grand hall, casting a soft glow on the marble columns, wooden benches, and the intricate central mosaic floor, creating a serene and historic atmosphere.
Input frames

Reconstruction: point cloud and boxes
Camera path: proxy and generated video
Context photographs and written prompt per chunk
The same ornate church interior with rows of wooden chairs on terracotta tile floors, arched doorways with red curtains, a high altar with a retable, and a painted ceiling with stars and angels, lit by chandeliers and wall sconces. The whole path: a smooth camera move in which the camera slides left 2.8 m and glides forward 2.3 m and turns right 59 degrees, then slides left 2.8 m and pulls back 1.9 m and turns right 117 degrees, then slides left 4.0 m and pulls back 3.5 m and turns right 31 degrees.








The camera glides forward 2.4 meters and slides left 0.9 meters, passing the freestanding altar table and revealing the sanctuary platform and sanctuary chair on the left, with the sanctuary rug centered underfoot. The camera slides left 0.8 meters and sinks 0.2 meters, turning right 18 degrees to reveal the wooden lectern and left sanctuary-side door on the right, while the middle nave chandelier is visible on the left. The camera pulls back 0.8 meters and slides left 0.7 meters, turning right 40 degrees to reveal the nave chairs on the left and the front left side altar on the right, with the sanctuary platform still visible on the right. The scene is lit by warm chandeliers and wall sconces, casting a serene glow on the terracotta tile floors and wooden chairs with beige cushions.









The camera slides left, pulls back, and turns right, revealing the high altar retable with its golden ornamentation and crucifix, then passes the wooden lectern and moves toward the sanctuary platform, showing the rows of wooden chairs with beige cushions and the arched doorways with red curtains in the background. The scene is warmly lit by chandeliers and wall sconces, casting a serene glow over the terracotta tile floors and ornate architectural details.







The camera slides left, pulls back, and turns right, revealing the high altar retable with its golden ornamentation and crucifix, then passing rows of wooden chairs with beige cushions as it moves toward the left side of the nave, where arched doorways with red curtains come into view. The scene is warmly lit by chandeliers and wall sconces, casting a serene glow over the terracotta tile floors and ornate painted ceiling with stars and angels.
Input frames

Reconstruction: point cloud and boxes
Camera path: proxy and generated video
Context photographs and written prompt per chunk
The same historic abbey complex perched on a rocky island, surrounded by a tidal flat and dense green trees under a clear blue sky. the main abbey church with its tall spire and gothic architecture dominates, with stone buildings and ramparts forming a fortified village below. the light is warm and golden, casting long shadows. The whole path: a smooth camera move in which the camera slides left 1.6 m and sinks 0.8 m and turns right 15 degrees.






The camera slides left, sinks slightly, and turns right, revealing the abbey church's upper body and spire centered in the frame, with the ascending village curtain wall and lower wooded slope on the right, and the lower village curtain wall and rampart corner tower appearing on the left as the view progresses. Warm golden light casts long shadows across the stone buildings and tidal flat under a clear blue sky.
Input frames

Reconstruction: point cloud and boxes
Camera path: proxy and generated video
Context photographs and written prompt per chunk
The same ornate rotunda structure with classical columns and a domed roof, surrounded by water and trees under bright blue skies, featuring pinkish stone and detailed carvings, with sunlight casting shadows on the architecture and water. The whole path: a smooth camera move in which the camera slides right 4.5 m and pulls back 2.6 m and turns left 45 degrees.









The camera slides right, passing the Northwest rotunda pier and revealing the North rotunda pier on the left and the West rotunda pier on the right, all made of pinkish stone with detailed carvings, while the rotunda's domed roof appears in the distance. The camera continues sliding right and pulling back slightly, revealing the West bank tree on the right and the Rotunda upper drum and roof centered in the frame, with the Northwest rotunda pier now on the left edge. Bright sunlight casts sharp shadows on the pinkish stone architecture and the surrounding water under a clear blue sky.
Input frames

Reconstruction: point cloud and boxes
Camera path: proxy and generated video
Context photographs and written prompt per chunk
The same brick lighthouse station with a tower, adjacent museum building, and flagpole under clear blue sky, featuring wooden benches, green shutters, and solar panels on the roof, lit by bright sunlight casting sharp shadows. The whole path: a smooth camera move in which the camera pulls back 2.8 m and sinks 0.7 m, then slides left 2.6 m and pulls back 1.2 m.






The camera pulls back and sinks slightly, revealing more of the brick lighthouse tower and museum building, showing the red brick, green shutters, and the American flag on its pole, with sharp shadows cast by bright sunlight. Bright sunlight casts sharp shadows across the scene under a clear blue sky.






The camera slides left and pulls back, passing the brick museum building which has green-shuttered windows and a sign, then revealing the red brick lighthouse tower on the right side of the frame, topped with a white lantern room, while the American flag on its pole remains visible. Bright sunlight casts sharp shadows across the scene under a clear blue sky.
Runs
One cold-start run per row on one eight-GPU node, all times in seconds. Wall clock covers the whole command. Model loading is the captioner and the condition encoder; prepare is the context selection, proxy rendering, prompt writing and condition encoding of all chunks; sampling is the diffusion steps and decoding of all chunks. What remains of the wall clock is loading the video model and the hand-over between stages.
| scene | chunks / frames | wall clock | reconstruction | path design | model loading | prepare | sampling |
|---|---|---|---|---|---|---|---|
| Office | 2 / 214 | 825 | 382 | 4 | 75 | 43 | 263 |
| Rotunda museum | 1 / 124 | 1160 | 859 | 4 | 68 | 33 | 129 |
| Mission church | 3 / 304 | 1868 | 1219 | 24 | 74 | 80 | 404 |
| Mont-Saint-Michel | 1 / 124 | 981 | 645 | 55 | 73 | 28 | 123 |
| Palace of Fine Arts | 1 / 124 | 922 | 580 | 51 | 71 | 30 | 133 |
| Brick lighthouse | 2 / 214 | 959 | 497 | 40 | 72 | 40 | 253 |
Blocking issues are the objects the reconstruction's final review still rejects. Clearance is the smallest distance between the camera path and a proxy surface. Coverage is the share of each chunk's view that its context photographs show, one value per chunk. Seeds are the path seed and the sampling seed.
| scene | environment | objects | blocking | path m | clearance m | coverage | seeds |
|---|---|---|---|---|---|---|---|
| Office | indoor | 21 | 3 | 2.9 | 0.52 | 0.89, 0.88 | 3153 / 85727 |
| Rotunda museum | indoor | 117 | 11 | 1.5 | 0.52 | 0.75 | 6034 / 46906 |
| Mission church | indoor | 255 | 6 | 11.8 | 0.48 | 0.71, 0.63, 0.70 | 3322 / 91758 |
| Mont-Saint-Michel | outdoor | 39 | 6 | 2.4 | 1.60 | 0.73 | 2398 / 43488 |
| Palace of Fine Arts | mixed | 37 | 9 | 5.4 | 0.32 | 0.56 | 2327 / 52015 |
| Brick lighthouse | outdoor | 15 | 5 | 5.9 | 1.57 | 0.44, 0.32 | 5114 / 31720 |
- Mission church: Second run. The first stopped in the reconstruction: the agent returned an empty scene program and the repair step could not fill it. The reconstruction now asks for the program once more in that case; same chunk count and seeds.
- Mont-Saint-Michel: Result from before the open-air fixes of the reconstruction; this scene is being run again.
- Palace of Fine Arts: Run again after the open-air fixes of the reconstruction (plain ground and water, fewer boxes, boxes cut back to what no view looks through); new random chunk count and seeds.
- Brick lighthouse: Run again after the open-air fixes of the reconstruction (plain ground and water, fewer boxes, boxes cut back to what no view looks through); new random chunk count and seeds.
Settings
- One command per scene:
python -m mundo_infer.run all --source <frames> --auto --auto-seed <seed> --seed <seed> --chunks <n>, with the default settings of the package and no reconstruction prompt. - Reconstruction: 48 frames per scene, Pi3X geometry, one editing round; indoors or outdoors is decided by the pipeline itself.
- Generation: 640×480, 124 frames for the first chunk and 90 more per further chunk, 32 steps of dpmpp2m, eight GPUs in tensor parallel, at most nine context photographs per chunk.