From raw frames to video.
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 fixed before the runs, and no result was picked from several.
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 whole path: a smooth camera move in which the camera pulls back 0.6 m and sinks 0.4 m and turns right 15 degrees, then glides forward 1.2 m and slides left 0.9 m and turns right 15 degrees.









The same office space with two l-shaped desks, a window bank with blinds, a high-back chair, and a whiteboard on the wall, lit by natural light and featuring light wood, blue, and white materials. The camera pulls back slightly, sinks lower, and turns right, revealing more of the right workstation desktop and the whiteboard on the wall, while the main desktop monitor and high-back chair remain visible in the frame. The scene is lit by natural light, casting soft shadows and highlighting the light wood of the desks, the blue of the chair, and the white of the walls and whiteboard.









The same office workspace with light wood desks, a whiteboard on the wall, and a window bank with blinds. a black monitor, keyboard, and mouse are on the desk. the room is lit by natural light. The camera glides forward, passing the black monitor and keyboard on the desk, then slides left to reveal the blue trash bin and the whiteboard on the wall, before turning right to show the high-back office chair and the window bank with blinds. The room is lit by natural light, creating a bright and calm atmosphere with visible reflections on the window blinds and the desk surface.
Input frames

Reconstruction: point cloud and boxes
Camera path: proxy and generated video
Context photographs and written prompt per chunk
The whole path: a smooth camera move in which the camera slides right 0.7 m and glides forward 0.5 m and turns left 54 degrees, then slides right 0.8 m and sinks 0.6 m and turns left 35 degrees.









The same small room with a window covered by sheer curtains, a wooden shelving unit holding various items, a bed with a dark mattress and a white pillow, a blue storage box, and a black backpack on the floor, lit by natural daylight. The camera slides right, glides forward, and turns left, passing the sheer white curtains, revealing the wooden shelving unit holding assorted items, then moving toward the bed with its dark mattress and white pillow, and finally showing the blue storage box and black backpack on the floor. Natural daylight illuminates the room, casting soft shadows and creating a calm, still atmosphere.









The same corner of a room with white walls, wooden herringbone floor, and a window covered by sheer white curtains. two wooden shelving units hold various items, and an office chair with a black jacket draped over it sits near a desk. natural light filters through the curtains. The camera slides right, sinks lower, and turns left, passing close to the sheer white curtains, revealing the wooden shelving units holding assorted items, and showing the grey office chair with a black jacket draped over it, positioned near the window radiator. Natural light filters through the sheer white curtains, casting soft illumination on the wooden shelves and creating a calm, still atmosphere.
Input frames

Reconstruction: point cloud and boxes
Camera path: proxy and generated video
Context photographs and written prompt per chunk
The whole path: a smooth camera move in which the camera slides right 1.5 m and turns left 15 degrees.









The same grand, circular rotunda with marble pilasters, a patterned tiled floor, and a central information stand, illuminated by warm ceiling lights and accented with red trim and gold railings. The camera slides right and turns left, passing the wooden bench on the left, revealing the central information stand, then moving past the main straight staircase on the right, with marble pilasters and gold capitals visible throughout. The scene is illuminated by warm ceiling lights, casting a soft glow on the patterned tiled floor, marble pilasters, and red trim, creating a grand and serene atmosphere.
Input frames

Reconstruction: point cloud and boxes
Camera path: proxy and generated video
Context photographs and written prompt per chunk
The whole path: a smooth camera move in which the camera glides forward 4.7 m and slides right 0.7 m and turns right 15 degrees, then glides forward 3.4 m and slides left 3.2 m and turns right 91 degrees, then pulls back 3.9 m and slides left 1.1 m and turns right 79 degrees.









The same interior of a church nave with high ceilings, arched doorways, and rows of wooden pews. a central aisle leads to a raised sanctuary with a high altar. warm light from chandeliers and wall sconces illuminates terracotta tile floors and cream-colored walls. The camera glides forward and slides left, revealing the high altar draped in white cloth on the right, while the presider chair with its red cushion appears on the left side of the frame, both set against the cream-colored walls and terracotta tile floor. The camera continues gliding forward and turning right, passing the central aisle and revealing rows of dark wooden pews on both sides, with chandeliers hanging above and the raised sanctuary platform ahead. Warm light from the chandeliers and wall sconces illuminates the scene, casting a serene glow over the terracotta floor and cream walls.









The same church nave with wooden pews, arched alcoves, and a chandelier, lit by warm artificial light and natural light from high windows, featuring cream walls, terracotta tile floors, and red curtains. The camera glides forward 3.4 meters, passing rows of wooden pews with dark brown finish, then slides left 3.2 meters and turns right 91 degrees, revealing the left side of the nave with arched alcoves framed by red curtains, a wall lantern with warm light, and a devotional image on the front left chapel wall. The scene is lit by warm artificial light from wall lanterns and natural light from high windows, casting a serene glow on the cream walls, terracotta tile floors, and wooden pews.









The same chapel interior with wooden pews arranged in rows facing a raised sanctuary platform. the walls are light-colored stucco with arched doorways and decorative moldings. a central reredos and devotional images are visible. warm, ambient light from wall lanterns and windows illuminates the space. The camera slides left and turns right, passing the left wall lantern and revealing the left seating row with wooden pews and light-colored cushions, framed by the arched doorway and light-colored stucco wall. The camera pulls back and slides left, revealing more of the left seating row and the front left chapel devotional image on the wall, all under warm ambient light. The camera pulls back further and slides right, revealing the raised sanctuary platform, central reredos, and sanctuary ceiling mural, with wooden pews and light-colored cushions in the foreground. Warm, ambient light from wall lanterns and windows illuminates the space, casting a serene glow on the light-colored stucco walls and wooden pews.
Input frames

Reconstruction: point cloud and boxes
Camera path: proxy and generated video
Context photographs and written prompt per chunk
The whole path: a smooth camera move in which the camera slides right 1.4 m and glides forward 1.2 m and turns left 9 degrees, then slides right 1.3 m and sinks 1.3 m and turns left 14 degrees, then slides right 1.6 m and pulls back 0.7 m and turns left 15 degrees.









The same abbey on a rocky island, surrounded by water and a village, with a prominent spire and scaffolding on its main body, under clear blue sky and warm sunlight. The camera slides right and glides forward, passing the long front rampart, revealing the abbey's main masonry body with its stone facade and scaffolding, flanked by green front and side slope woodlands, while outer village buildings made of stone and brick remain visible in the background. The scene is bathed in warm sunlight under a clear blue sky, casting soft shadows and highlighting the textures of the stone and foliage.









The same abbey on a rocky island, surrounded by a village and tidal flats under clear blue sky. stone abbey with gothic spire and scaffolding, village houses with dark roofs, green trees, and sandy shore. warm golden light. The camera slides right and sinks lower, revealing more of the front slope woodland on the right side of the frame, while the abbey's main masonry body remains centered. As it turns left, the side slope woodland on the right becomes more prominent, and the long front rampart on the left comes into view, all under warm golden light. Warm golden light bathes the stone abbey, green trees, and sandy shore under a clear blue sky.









The same abbey on a rocky island, surrounded by a village and tidal flats under clear blue sky, with stone walls, gothic architecture, and green trees illuminated by warm sunlight. The camera slides right, pulls back, and turns left, revealing the abbey's main masonry body with its tall stone spire, the front gabled village block to the left, and the side slope woodland to the right, all under warm sunlight. The scene is bathed in warm sunlight under a clear blue sky, with the stone abbey, village buildings, and green trees illuminated against the tidal flats.
Input frames

Reconstruction: point cloud and boxes
Camera path: proxy and generated video
Context photographs and written prompt per chunk
The whole path: a smooth camera move in which the camera slides right 5.3 m and pulls back 1.4 m and turns left 45 degrees.









The same rotunda with pink columns and stone arches, surrounded by water and trees under bright blue sky, with ornate carvings and a white dome visible. The camera slides right and rises, passing the northeast rotunda support and revealing the relief-covered projecting pedestal, all under bright blue sky with pink columns and stone arches visible. The camera pulls back and turns left, revealing the southeast and south rotunda supports, with the rotunda entablature and roof centered in the frame, surrounded by trees and water. The camera slides right and sinks, passing the northwest rotunda support and revealing the west bank tree, with the rotunda entablature and roof still visible. The camera slides right and sinks again, revealing the south bank tree and rear approach tree, with the rotunda entablature and roof centered in the frame. Bright sunlight illuminates the pink columns, stone arches, and white dome, casting sharp shadows and highlighting the ornate carvings against 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 whole path: a smooth camera move in which the camera pulls back 2.9 m and sinks 0.2 m, then pulls back 2.2 m and slides right 2.1 m.









The same brick lighthouse tower and adjacent museum hall under bright sun, with a flagpole bearing an american flag, a lantern platform slab, and railings on a paved area, all in red brick and white trim. The camera pulls back and sinks slightly, revealing more of the paved area and railings, while the red brick lighthouse tower and adjacent museum hall remain central, with the American flag on its white flagpole visible in the frame. Bright sunlight illuminates the red brick structures and white trim, casting clear shadows on the paved ground.









The same brick lighthouse tower and adjacent museum hall under bright sun, with a flagpole and american flag, a wooden bench, and railings visible. red brick, white trim, and dark roof dominate. sunlight creates lens flare. The camera pulls back 2.2 meters and slides right 2.1 meters, revealing more of the brick museum hall on the left and keeping the lighthouse tower centered in the background, with the American flag on its pole visible in the mid-frame. Bright sunlight creates lens flare, casting strong highlights on the red brick, white trim, and dark roof of the buildings.
Runs
One cold-start run per row, all times in seconds: the reconstruction on one GPU, the generation on one eight-GPU node. Wall clock is the sum of the two commands. 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 | 900 | 448 | 0 | 79 | 44 | 270 |
| Bedroom | 2 / 214 | 964 | 504 | 0 | 74 | 45 | 270 |
| Rotunda museum | 1 / 124 | 1222 | 925 | 0 | 72 | 33 | 132 |
| Mission church | 3 / 304 | 1689 | 1068 | 0 | 68 | 80 | 410 |
| Mont-Saint-Michel | 3 / 304 | 1112 | 511 | 0 | 70 | 60 | 410 |
| Palace of Fine Arts | 1 / 124 | 913 | 613 | 0 | 73 | 33 | 132 |
| Brick lighthouse | 2 / 214 | 964 | 509 | 0 | 72 | 43 | 271 |
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. The verifier column gives, per chunk, the judge's rating of how well the frames match the photographs and the layout, then of visible artifacts, each from 0 (none) to 3 (severe). Seeds are the path seed and the sampling seed.
| scene | environment | objects | blocking | path m | clearance m | coverage | verifier | seeds |
|---|---|---|---|---|---|---|---|---|
| Office | indoor | 22 | 3 | 2.9 | 0.32 | 0.87, 0.87 | 0 / 1, 1 / 1 | 3153 / 85727 |
| Bedroom | indoor | 23 | 1 | 2.0 | 0.39 | 0.83, 0.69 | 1 / 1, 1 / 2 | 0 / 42 |
| Rotunda museum | indoor | 111 | 7 | 1.8 | 0.67 | 0.79 | 0 / 1 | 6034 / 46906 |
| Mission church | indoor | 180 | 15 | 12.3 | 0.32 | 0.66, 0.59, 0.63 | 2 / 1, 0 / 1, 0 / 2 | 3322 / 91758 |
| Mont-Saint-Michel | outdoor | 29 | 5 | 6.0 | 1.43 | 0.74, 0.63, 0.44 | 0 / 1, 0 / 1, 2 / 1 | 9564 / 23265 |
| Palace of Fine Arts | mixed | 41 | 11 | 6.0 | 0.32 | 0.56 | 2 / 1 | 2327 / 52015 |
| Brick lighthouse | outdoor | 14 | 5 | 5.9 | 1.35 | 0.57, 0.32 | 0 / 1, 0 / 1 | 5114 / 31720 |
Settings
- Two commands per scene:
python -m mundo_infer.run reconstruct --source <frames>, thengenerate --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.