Use case: a five-shot interior design film from still images with local MiniMax H3

MiniMax H3MiniMax H3 is free open weights (runs locally)

The Astravisualize team turned AI concept images of one bathroom into a 24-second vertical film by animating five separate views with MiniMax H3 running locally in ComfyUI, then cutting them in FFmpeg. Their write-up shares the exact camera prompts, the local settings and the takes they threw away.

What it does

Each five-second shot got one camera job (move forward, push in on the faucet, rise along the tiled wall, move into an overhead detail, glide over the mosaic floor) and its own starting image. Instead of cropping the overview, they asked for new camera positions with the approved bathroom as the visual reference, then joined the clips with four quarter-second dissolves.

Results and limits

The first vanity prompt asked for an extremely slight move and came out almost static; spelling out an immediate forward dolly toward a target with visible foreground-to-background parallax fixed it. An attempt with a cropped end image broke the geometry, so they went back to first-frame-only, and one overview was dropped because it invented a mirror. The final takes took about 6 to 6.5 minutes each locally (one took 13.1 minutes), using the int8 FL2VA checkpoint with the 8-step turbo LoRA at 768×1344, 124 frames at 24 fps.

Worth knowing

The team says the numbers in the prompt (60 centimetres, 25 percent larger) are instructions, not measured camera paths, and that new viewpoints can invent hidden geometry, so check them against real photos or renders before showing a client. The final file is an upscale to 1080×1920, not native 1080p. Astravisualize sells a design visualisation product, but this film was made with built-in image generation and local H3 outside it.

Try it yourself

A five-second architectural video with a CLEARLY VISIBLE FORWARD DOLLY. Starting immediately, the camera travels smoothly forward about 60 centimeters toward the brass faucet and marble basin over the full five seconds. The near marble edge passes visibly toward the lower frame edge, while the far doorway shifts more slowly, creating unmistakable depth parallax. One continuous purposeful camera move at normal speed, no pause at the start or end. Preserve the existing bathroom design, material colors, solid geometry and arrangement of all objects. No person, no cut or transition. The camera moves; the countertop and fittings remain physically stationary.
Read the primary source >

More coverage

Related entries

< Back to all entries