Open editor →

Tiny world, full soundtrack

A miniature scene becomes a short film with sound drawn from its motion.

Sampled frames and model-assisted audio review · 2026-09-12 · Reported saved run: $0.18176. meta/muse-image/text-to-image · wan-video-image-to-video · mirelo-ai/sfx1.6/video-to-video

Muse builds the first frame, Wan animates that image, and Mirelo adds foley from the generated video. The sound model receives the actual take, without the original world description. The graph outputs the first frame and film with foley. The saved sample also includes the intermediate silent take for comparison.

Saved result

1. Muse: the miniature world — Tiny world, full soundtrack
1. Muse: the miniature world · Download
2. Wan: the actual silent intermediate · Download
3. Mirelo: the clip with generated foley · Download

Muse builds a walnut-shell watermill; Wan animates that image; Mirelo watches the actual silent take and adds its sound. The 480p result is five seconds long. Sampled frames retain the mill and turning wheel; a separate audio model describes rushing water and faint friction, without speech or music. The audio ends sharply. Exact audiovisual sync and seamless looping were not established.

Another run: The watchmaker’s mothSee what the same workflow made with different inputs ↓

Sampled frames and model-assisted audio review · 2026-09-12 · Reported saved run: $0.18176.

Muse creates a brass moth on a wooden spool; Wan animates its wings; Mirelo adds sound from that actual silent clip. The five-second film keeps the moth and spool, but the orange wing underside changes the intended folded-metal material. A separate audio-model review describes rhythmic metallic friction and clicks without speech or music, and notes an abrupt ending. That review did not see the video: precise audiovisual synchronization and seamless looping are not established.

1. Muse: the mechanical moth — The watchmaker’s moth
1. Muse: the mechanical moth · Download
2. Wan: the actual silent intermediate · Download
3. Mirelo: the film with generated foley · Download
Separate model-assisted audio review
Here is a review of the audio clip: ### Timeline of Events - **00:00.00 – 00:04.90:** A recurring sequence of metallic, rhythmic friction and clicking sounds. Every approximately 1 to 1.3 seconds, there is a distinct metallic scraping/draw sound followed by sharp clicks and an impact (at roughly 00:00.3, 00:01.5, 00:02.8, and 00:04.0). - Between these main rhythmic strikes, a faint, high-frequency hiss/air noise persists in the background. - **00:04.90 – 00:05.10:** The clip ends abruptly in silence right after the fourth mechanical cycle decays. ### Speech and Music - **Speech:** None. - **Music:** None. ### Glitches, Clipping, or Abrupt Cuts - **Clipping/Distortion:** None detected; the high-frequency transients are crisp without digital distortion. - **Abrupt Cut:** The audio cuts off sharply at the end of the file around 00:05.10.
Separate model-assisted audio review · Download
Inputs for this version

World

A handmade miniature brass moth with two broad folded-metal wings perched on a wooden spool in a sunlit dry watchmaker workshop. A tiny winding key protrudes from its back. Sand-colored wood, amber brass, pale blue painted wing tips. Tactile stop-motion set, macro lens, entire moth clearly visible, horizontal composition. No water, plants, windows, lettering or people.

Motion

One continuous locked-off macro shot. The brass moth gently opens and closes both hinged wings twice while staying on the spool. Its winding key makes a small turn. Keep the same body, pale blue wing tips and dry wooden table; no flight, extra wings, camera travel or cuts.

Sampled models: meta/muse-image/text-to-image, wan-video-image-to-video, mirelo-ai/sfx1.6/video-to-video.

Open this version →Download this versionRun record

Make it yours

  1. Open the graph — save first if the canvas already has your work. Undo brings yours back.
  2. Change World and Motion together. Describe a clear miniature subject and one visible movement. The default animation requests 81 frames at 16 fps and 480p, about five seconds. Keep these settings for the documented short format.
  3. Check the editor estimate, then run with your NanoGPT key. Three model calls. Catalog rates checked on 12 September 2026 were $0.01 for the image, $0.20 for animation and $0.01 per second for foley, about $0.26 combined. Duration rounding and higher settings can change the total. The saved run's observed cost appears above.
  4. Compare First frame with Silent take, then play Film with foley with sound. Check that the scene remains recognizable, the requested motion happens and the sounds fit the visible events. Review the whole clip; exact synchronization, music and speech are not promised.

Download workflow · Original sampled graph · Explore MCP tools · awesome-noodles