The reel that never uses a video model
An MIT skill that composites a still into a short vertical reel, and keeps the sway inside twelve pixels.
- Category
- Motion
- Who it is for
- Someone with a still who wants a short vertical clip and does not want a video model to invent the scene.
- Possible model
- Open source
- Difficulty
- Medium
- Inspired by
- image-to-reel · image-to-reel
- Original launch
- Watch the launch · Nov 18, 2026
Difficulty is an editorial estimate of technical and operational complexity, not a score.
Startup idea analysis by Startup Ideas. Last reviewed Nov 18, 2026. How we read a launch.
The startup idea
Swapnil posted image-to-reel on November 18. He said it is a Claude skill that turns a still into a vertical reel for Instagram or YouTube, with no video-generation model. Every frame is composited from layers in Python, OpenCV, and ffmpeg. The repository is swapnilbhange/image-to-reel. The license is MIT. The only model step is a cutout, Segment Anything through rembg, and only when something has to move in front of something else. The motions are named and small: clouds, ripples, a waterfall, cloth, a halo, glow, falling particles, a slow push-in. Faces are never warped. A frame stays locked. The readme keeps the amplitudes tight. A sway is 5 to 12 pixels. Water is 1 to 5. A sky drift is 4 to 7. A rigid part, a pole or a standing leg, needs a sway weight of zero. The output is 1080 by 1920, 30 frames a second, 9 to 15 seconds, with synthesized music and sound. He says a 14 second reel renders in about 75 seconds on two workers. The post shows a picture that moves. The movement is a few pixels of compositing. It is not a shot a video model invented. This page did not render a reel.
The opportunity is the still that moves a little, with the model left out of the frames. A smaller product is one still, one motion type, and the line that no video model ran. A reel that claims the scene was generated is not this readme.
Do not warp a face. Do not call twelve pixels of sway a generated shot.
The problem
A still does not move, and a generated shot is a different product from a few pixels of drift.
Who would pay for this?
- Primary
- Someone with a still who wants a short vertical clip and does not want a video model to invent the scene.
- Secondary
- A person who will read that the sway stops at twelve pixels.
- Early adopter
- One image, one motion, the no-model line.
How could it make money?
Possible models. Nothing here is a claim about what image-to-reel charges.
- No price. MIT.
- You still pay for Claude if you use the skill that way.
- The post names no fee.
What would the MVP look like?
- A still.
- One named motion.
- A vertical mp4.
- The sentence that no video model ran.
How to validate the idea
- Ask five people whether a model generated the moving water.
- If they say yes, the line failed.
- Do not add a video model and keep the name.
How to differentiate
- The frames are composited.
- The face does not warp.
- MIT.
- Do not become a video model.
Original product launch
This launch is inspiration and evidence that someone shipped, not a partnership. The idea above is our reading of the broader opportunity. image-to-reel did not write it.
- Product
- image-to-reel
- Source
- image-to-reel
- Film
- image-to-reel, the sway is twelve pixels
- Posted
- Nov 18, 2026 · 0:14
Product that inspired this idea
Swapnil's post says he open-sourced image-to-reel, a Claude skill that turns a still into a 9:16 reel. No video model. Every frame is layers in Python, OpenCV, and ffmpeg. The repository is swapnilbhange/image-to-reel. The license is MIT. The only model step is a cutout, through Segment Anything, when something has to move in front of something else. Clouds drift, water ripples, cloth sways, particles fall. Faces are not warped. The readme keeps the motion small: a sway is 5 to 12 pixels, water is 1 to 5, a sky drift is 4 to 7. A pole or a standing leg needs no sway at all. The sound is synthesized. A 14 second reel, he says, renders in about 75 seconds on two workers. The post shows a picture that moves. The movement is a few pixels of compositing, not a generated shot. This page did not render a reel.
Posted by @swapnil_1_618. Watch the original launch. No separate product site is listed unless the post itself is the source.
- video
- claude
Related startup ideas
- Ask a coding agent to direct the launch film
A skill that turns a short prompt into direction for a launch film, using the coding agent as the director.
- Make videos with the coding agent you already use
A free Mac app for video, built so Claude, Codex, or another agent can drive it.
- The lower third, from the chat
An MCP server that builds and renders in a live After Effects project, with the tool count left as the disagreement between 65 and 73.
- A shared bench for short drama production
Script, assets, storyboard, and a video preview in one project more than one person can open.
- The cut list Claude wrote
An MIT cutter for one channel that keeps the picture and lets Claude choose the sentences.
- Voice, music, image, and video in one project
One project for the pieces of a video — voice, music, image, captions — instead of four exports.
Looking for more ideas?
The full list is startup ideas for 2026, each one tied to a launch you can watch.
Browse startup ideas for 2026