Gen AI
The generative bench. Two tools, at two different stages: one is finished and one is deliberately not, and the difference is written on each of them rather than left for you to discover.
Video → image
Pull still frames out of a video file, in the tab. The browser's own codecs do the decoding, a canvas does the encoding, and the file never leaves the machine. Pick one moment, or take a strip spaced evenly across the whole clip.
Drop a video here
or choose a file — MP4, WebM or anything else this browser can decode
Scrub to the frame you want, then grab it at full resolution.
Spaced evenly across the clip, inset from both ends — the first and last timestamps are usually undecodable.
Frames come out at the video's own pixel dimensions.
The video is decoded by this tab and never uploaded. Browsers ship different codecs, so a file one browser refuses may open in another.
Image → HTML
This does not turn a design into markup. Doing that means inferring layout from pixels, which needs a vision model — and shipping one would mean either uploading your image or loading hundreds of megabytes of third-party runtime into the page. Neither is something this site does.
What it actually does
Everything about an image that is honestly computable in a browser tab, and nothing more: its true pixel dimensions, the colours it is mostly made of, and a self-contained HTML scaffold with those values already filled in and the image embedded.
The generated file carries a comment at the top saying which parts are measured and which are not, so it cannot be mistaken later for a conversion of the design. When a local vision model is ready — that is what the small models work is for — the layout half lands here and this notice goes away.
Drop a screenshot or mockup here
or choose a file — PNG, JPEG, WebP or AVIF, up to 40 megapixels