Client
Text Behind SubjectTimeline
Product developmentProduct
In-browser photo and video editorIndustry
Creator ToolsDesign SoftwareMedia ProcessingPutting text behind the subject of a photo means cutting the subject out first, and the tools that do it well send the image to a server. That rules the technique out for anything sensitive, adds an upload wait, and attaches a monthly fee to an effect that is really one composite. Text Behind Subject had to run segmentation, text compositing, video export, and object removal on the user's own machine, with no build step, no account, and no API key.
Composited in three layers, background then text then the alpha-cut subject on top, so the whole effect is a draw order rather than a per-image mask edit.
Shipped six segmentation models and picked the default by measuring what the browser can actually do, returning BEN2 where WebGPU reports shader-f16 and BiRefNet lite 512 everywhere else.
Added a depth mode for scenes with no single subject, and a paint-by-hand path for images where neither model asks the right question.
Made geometry resolution-independent, with position as a fraction of the image and sizes as percentages of font size or image height, so the preview canvas and the 4K export run identical draw calls.
No image or video leaves the user's machine, and there is no account, no upload queue, and no API key.
Export at full native resolution, up to 8192px for stills and 4K MP4 for video.
About 5,900 lines of vanilla ES modules across nineteen files, with no bundler and no install step.
Undo that counts a typed word as one entry and a slider drag as one entry, instead of eight and a hundred.
