Text Behind Image

About the Text Behind Image

Make the popular “text behind the subject” look for posters, thumbnails and posts: a big title that sits behind the person, the pet or the mountain in your photo but in front of the background. Find the person cuts people out automatically with an AI portrait model (MODNet); for anything else, Tap an object and MediaPipe’s MagicTouch selects it — tap more objects to add them, or tap a part to take it out.

Add as many texts as you like and style each one: font, size, colour or gradient, opacity, rotation, shadow or glow, outline, letter spacing and even a curve. Drag them live; each can go behind the subject or in front of it. Save a full-size PNG, JPG or WebP. The models run on your device — your photo is never uploaded.

Common uses

• YouTube thumbnails and video covers with a bold title behind the presenter.
• Instagram posts, stories and reels covers with the “text behind subject” style.
• Event, product and travel posters.
• Magazine-style covers, album art and wallpapers.

How the subject is cut out

• Find the person uses MODNet, a portrait-matting model (Z. Ke et al., AAAI 2022; Apache-2.0), refined along the photo’s own edges so hair and shoulders stay natural.
• Tap an object uses MediaPipe’s MagicTouch interactive segmenter (Apache-2.0): the object under your tap is selected. Tap several times to add more objects, or switch to Tap to remove to take a part away.
• Edge: shrink or grow removes a halo or a missed rim; Edge: soften blends the outline (1 px is a good start).

Both models run in your browser: the photo is never uploaded. The portrait model (13 MB) runs in ONNX Runtime with WebAssembly (14 MB); MagicTouch (6 MB) runs in MediaPipe’s WebAssembly runtime (12 MB). They are downloaded from this site once.

Making the effect look good

• Use big, bold letters: Anton, Bebas Neue or Montserrat Bold at 20–40% of the photo’s height.
• Let the subject cover a part of the text — the top of a head over the middle of a word works best. If the subject covers too much, move the text up or make it smaller.
• Pick a colour that stands out from the background, and a soft shadow so the letters do not look pasted on.
• Opacity below 100% lets the background show through for a subtler look.
• Photos where the subject stands in front of an open background (sky, a wall, a street) work better than crowded scenes.

Limitations

• The models can miss parts of the subject (fine hair, see-through or very thin parts) — check the edge, and tap to add or remove parts.
• MagicTouch selects the one object under a tap; tap again for other parts.
• Photos over 16 megapixels (8 on phones and tablets) are reduced to that size.
• The built-in fonts cover Latin script; Noto Sans and Poppins also Devanagari (Hindi, Marathi), and Noto Sans and Inter also Cyrillic and Greek. Other scripts use your device’s fonts.
• The first cut-out downloads 18–27 MB. Edits are kept only while the page is open.

Privacy

Your photo is processed in your browser and never sent to a server. Finding the subject downloads the AI engine and a model from MySmartCoPilot — not from a third party — the first time; after that your browser’s cache serves them. Saved images contain no EXIF data from the original, such as the camera or GPS location.