Hand photo guide: lighting, framing, and privacy

Use even front light

Light the full hand evenly from the camera direction or slightly to one side. Brighter is not always better: soft daylight or a diffused room light usually preserves finger outlines more effectively than one small, intense lamp. Check the preview to ensure the wrist, joints, and fingertips do not disappear into dark areas.

Avoid glare, deep shadow, and backlighting. Shiny skin, a glass table, or direct flash can produce white highlights that erase joint edges. A window much brighter than the hand can turn the hand into a silhouette. Moving the light or the hand is usually more useful than applying an image filter.

If automatic exposure keeps changing, hold the hand still for a moment before taking the photo. Skin tone is not a success criterion. The important qualities are a visible boundary between hand and background, highlights that retain detail, and shadows that are not crushed. HandFuture does not evaluate anyone's skin appearance.

Frame one complete open hand

Keep everything from the wrist through all five fingertips inside the frame. Open the hand naturally and leave some space between fingers. Aim the camera toward the palm and avoid bending fingers sharply toward the lens, which can create overlap. A margin around the hand gives the model more context than a tightly cropped fragment.

Choose a plain background with contrast against the hand. Patterned fabric, similarly colored clothing, and unrelated objects add visual clutter. Include only one hand and avoid overlap from a second hand, another person's hand, or an arm. The interface treats multiple-hand output as a state that needs a new photo.

Rings, watches, and bracelets do not always prevent detection, but move or remove them if they cover the wrist or joints. Never force an injured or uncomfortable hand to straighten for a photo. Comfort and safety matter more than detector success, and the tool does not require a special gesture.

• Show one hand completely from wrist through every fingertip.
• Open the palm naturally and minimize overlapping fingers.
• Use a plain, contrasting background.
• Move jewelry or objects that cover joints.

Files and privacy

The implemented uploader accepts JPEG, PNG, and WebP with a 10MB limit for one file. Both the extension and file contents must describe an image the browser can decode; renaming another format does not convert it. If a phone photo is too large, use the device's built-in tools to reduce its dimensions or export a supported format.

After selection, FileReader reads the image in the current browser page, and HandFuture has no photo upload endpoint. The image remains in application memory for the model during this session. Resetting, closing the tab, or refreshing removes that image state. Disclaimer acceptance is the only related preference stored separately on the device.

Data minimization still matters even when a photo is not sent to HandFuture. Keep faces, identity documents, address labels, screens, and identifying surroundings out of the image. Do not use a hand photo of someone who has not agreed. On a shared device, close the tab when finished.

Troubleshooting

For a no-hand result, first confirm that the wrist and fingertips are visible, then move to even lighting and a more contrasting background. For a multiple-hand result, reframe so only one hand remains. Both messages are detector outputs, not judgments about image quality or the person shown.

If the browser cannot read the file, check the supported formats and 10MB limit, then try exporting the image again. A file downloaded from a social app can have a familiar extension but incomplete contents; opening it in the photo library and saving a new copy may resolve some encoding problems. HandFuture does not preserve a half-finished analysis after an error.

The model must load required resources on first use. A network interruption, workplace or school content policy, privacy extension, or CDN block can prevent initialization. Refreshing or trying later may help. If the environment continues to block the model, do not repeatedly submit sensitive images; all guides remain readable without detection.

Better lighting and framing can give the model a clearer standard hand shape, but cannot ensure success for every photo. Device performance, camera blur, unusual poses, and model limits all affect output. Successful detection means that joints were located; it does not mean palm creases were identified or interpreted.

Sources

• MediaPipe Hand Landmarker
• MDN FileReader