All Identifiers

How photo identification works

One standardized pipeline supports every tool, while the analysis instructions and response fields stay specific to the subject in the photo.

Methodology

How a photo becomes a result

Each identifier follows this shared pipeline while keeping its own subject-specific instructions, fields, and result presentation.

  1. 01

    Prepare the photo in your browser

    The selected image is decoded, scaled so its longest side is no more than 1024 px, placed on a white canvas, and prepared as a compressed JPEG for analysis.

  2. 02

    Apply the topic-specific method

    The server selects the topic-specific prompt, image-detail setting, model parameters, and strict response schema assigned to that identifier. A bird photo and a banknote photo therefore ask for different evidence and fields.

  3. 03

    Build structured candidates

    The response is parsed and normalized into the format defined for the topic. Depending on the identifier, that can be one best match or several plausible candidates with evidence, details, and an optional confidence estimate.

  4. 04

    Add reference media when available

    For supported topics, up to the first three results may be enriched with a separate reference image. Media is accepted only from configured providers and allowlisted HTTPS hosts; the text result still works when no reference image is available.

Reading a result

Evidence first, certainty second

A result is based on details visible in a single image. Lighting, angle, resolution, missing context, regional variation, and similar-looking subjects can all change the candidates returned.

Candidate lists are useful when several matches remain visually plausible. Single-result tools use a narrower schema when the task calls for one best match. If the required subject cannot be recognized, the structured result can be empty instead of forcing an identification.

Reference images are supporting material fetched after a candidate is named. They are not evidence that the uploaded subject and the reference are the same, and they do not change the model's confidence estimate.

Browse all identifiers →