What counts as similar content on Adobe Stock?

Adobe Stock refuses photos that are too similar to each other. Its examples include the same subject from slightly different angles, flipped or mirrored copies, crop variations, exposure brackets and filter or preset variations. Adobe suggests at most three variations of one photographed scenario, and says similar content is one of its top reasons for refusing content.

Key points

• Adobe calls similar content one of the top reasons it refuses content submitted to Adobe Stock.
• For photos, Adobe names five variations that do not count as distinct: flips, crops, exposure brackets, background colour changes and filter or preset changes.
• Adobe also counts a slightly different angle or distance, and sequential shots from the same moment, as too similar.
• Giving different images identical titles and keywords is itself listed as a similarity pattern.
• Adobe's pages give examples, not numbers, so there is no percentage of difference you can aim for.

What Adobe lists

Adobe's general page says it considers content too similar when a submission contains any of these:

Adobe's page on photos adds a list of technical variations that, although they are different files, do not create enough visual distinction to justify several submissions from one shoot:

Adobe's advice is to limit a single photographed scenario to three variations unless a clearly defined change, such as different lighting, a shift in setting or a distinct story beat, justifies more. It also asks you to tailor the title and keywords of each file to what is unique about it.

What we tested

A near-duplicate checker can only act on pixels. Ours computes a perceptual hash of each photo: a short fingerprint derived from the content of the image, which the pHash project describes as a fingerprint of a multimedia file derived from various features from its content. We use a DCT-based hash and a difference hash to find candidates, and a longer detail hash to confirm them. A mirrored copy is compared through the hash of the flipped image.

Generated scenes. We ran a generated photograph through the checker next to a set of variants, and repeated it on three different generated scenes: a landscape, a scene of objects and a dark, low-key scene. The result was the same in all three.

Real photographs. We repeated the test on 14 openly licensed photographs from Wikimedia Commons, each with eight variants made the same way. We also checked six pairs from a public noise dataset, each pair being the same scene shot at ISO 200 and at ISO 6400.

What the gap means

Adobe lists crop variations and slightly different angles as too similar. Our checker groups small crops but misses a 10% or 12% crop and a re-framed shot. So a batch with no groups reported is not a batch Adobe will see as distinct. The checker finds copies that were exported, resized, flipped or re-graded. It cannot find two shots of the same subject from a metre apart, which is the first thing on Adobe's list.

Two more limits follow from how it works. It compares only the files in the batch you give it, not your existing portfolio and not Adobe's collection. And Adobe's own comparison criteria are not public.

What to do about it

1. Keep the strongest frame from a scenario and leave out its flips, crops and filter versions.
2. If you submit more than one, change what the photo is about: framing, the subject's interaction with the scene, the setting or the story, as Adobe's page describes.
3. Write a separate title and set of keywords for each file.
4. Use the checker as a first pass on the batch for the copies it can find, then judge the different-angle shots yourself.

What to do next

Run the batch you plan to upload through the checker, then read Adobe's two pages again before a large submission, because they can change after the date we read them. For the file requirements Adobe checks first, see What are Adobe Stock's technical requirements for photos?. For how our checks work and what they leave out, read how Check Before Submit works.