The main change in 1.4: when Enhanced quality is on, AutoKeyWorder now runs a second AI call after the first one finishes. It looks at the actual image again, checks every title word and keyword against visible pixels, and fixes what is wrong before anything gets sent to your upload row. Full disclosure, we built AutoKeyWorder.
Here is what it catches.
The problem it solves
The first AI pass is good at describing what it sees at a glance. The failure mode is confident fabrication: terms that sound plausible for an image but are not actually shown in it.
The most common category is location. A portrait of a giraffe generates keywords like “savanna”, “africa”, “grassland” because that is where giraffes live, not because the savanna is in the frame. The background is blurred green foliage. There is no open terrain visible anywhere. Adobe’s content policy flags inaccurate metadata, and these geographic claims are exactly the kind that trigger rejections.
Other patterns: “white background” on an image with a light gray gradient. “Plastic plate” on a paper plate. “Aerial drone shot” on a video clip where nothing moves.
The second pass has one job: judge only what is visible. If a term cannot be confirmed against the actual pixels, it gets dropped.
What changes for photos
Location and biome terms get cut when not shown. Savanna on the giraffe portrait goes. Forest behind a person when the actual background is blurred bokeh goes. Ocean on a boat product shot where no water is visible goes. These are the terms most likely to earn a rejection flag and least likely to bring real buyer traffic anyway.
Material and background are corrected, not just dropped. If the plate is paper, the keyword changes to “paper plate” and “plastic plate” comes out. If the background is red and the title says “white background”, both get corrected. The title and keyword list stay in sync: a term the second pass removes from keywords also comes out of the title, so you never end up with a word in the title that has no matching keyword.
The top 10 slots get the right terms. Stock platform indexing weights the first 10 keywords most heavily. The second pass checks whether the primary subject is actually in that window. A close-up texture asset where “texture” ended up at position 22 gets it promoted to the top. An asset with a clear medical examination context gets the relevant concept verified and placed where it has the most search impact.
Synonym clusters get trimmed. If the first pass writes “peeling”, “flaking”, and “chipped” for a single peeling paint texture, two of them come out. The concept is real and stays; the redundant variations do not.
A typical clean batch sees 0 to 3 changes per image. The second pass is a scalpel, not a rewrite. Most assets come back with a few dropped location terms, occasionally a corrected material, and the primary subject nudged into the top 10 if it drifted.
What changes for videos
Video metadata gets two additional checks that photos do not.
Motion is verified against the frames, not assumed from the subject. AutoKeyWorder samples frames from the clip and compares the first against the last. If the title says “aerial drone shot” but the frames show a static locked-off angle with no position change, the aerial and drone keywords come out and the title is corrected. The reverse also applies: a genuine slow aerial hover gets kept even if the frame-to-frame difference is subtle.
Titles are rewritten to lead with the action. Buyers searching for video footage filter by camera move and subject motion. A title like “Mountain valley at dusk” does not tell them anything about what the camera does. The second pass rewrites video titles to verb-first format: “Holding static over a mountain valley at dusk” or “Panning across a city skyline at dusk”. The subject is still there; it just comes after the actual motion, which is what a video buyer is searching for.
The three keywords the second pass always removes from video: “video”, “footage”, “clip”. They are implied by the medium and add no search value.
How to use this
Enhanced quality needs to be on in your popup. If the toggle is off, only the standard single-pass runs. Flip it on before your batch and the second pass runs automatically on every asset, including video.
Enhanced quality costs two credits per image instead of one. On a tight credits budget, standard mode is accurate enough for most batches. The second pass earns its cost on conceptual images (medical, wellness, business), nature photography with blurred backgrounds, and all video work where motion terms and title format directly affect where the clip ranks.
For a refresher on every popup control, the complete AutoKeyWorder guide has the full breakdown including platform-specific toggle behavior. If you want to understand the title changes that shipped a few updates back, that post covers the Context field and how typed titles shape your keywords.