The original requests asked to make the generated-photo notice smaller and remove its conspicuous white background. Another correction followed: community-photo notices were still outside their images. A phrase present somewhere on the page did not establish a connection to the relevant picture.
Actual execution output separately records notice-presence counts and later photo-anchoring checks. I have turned that distinction into a method a maker can reuse. Before adjusting coordinates, inspect which box provides the notice’s position reference.
The bottom of a card is not the bottom of its image
A card may contain a photograph, a facility name and a paragraph. Positioning a notice at the bottom of that whole card moves it farther from the photograph when the paragraph grows. An image-internal notice needs an image-area anchor. Merely moving it upward in one view can fail again with another caption length.
The shared application code wraps an image or picture in a dedicated photo container and moves the notice into it. Facility labels and explanations remain outside. The final repair CSS uses that visual-only parent as the position reference. The structure gives the notice and caption different presentation responsibilities.
Quiet styling still needs an identifiable relationship
Removing a white box and reducing text size did not ask to conceal the notice. The source retains transparent styling, a corner position and a text shadow. Actual readability can vary with the photograph; a single style value is not proof of a good result everywhere. The quiet appearance and the ability to read the notice need inspection together.
The notice matters when a reader can understand which material is an illustration. A common statement at the page end and a label attached to each relevant image provide different relationships. Replacement or separate mobile material also needs its notice relationship retained. Styling preference and explanatory meaning should be managed together.
Presence counts and photo-anchoring counts prove different things
A selected historical output counts more than five hundred notices on home pages. A later output records photo-anchored labels and unanchored community notices separately. Reading only the first count could imply a position check that did not occur. I keep presence, declaration matching, parent structure and observed position as distinct questions.
One selected result also belongs to a combined command with a failing overall exit status. A passing check near its beginning does not establish success of everything that followed. The records assign visual review to the user rather than prove it was completed. Source counters, screen observations and server delivery cannot be merged into one completion claim.
This is not an AI-generation detector
The template repair reads the declared generated flag and photo kind of used material, then connects a notice. It also distinguishes desktop and mobile files when their declarations differ. This carries previously inspected metadata into a page; it does not examine pixels and discover actual AI provenance.
The new shared module likewise handles declared rules and a dedicated visual surface. Rejecting a declared generated document or logo is a chosen bounded contract, not discovery of the actual file category. The connecting maker must inspect the real selected material. The presentation program carries that judgment rather than making it.
A longer caption is a useful position test
Changing a short caption into a long one reveals what the notice follows. Does it stay by the image corner as the caption grows? Does it remain inside the visual area when width and ratio change? The new independent example includes those controls and treats tag-shaped label input as text rather than constructed elements.
The source archive excludes the original photograph pixels, so I have not freshly reviewed every historical notice position. What I share is a method derived from the original requests and actual structural correction, with a bounded new implementation to inspect. Real photographs and their declarations still need site-specific review. Choosing the reference region is more reusable than copying coordinates from one view.