Review AI analysis of photographs and faces
What you will do
Section titled “What you will do”Run Arams photo and face analysis on a representative batch, use the visible quality filters to find photographs that need attention, and inspect the original pixels before deciding what to keep. The verified example compares a photograph that Arams included under Closed Eyes with the next photograph from the same burst.
This workflow was verified in Arams 1.10.2 on Windows with the English interface. The project contained 80 full-resolution JPEGs, including maxine-01.jpg and maxine-02.jpg at 6064 × 4040. Arams reached Analyzed: 80 / 80 and Analysis completed. The Closed Eyes filter returned 15 photographs.
The result is a review queue, not an automatic keep-or-reject decision. In the verified batch, many results visibly had closed or lowered eyes, while milana-dsc08137.jpg had open eyes and was still included. Treat every analysis label as an AI suggestion and check the photograph yourself.
Before you start
Section titled “Before you start”- Sign in to Arams and use photographs that you are authorized to send for server analysis.
- Keep the source photographs at full resolution. Do not resize face photographs before analysis.
- Keep the originals unchanged. This procedure does not require retouching or export.
- In Settings > General, make sure Automatic photo analysis and Automatic face search are on. If Arams shows its server-analysis notice, read it and choose I Agree only when sending the selected photographs is appropriate.
- Use a short, varied batch rather than a single portrait. Nearby frames from the same burst are useful for checking expressions, focus, and motion.
The verified run used neutral, full-resolution JPEG conversions from authorized originals. It also included maxine-01.jpg and maxine-02.jpg without resizing. No photograph was retouched or altered to create an analysis result.
Run photo and face analysis
Section titled “Run photo and face analysis”- Create a practice project and import the full-resolution photographs.
- Select the photographs that you want to analyze.
- Check the analysis control. If it reads Analyze, choose it. If automatic analysis has already started, let it continue instead of starting another operation.
- Wait until the analyzed counter covers the complete batch and the button reads Analysis completed. Do not start Retouch preview.
- Confirm that the project count still matches the imported batch.

In the verified run, automatic analysis covered all 80 photographs. Completion confirms that Arams finished the requested task; it does not confirm that every label or displayed score is correct.
Inspect image- and face-quality filters
Section titled “Inspect image- and face-quality filters”- Expand Image Features in the left panel. On the verified release, the visible rows were Overexposed, Underexposed, NSFW, Plain Background, Textured Background, Bad Pose, Motion Blur, Kissing, Outdoor, and Aesthetics.
- Expand Face Features. The visible rows included Occluded, Cropped, Bad Light, Out of Focus, Bad Mouth, Bad Eyes, Eye Contact, Smiling, Profile, Turned Away, Looking Down, and Closed Eyes.
- Use the eye controls on one row to show or exclude that result, then check how the photo count and visible filenames change.
- Reset that row before trying another filter.

The number shown beneath a thumbnail is a displayed gallery score. Do not interpret it as the confidence of the active filter: Arams did not expose a documented confidence value for Closed Eyes in this check.
Find closed-eye candidates
Section titled “Find closed-eye candidates”- Under Face Features, use the show control on the Closed Eyes row.
- Check the project count and record the filenames that remain. The verified batch changed from 80 photographs to 15 candidates.
- Review the grid only as an index. A small thumbnail is not enough to judge an eyelid or eye direction.

The verified results included clear closures, partly closed or lowered eyes, and at least one incorrect result. A filter match means “inspect this photograph,” not “reject this photograph.”
Inspect a face at full resolution
Section titled “Inspect a face at full resolution”- Open one candidate in the large viewer.
- Wait for Loading… to disappear. With the photograph viewer focused, press Z to toggle zoom, then use the mouse wheel over the face to choose a useful scale. Z is the default shortcut; if you have customized your keys, check keyboard shortcuts.
- Judge what the pixels show: closed, partly closed, obscured, looking down, or open.
- Record your visual judgment separately from the Arams filter result.

The illustrated candidate is milana-dsc07266.jpg. Arams included it under Closed Eyes, and visual inspection of the full-resolution 4024 × 6024 JPEG confirmed that both eyes were closed or lowered in this frame.

At 583%, the real Arams viewer keeps the filename, active Closed Eyes filter, and surrounding interface visible while making both eyes readable at normal documentation width. This enlarged view is still the original photograph in the application, not a cropped or enhanced derivative.
Check faces in the right panel
Section titled “Check faces in the right panel”You can also inspect a face without zooming the whole photograph:
- Select the photograph and click the face-outline icon in the far-right toolbar to open Details and Faces.
- Expand Faces if necessary. Beside its heading, choose the split-rectangle icon to switch to the larger face view.
- Inspect the eyes in the enlarged face preview. Check the filename so that you are judging the intended photograph.
- Look at the analysis icons associated with the face. They indicate detected issues, such as closed eyes, and help direct your attention to details that need checking. Compare each indication with the visible face: an AI warning is a suggestion to inspect, not a reason to reject the photograph automatically.
This is especially useful when reviewing Closed Eyes results: keep the overall composition visible while checking the face in the right panel. The face preview is not a guaranteed 1:1 view of the original pixels. For precise image inspection, use the main viewer and zoom. See Inspect detected faces for the illustrated panel controls.
Compare a nearby open-eye photograph
Section titled “Compare a nearby open-eye photograph”- Return to the gallery and reset Closed Eyes so the full batch is visible again.
- Open a nearby frame from the same burst.
- Compare the eyes, expression, focus, motion, and obstruction around the face.
- Prefer the stronger photograph only after checking the full image; do not base the decision on the filter or score alone.

The nearby frame milana-dsc07267.jpg is the next file in the burst. Its eyes are visibly open. This pair demonstrates a useful review pattern for these two photographs only; it is not an accuracy claim for other shoots.

The comparison uses the same 583% viewer scale and the same face position. That makes the visible eye state comparable without turning either photograph into evidence that the AI label is generally reliable.
Review borderline or incorrect results
Section titled “Review borderline or incorrect results”- Inspect every remaining candidate in the large viewer.
- Keep partly closed eyes, side glances, lowered eyes, occlusions, and hard-to-read faces in a borderline group until you make a visual decision.
- Record matches that visibly disagree with the photograph. In the verified batch, milana-dsc08137.jpg had open eyes looking to the side but appeared among the 15 Closed Eyes results.
- Reset the analysis filter and confirm that all 80 photographs are visible before continuing the cull.
- Use ratings and selection only after this visual review.
The operator review retained a representative set from the 15 results, including clear closures, lowered or partly closed eyes, the open-eye comparison, and the incorrect open-eye result. No result was edited or injected to make the example succeed.
Check the result
Section titled “Check the result”The analyzed counter covers the intended batch, Image Features and Face Features expose the observed filters, and each candidate used for a selection decision has been opened at a useful size. The closed-eye example and its nearby open-eye frame have distinct filenames, and any borderline or incorrect matches remain identified as AI suggestions rather than facts.
If it does not work
Section titled “If it does not work”- Analyze is unavailable: check whether the control already says Analysis completed and whether the counter covers the batch. Otherwise, confirm that the photographs are selected and record the exact disabled state or message.
- Analysis does not complete: preserve the visible counter, task state, and error. Do not replace the result with a simulated completion state.
- The analysis filters are missing: confirm that Automatic photo analysis and Automatic face search were enabled before import and that analysis completed. Reset ordinary filters and reopen Image Features and Face Features.
- Closed Eyes returns no photographs: do not manufacture an example. Review other observed indicators or use another authorized batch.
- A candidate has open eyes: keep it as an incorrect result and make the culling decision from the photograph.
- The face is too small to judge: open the large viewer and inspect the source at a useful scale. Follow Inspect detected faces when you need the face panel.
- The account cannot analyze the batch: record the exact capability or server message. Do not purchase credits, top up the account, or change subscription settings.