Official visual evidence
Inspect a labeled target-reference-result workflow, a mapped group-photo workflow, two identity directions, and a reproducible decoded-pixel audit. Every example includes visible provenance and a clear limit on what it can prove.
Evidence disclosure
A useful example page should make provenance and limits as visible as the images. These assets explain the intended identity-and-scene relationship without turning illustration into an unsupported performance claim.
The six visual assets are DeepSwapAI illustrations: one single-photo target-reference-result composite, one mapped group-photo composite, two studio identity references, and two scene directions used in the current product experience.
No image on this page is presented as a customer result, user submission, testimonial, or record from a paid generation task.
The examples do not establish an accuracy score, average quality, generation speed, success rate, or guaranteed outcome.
Compare identity landmarks, hair boundaries, facial light, foreground overlap, and scene detail. Then test your own representative input before scaling.
Photo workflow example
The target image establishes the body, hair, pose, lighting, and background. The identity reference supplies the facial cues to place into that scene. The labeled result panel makes that input relationship visible before you test your own media.
Reproducible before-and-after audit
The source illustration is split into a 622 x 701 target panel and a 622 x 701 illustrative-result panel. Three aligned regions remain clear of the central identity-reference portrait and arrow overlays. The table reports decoded-pixel differences at native resolution; it does not score the face, run a generation task, or infer model quality.
| Region | Aligned window | Mean RGB difference | Pixels above delta 12 | Luminance correlation | Gradient correlation |
|---|---|---|---|---|---|
| Upper scene and face band | 0,0 · 622 x 230 |
16.658 | 74.963% | 0.93 | 0.378 |
| Face and hair window | 180,8 · 260 x 270 |
20.584 | 68.372% | 0.901 | 0.3097 |
| Lower subject and scene band | 0,480 · 622 x 221 |
13.996 | 64.64% | 0.9319 | 0.4188 |
The upper band contains the aligned kitchen scene, hair, and face while remaining above the reference-circle and arrow overlays.
Target (0, 0, 622, 230); result (622, 0, 622, 230)
The central window isolates the visible face and surrounding hair without including the reference-circle or arrow overlays.
Target (180, 8, 260, 270); result (802, 8, 260, 270)
The lower band contains clothing, arms, hands, and room detail below the reference-circle and arrow overlays.
Target (0, 480, 622, 221); result (622, 480, 622, 221)
The JSON contains the source fingerprint, panel layout, formulas, exact region coordinates, measurements, rights boundary, and limitations. CSV supports tabular analysis, and BibTeX provides a stable citation.
For a real generated output, use the separate human-review scorecard. For the complete machine-readable inventory, use the research data catalog.
Mapped group-photo example
This official composite illustrates the relationship between one group target, four separate identity references, and one multi-person result direction. The live multiple face swap workspace detects people first and displays the numbered reference-to-target mapping before submission.
Use the group-photo preparation guide to check visibility, spacing, numbered order, the 10-reference limit, and the current 3-credit-per-selected-face cost.
Identity direction A
The studio portrait supplies the visible identity cues. The scene direction places a related facial identity among flyaway hair, flowers crossing the lower face, bright sky, and dappled light.
Use the eyes, brow line, nose, mouth, jaw, and overall age cues as the comparison anchors.
Identity direction B
The second studio portrait supplies a different set of identity cues. The related scene direction uses the same flower-field language so viewers can focus on facial identity without confusing it with a completely different environment.
Compare the eye shape, brow arch, nose, mouth, jaw contour, and overall facial proportions.
Interpretation guide
Use the images to form a test plan, not to infer evidence that is not published.
| Question | What is visible here | Limit |
|---|---|---|
| Can I compare identity cues? | Yes. The reference and scene direction expose eyes, brows, nose, mouth, jaw, and age cues. | This is a perceptual review, not a biometric similarity score. |
| Can I inspect the target-reference-result relationship? | Yes. The photo workflow composite labels the scene target, identity reference, and illustrative result. | It explains input roles but is not a controlled before-and-after benchmark. |
| Can I judge video or GIF continuity? | No. These are static illustrative assets. | Test turns, blur, occlusion, cuts, and loop seams in the relevant motion workflow. |
| Can I predict my output or speed? | No. Results vary with the supplied identity reference and target media. | Run one representative image or short clip and review the actual result before scaling. |
From example to test
For deeper preparation, use the face swap quality guide. After generation, record the same visible criteria with the face swap quality scorecard. For consent, retention, and sharing decisions, use the privacy checklist. The verification methodology explains how public claims are reviewed.
Choose a workflow
Validate one identity and one target image.
Open photo workspaceTest consistency before processing a larger photo set.
Open batch workspaceReview continuity through turns, blur, and occlusion.
Open video workspaceInspect fast motion, short loops, and seam continuity.
Open GIF workspaceUse a clear identity reference and a target that contains the real lighting, pose, hair, or occlusion you need to evaluate.
