Source: https://restorebunny.com/guides/use-reference-photos-as-evidence/ # Use Reference Photos Without Turning Similarity into Proof Build a corroboration sheet for related family photographs, separating reference identity, independence and the particular feature each source supports. Family photo projects4 min readUpdated 11 October 2026 ## Try it on your own photo Repair a copy of your photo, then compare faces and details with the original. [Restore my photo]()1 credit per successful photo ยท Account required[See credit packs]() Quick answer ## The short version Use references to answer specific questions, checking their identity, date and independence first. Compare the unedited sources and record what each reference actually supports. A similar face, jacket or building can suggest a lead without proving that an uncertain photograph shows the same person or event. ## Choose references around a question A folder of attractive pictures is not automatically a useful evidence set. Start with the question: does this image show the same house, is the inscription readable elsewhere, or is a claimed person's appearance consistent with a documented portrait? Choose references that can address that question and record where each came from. Leave a reference whose identity is uncertain labelled uncertain. The Library of Congress' primary-source analysis materials encourage testing hypotheses and revising conclusions with additional sources. Apply that approach to family photos. Do not ask a reference to prove more than it records. A labelled portrait can help compare visible features, while a picture from another decade may be poorly matched for hairstyle, age or an accessory. ## Check whether the references are independent Two files may be crops or edits of the same scan. They repeat the same source information even if their names differ. Track the parent image before treating several views as corroboration. A restored output can introduce apparent similarity that was not in the capture, so return to masters for identification-related comparison. Then examine dates, pose, lighting and source quality. A profile and a frontal portrait show different features. Small, blurred faces can look alike because the distinguishing detail is absent. An object such as a coat or necklace may belong to more than one person. These limitations do not make comparisons useless; they define what remains a clue rather than a confirmed identification. ## Worked example: three images provide two sources A fictional research set contains a labelled 1946 portrait, a cropped version of that same portrait and an unlabelled group photo believed to be from 1948. The two portrait files are one underlying source. The group photograph is a separate object, but its date and names need their own evidence. Your sheet records two sources, not three independent confirmations. The labelled portrait supports a comparison of a visible ear shape and spectacles, while the crop adds only viewing convenience. If the group face is too small to show those features, write that the comparison is inconclusive. Do not use an AI-enhanced group face that resembles the portrait to close the gap: the enhancement may have supplied the feature being compared. ## Make a reference corroboration sheet 1. State the question and the target photo ID. Identify the precise person, object or text region under review so the comparison is reproducible. 2. List each reference ID, provenance, date basis and identification basis. Mark unknowns explicitly instead of letting a familiar filename stand in for evidence. 3. Trace dependencies: source, crop, restored derivative or repost. Count underlying independent sources separately from convenient viewing files. 4. Compare one feature at a time at matching scale. Record visible agreement, contradiction and features that cannot be assessed due to pose or quality. 5. Write a bounded conclusion and next check. 'Consistent with the same building, entrance not visible' is more useful than declaring a complete match from a general resemblance. ## Keep evidence use separate from visual inspiration A reference can also inspire an interpreted restoration, but that is a different use. If you supply another portrait to guide an artist or supported editing workflow, document how it was used and check the result against the target source. A generated image resembling the reference does not strengthen the historical claim about the original. Keep the reference sheet with research notes and share a simplified comparison only when contributors permit it. Avoid exposing unrelated personal information from captions or back notes. A good reference set narrows a question, reveals a contradiction or explains why you cannot decide yet. It does not have to produce a confident name to be valuable. Decide in advance what would count against the proposed match. If every difference is explained away but every similarity is accepted, the comparison cannot meaningfully test the idea. Put contradictory features in the same worksheet as supporting ones and allow an inconclusive outcome. A reference set should make revision easier, not defend an initial guess. ## FAQs Do multiple crops count as independent evidence? No. They derive from one source, even when separately exported or labelled. Can an AI-enhanced face confirm a match? Use the unedited source for the evidence comparison. Generated detail may create the resemblance you are trying to test. Must a reference be from the same year? Not always, but date and age differences limit which features it can support. Record that limitation. ## Sources Preservation and technical guidance reviewed for this article. - Library of Congress: [Primary source analysis tools]() ## Related - [Trace which files share a source]() - [Compare edits with source identity clues]() - [Make an identification contact sheet]() ## Ready to restore your photos? Upload an image, choose your settings, and preview results with a before/after slider. [Restore my photo]()[View Pricing]() RestoreBunny Team Photo workflow editors Guides follow our [editorial policy]() for sourcing, product claims, AI limitations, updates, and corrections.