We have thousands of photos in our collection, and if I were to guess, less than 25 percent are labeled or identified. Finding a labeled photo is certainly a treat, but most often, we are left guessing at who the person might be. If the photo is in a collection, that might help narrow down to a family. Clothing can help with a time period, as can vehicles, hiar styles and so forth. And, if it is stamped with a photographers name, it could help with a place.

After our Genelogical Gatherings talk about using AI, Kathy D set about using it to help identify photos in her collection and this is something I had not played with. So I dived in, just playing and learning a bit.

First, I asked ChatGPT which AI agent I should be using and it suggested this; No single AI does this perfectly. The best results come from a workflow using multiple tools, each doing what it’s good at:

• Human-guided visual analysis (clothing, setting, props, photographer marks)
• AI-assisted image analysis (objects, dates, locations, relationships)
• Careful facial similarity clustering (not true identification)
• Human validation and annotation

For a historical society, ChatGPT + a dedicated face-clustering tool + photo-archival software is the strongest and safest approach.

What AI can (and cannot) do with old photographs. AI does well at:

• Clothing and fashion era clues
• Studio vs. candid context
• Props (books, tools, medals, furniture)
• Image condition and format (daguerreotype, tintype, cabinet card, etc.)
• Facial similarity clustering (who might belong to the same family)

AI does not reliably do:

• Positive identification of historical individuals
• Genealogical relationships without external records
• Ethnicity or identity inference (and shouldn’t)
• This is important for a museum-grade workflow.

Best AI tools by task:

1. ChatGPT (image analysis) – Historical clue extraction

Best for:

• Analyzing each photo for historical signals
• Writing structured metadata
• Training volunteers how to “see” clues
• You can upload a photo and ask (you are assigning its role): “Analyze this photograph for date range, setting, clothing style, social class indicators, photographic process, and any clues that suggest family relationships or era.”

Strengths

• Excellent contextual reasoning
• Understands 19th–early 20th century clothing and studio practices
• Can output catalog-ready notes

Limit

Cannot do face recognition or large-scale clustering on its own. This should be your first pass tool.

2. Face-clustering tools (NOT identification)

These tools group similar faces without naming them.

Best options:

• Google Photos (private, non-public collections)
• Amazon Rekognition (face similarity only)
• OpenFace / FaceNet (technical, but powerful)

Use case: “Show me which photos may contain the same person or related people.”

Important
For a historical society, you must frame this as: Facial similarity clustering, not identity confirmation, and no publishing of names unless independently verified.

Recommended workflow (this is the key part)

Step 1 — Human + ChatGPT analysis

For each image:

• Upload photo

  Extract:

  • Estimated date range

  • Photo type

  • Clothing clues

  • Props

  • Setting

  • Photographer marks

• Save structured notes

Step 2 — Face similarity clustering

• Run photos through a face-grouping tool

• Assign neutral cluster IDs:

  • Group A

  • Group B

  • Possible sibling group

• No names yet

Step 3 — Cross-reference with records

• Compare clusters to:

  • Census age ranges

  • Studio locations

  • Known family donations

  • Other labeled images

AI suggests; humans decide.

Step 4 — Annotation and confidence levels

Every conclusion gets a confidence tag:

• Confirmed
• Probable
• Possible
• Unknown

Ethical and legal guardrails

• Do not label living persons
• Do not publish AI-identified names without documentation
• Be transparent: “Facial similarity analysis suggests these images may depict related individuals.”

1. Photographic format & studio practice (strong evidence)

All three images appear to be albumen prints mounted on card stock, almost certainly carte-de-visite (CDV) or possibly early cabinet cards.

Key shared traits:

• Identical or near-identical studio chair

  • Same carved wooden arm, same woven or textured upholstery

  • Positioned at the same angle relative to the sitter

• Plain studio backdrop, lightly mottled, no painted scenery

• Very similar floor–wall transition line

• Consistent lighting: soft, frontal light with minimal shadow modeling

👉 This combination strongly suggests the same studio setup, not just the same era.

2. Border, mount, and print style (moderate to strong evidence)

• All three have simple rectangular borders, no elaborate embossing

• Tonal range is consistent: mid-gray backgrounds, slightly washed highlights

• Contrast and grain structure match what you see from albumen paper sourced in the same period

While mounts can be swapped later, the printing characteristics look uniform—this argues for one photographer or studio using the same materials.

3. Clothing & hairstyle dating (strong evidence)

Approximate date range: 1865–1875

Here’s why:

Children’s clothing

• Boy (first image):

  • Short jacket with decorative buttons

  • Bow tie (not a long necktie)

  • Lace-up boots

  • Hair neatly parted—post-Civil War style

• Girls (second & third images):

  • Dresses fall mid-calf (not floor-length → pre-1880)

  • Simple bodices with high collars

  • Hair parted center and smoothed back (common in the late 1860s–early 1870s)

  • Boots and stockings typical of the period

Accessories

• The small basket or reticule held by both girls is especially telling

  • This was a studio prop frequently used for children in the late 1860s

  • The same prop appears in at least two images → strong studio link

4. Posing conventions (very strong evidence)

The poses are formulaic in a way photographers of the period often were:

• One hand resting on the chair

• Other hand posed deliberately (holding prop or resting at waist)

• Subjects positioned slightly off-center

• Neutral expressions (typical of long exposure times)

These are not casual portraits—they follow a house style.

👉 This is one of the strongest indicators of a single photographer or studio.

5. Age progression possibility (suggestive, but not provable)

You may be instinctively seeing this too:

• The two girls could plausibly be the same child, photographed a few years apart

  • Similar face shape

  • Same studio prop

  • Similar hairstyle and posture

⚠️ Important caution:
Without identifiers on the mounts (photographer name, town, date), this remains hypothesis, not conclusion—but it is a reasonable one.

6. What would clinch it (next steps)

If you have access to the backs of the cards, look for:

• Photographer’s imprint (name, town)

• Decorative stamp or logo

• Same card thickness and corner rounding

• Pencil inscriptions by the same hand

Even a partial imprint (e.g., just a town name) can anchor this precisely.

Bottom line (clear conclusions)

✔ Time period: Very likely circa 1865–1875
✔ Same photographer or studio: Highly likely
✔ Same family group: Plausible, especially among the children
✔ Same session: Unlikely (ages differ), but same studio over several years is very plausible

My conclusions – So, It did pick up on my notion that it is the same girl and she has the same bag. It also picked up on the chair. It helped me settle on a date range. From here, I can look for other images with the same chair or photographic style to at least group family members. I am no closer to knowing who they are. My next step will be to use either Google Photos or another agent to group faces.