Have you tried MyHeritage’s Scribe AI since they launched it in March?   Unfortunately, I was not at the debut at RootsTech, but I’ve played around with it a bit since then.  If you’ve tried it yourself, or if you are wondering what all the buzz is about, read on!

Scribe AI, in a nutshell, is AI without the prompting.  Prompts are the instructions that you give to your AI program to tell it what to do.  Writing effective prompts is a skill that takes some practice, but MyHeritage has taken all the guesswork out of the system.  They’ve already written the prompts.  You just upload a file to the Scribe AI page on their website, and the program will analyze and report back.

For this newsletter, I tried uploading a photo of the cutest brick wall that I’ll probably never solve, since the subject is so young.

This was one of the collection of 1500 or so family photos you may have heard me talk about earlier.  I have no idea who this little boy is, and I had no expectation that Scribe AI would be able to give me a name.  But I was hoping that it would generate some clues for further research. 

The vast majority of the photos in my collection are from the New York City area, which is where this branch of my family settled in the United States.   Probably about 10% are from various travels, either for vacations or military service.  A handful were sent from relatives overseas.  I suspected that this photo was taken in Brooklyn, NY.

The middle building in the background of the second photo had a unique and recognizable tower on its roof, which allowed me to identify the area as the Coney Island neighborhood. Cross-referencing available census data allowed me to hypothesize which branch of the family took this photo.  Of course that tells me nothing about who this little boy is.  A cousin?  A neighbor’s child?  Several generations of my family lived in this neighborhood from the 1920’s through the 1940’s, and I wasn’t sure which generation I was even looking at.

After uploading the photo, Scribe took about 30 seconds or so to think.  It then returned data divided into the categories below. For a “normal” program, such as Claude or ChatGPT, I would have to write my own prompts to get this data, but Scribe has them baked in:

Key Findings – A summary of all the details in the categories below.

Historical Background – Identified the photo as 1930’s-40’s United States, based on clothing, building structure, environment, etc.

Estimated Date – 1935-1945, based on the boy’s clothing and the vintage baby carriage in the background.

Estimated Location – without any prompting from me, Scribe AI narrowed the area down to “North American city (likely East Coast or Midwest, such as New York City, Chicago, or Philadelphia)”, based on the architecture of the buildings in the background.

Visual Clues – a narrative description of the photo, which included things I hadn’t noticed myself, such as the baby carriage in the background and the young girl writing on the sidewalk to the far right.

Technical Details – this discusses the type and condition of the photo itself.  “The image is a faded black-and-white (now sepia-toned) gelatin silver print snapshot with a white border. ”

In general, Scribe AI confirmed what I suspected in terms of the era and location of the photo.  I was glad that we were in agreement, but had hoped for a little more direction.  I had a second photo taken at the same time, so I uploaded it.

I need to reiterate: cutest brick wall ever.

Scribe AI analyzed this photo and returned much of the same information as in the previous one.   This isn’t surprising since the basic background and the boy are the same.  There was a small shift in the background that revealed the restaurant sign, and of course, there was the little girl.

The program didn’t do much with the girl. There was a description of her age, clothes, and the body language between the two children, but nothing that I could use for further research.

The larger addition was the program’s analysis of the restaurant sign in the background, although it wasn’t terribly accurate.  Scribe AI transcribed it as “Tarvernette Restaurant”, although the actual wording is “Tarvernelle Restaurant”.  There also seems to be some markings, perhaps another word, after Restaurant.  I haven’t been able to conclusively decipher what it is, but Scribe AI didn’t mention its existence.  What I did get were some references for further research that I hadn’t necessarily thought of, such as checking for the restaurant in a previous census.

I was a little disappointed to note that Scribe AI did not let me link these two photos in any way.  I knew that they were from the same set, and if I had a way for the program to associate the two, I may have gotten something new.  Who knows?  It is a feature that I hope makes it into a future release.

Scribe AI can also interpret and analyze documents, using the same method that it does photos. The categories of data returned are different, which is to be expected. For documents, they are:

• Details Mentioned
• Key Findings
• Suggested Next Steps
• Full Text Transcription
• Translation (if the document is in a foreign language)

Again, these prompts are pre-configured by the program developers. No direction needed on your part!

Have you used Scribe AI at all? What did you think? Let’s talk about it in this month’s

FTT Office Hours on June 20th, at 2pm EDT. The link to the Facebook event is here. If you’re not on Facebook, send an email to info@familytreetech.com, before 1pm on Saturday the 20th, and I’ll send you the link.