I think the image entropy option is VASTLY superior to other methods (TONS of entropy, very simple, no proprietary RNG hardware, trivially easy & fast for the user, no opportunity for user to screw something up) but with one caveat: you can't take the input data and verify the results in any other tool.

For example, you can enter your dice rolls into a SeedSigner, Coldcard, and iancoleman.io (with the right settings) and get the exact same result.

But, by design, there's no way to export the image entropy data that's used to create the seed (we didn't want to create the temptation to save the image to a file and use that jpg as a backup; bad idea).

So if you can't verify individual results, it becomes very important to at least verify that the release img is authentic and, to whatever extent possible, review the code itself or talk with someone you trust who can review the code.

Also note that we actually incorporate a lot more than just the pixels of a single image into the entropy used to create the seed. It's really bonkers overkill.

https://i.nostr.build/lLMln.png 

 Awesome, thanks Keith for the input.  Yeah, that's what I was thinking too, makes sense that the photo is super random.  

I've been experimenting with the QRs too... You guys ruled out doing a straight bip39 English words qr? My concern is if I'm not around in the future and wife will try to extract seed using phone cam qr program.  ... Maybe practice and education is the solution there. 

 The SeedQR formats are seeing some pretty good adoption across the bitcoin ecosystem (e.g. Sparrow can directly read them and turn them into a hot wallet, if need be). Plus the format is well documented; the "Standard" SeedQR could even be decoded by hand if you just scan it with a cell phone that'll read out the digit stream. The "Compact" format is a little harder for humans to work with.

The actual English words would bump the QRs up to the next size or two since words take up more data space than digits. If you export a test seed as Compact SeedQR, Standard SeedQR, and if you just paste the full mnemonic into a generic QR encoder, you'll see what I mean when you compare the three QRs.

And with SeedQRs size matters since the point is to mark them by hand. A Compact SeedQR for a 12-word mnemonic is just 21x21. Takes like 5 minutes to with a Sharpie. But a 24-word Standard SeedQR is 29x29. Huge difference. That takes more like 20+ minutes to transcribe. Not fun.