But why can’t they be vector images? The pictures in the tweet looks like they could easily be replaced by vectors and the difference would be nearly unnoticeable.
Takes a lot of models and instructions to burn hundreds of megs that way.
But they should still be pulled on demand.
For reference, my own photos downscaled to 3840x2560 AVIF for sharing purposes range from 500KB to 2MB depending on how "busy" the image is. The off-camera 6000x4000 JPEG is ~7MB in comparison.
I stand by my opinion that a few hundred KB per image is plenty at the required resolution for model photos in an app.
So only twice the size of your JPEG example. Looking at the area chart of the files, it looks like the majority of the assets are less than that 2-3MB.
https://xcancel.com/pic/orig/media%2FGWAYmxabsAApjwu.png
There's probably over 200 or so assets in that 233MB archive.
That's slightly less than half the height and width of your phone's screen respectively, so a decent representative of a model photo dimensions (which are usually widgets smaller than the screen).
Model images would compress much better as they are edited to be much cleaner with smoothed solid colors and either solid or transparent backgrounds, and for something like a toothbrush you'd have an incredibly narrow image geometry making the final image likely much smaller.
Even if you'd consider this image too small or too low resolution, bumping those would not push us past 150 or maybe maximum 200KB.
(That repo has plenty of examples of other configuration examples if one is curious.)
Otherwise they save all the traffic and put it on the appstores and users and when they use some service for analytics they don’t need to run any server themselves. Oral-B probably sells worldwide putting everything in the app instead of fetching it maybe saves them good money.