Disclosure I am still exploring this issue - but it certainly is not straight forward!
We have successfully been making image services from source imagery data files on premise for 6+ years. Big fan of the product and capabilities such as functions. We are migrating to the cloud.
Back when esri started down the Imageserver path it was sold as something where you could web enable you images "at source" and the power of the mosaic datasets and extension/now server would allow for .prj, mosaicking etc… no transformations, no copies of the data.
We just want create an image service - From mosaic dataset in file geodatbases stored on S3. We cannot connect to the S3 bucket from ArcGIS Pro or ArcMap as the as folder functions don’t support web accessible folders – ie it must be a mapped drive. have yet to figure out a mapped drive to an S3 bucket.
I have seen the optimizeraster repo - and support pushes me there. I am not strictly against that but there are a large number of imageservice that we have that are not huge and the data can transfer quite easily – even via the S3 browser – thus the optimize raster functions and transformations are just another list of steps (and derivations from the source data) and additional "software" on Github.... that should not be required for many (smaller) raster datasets to be web enabled from rasters in their original format in S3 storage.
#1 Optimize image publishing flows in core products from cloud stores
#2 If I must (but I don't like it) integrate Optimizeraster with core products - as it seems poorly integrated with the core product line and publishing flows of my desktop/server tools