We are producing ~100,000 x 100,000 pixel TIFFs with 60+ channels (~1 TB file size!) without problems. One key is to use the tiled storage format and ensure your processing tools can operate tile-by-tile to keep the RAM usage tractable.

Agree that works to a certain point, especially if you are working with mostly relatively planar data which can be processed slice by slice. Many issues only arise at scale and with certain types of data (e.g. ExM, EM) that can be dozens of TB in size. With large image volumes the tile-based approach is often insufficient. Instead you often want to perform operations on a chunk-by-chunk basis and reading or writing 3d chunks out of a set of TIFF slices is very slow. Tiled format and other optimizations can help, but ultimately a TIFF just doesn’t support the level of distributed parallelization that is needed to process and visualize these large data sets. This is why formats like N5 and Zarr were invented, and hopefully OME-NGFF will unite them into a standard going forward. But your point is well taken -- there are many types of microscopy workflows that don’t need to scale this way and probably never will. TIFF (or OME-TIFF) probably has the most amount of support through-out the tooling, and it makes sense to build modules for it.

Oh, excellent point about relatively planar data!