I’m not sure, but this may be the issue you guys are coming across:
When converting a floating point image to 8bit, if you don’t introduce any error diffusion (or dithering) you can end up with banding, which is really rounding issues on subtle gradients. It’ll look like this:
[/u] However once you introduce some some error diffusion/dithering (which is really just noise), before converting to 8bit, it will fix the rounding issues and get rid of the banding:
A lot of programs like nuke do this inside the viewer (when its set to 8 bit), but I always add a dither node to clean it up for final output.