Thanks for the samples, Moose. Topaz Photo AI technically looks very
much like the Codeformer and GFPGAN I'm playing with. There are
certainly lots of room for improvement but when it works, the result is
just brilliant. I have hundreds of old images run through both
Codeformer and GFPGAN, around 90% of image works better with Codeformer
and 10% works better with GFPGAN. For the one with trouble I fix them in
post processing and sometimes pre-processing is required.
With normal sharpening tool (including Topaz Sharpen AI) the result is
much coarser. I think the future is Photo AI, it just need more time to
On 10/2/2023 17:41, Moose wrote:
I've been messing about with this. My interest is in how this tool
compares to those I already have.
This is my elder son and his elder daughter 10 years ago, Christmas
2002. My first digicam, 1.9 MP Canon S110 P&S.
Photo AI does pretty well with Ella, cleans Eric up too much. Subtly
changes both noses, but not enough to worry about.
Two rounds of Topaz Sharpen AI, with different settings does, to my eye,
slightly better than Photo. With hand clean-up of tongue reflection and
red-eye, it's a huge improvement over the original.
CodeFormer, at the default W-0.5 does some serious damage to the people.
All new eyes, a larger nose, puffier cheeks and, mistaking the saliva on
tongue reflection of a tooth, gives her a whole new lower set of teeth.
Eric's nose, mouth and eyes are all changed a lot. W= 1.0 is less
drastic, but still off.
This round to Topaz Sharpen AI
The flurry of posts, crossing each other left what may be some
Topaz Sharpen AI has 11 different modes, with settings within each. With
the four-up comparison view, it's possible to home in on the best. Using
it on images from one's well know cameras and lenses makes it much
easier. The Auto settings are poor. It requires some active
participation for decent results.
Code Former ONLY works on faces. The other processing you see from this
tool are by another set of code.
Topaz Photo AI is new, and in flux, with frequent updates. Like CF, it
pays special attention to faces, but doesn't alter their basic details.
It also selects an overall 'subject' to work on. In the current and last
versions, leaving this feature on results in a slight offset of the
selected area from faces and background. I imagine that will go away.
More to Come Moose
Themed Olympus Photo Exhibition: http://www.tope.nl/