Reactor Praise and problem experience

[drdancm]

I'm very grateful for getting Reactor working with the pre-built package which removed the need to load Visual Studio etc. At first I couldn't get it to work, but after some time I got it working inside Stability Matrix (the last roadblock was perhaps removed by a Stability Matrix update).

The interface is, in my opinion, better than Face Swap Lab, which I had also tried after getting the pre-requisites working.
I did have some problems with it, and still some problems with Reactor, which I will discuss in case it is helpful to someone else.

1 Reactor would work sometimes, and fail other times to make any swaps, and at first I could not figure out what I was doing wrong.
After some time I learned that the Face Swap will not work with all models. It appears to work with DreamShaper 7 which is I think what you show in the demo pictures. With some other models it appears to fail. Anyone else have that experience ?

2 Other times when it failed to work, it was not until I looked at the execution page, that I realized, because it is clearly shown, that it failed because of a gender mismatch of some sort between source and target. So it is important to check out the lines of execution for reasons why things fail. This is obvious to experienced users, but not so for others.

3 On occasion Reactor misidentifies a female for a male face, even when the target appears clearly the appropriate gender and you would not expect the program to make such a mistake. We may find this very puzzling but we must keep in mind that the program probably only looks at faces (is this correct?), while we not only write the prompts, so we almost certainly know the characters, and but we probably forget that when we look at a picture we judge gender based not simply based on the face, but the entire figure.

4 I assume all users have noticed that replacing target faces deviating from looking straight ahead at the viewer/camera become gradually less satisfactory, with a complete Left or Right Profile also looking less good. Would the profile work better if the Source image was a profile? or would it not make a difference.
I'm guessing that this is a problem inherent in the tools used by Reactor, rather than an issue with Reactor. A lot of face recognition applications (Portrait Pro as well as lots of photographic management programs) have problems with face recognition of profiles.

Overall, the program is EXCELLENT and I am delighted in having such a great tool. I would love to have it be available in Fooocus also, but that is up to the developers of that package.

I just noticed that there is a new version called Reactor-Force. Should be remove Reactor, before adding Reactor Force, or can both be loaded at the same time? I'm guessing I should probably unload the Reactor extension and then load the new version since my nVidia card has 8 Gb.

Thanks again.

[drdancm]

I'm having problems with close up portraits even though when the original face is smaller in the image things work perfectly well. By close up I mean instances where the face and neck or upper chess fills the entire image.

It almost always, or probably always fails to produce a good face. Eyes are often blurry and in many examples it looks as if the swapped in face is smaller than the original face. You can see smaller pair of eyes superimposed on the original and the mouth shows up right under the nose. Since I'm using Reactor-Force (which is still beta?), I don't know if this is also a problem in the Reactor, since I haven't had a chance to switch back.