FACE RESTORATION V1.3
After countless iterations you finally craft the perfect prompt for your favourite Text 2 Image AI… only to get two squint uneven eyes and a nose with three nostrils staring back at you. Now you could face a few hours in Photoshop (or GIMP) to get your AI portrait looking the way you want. Time you could spend doing something else, crafting more elaborate prompts probably. Welcome to the emerging Magik that is PromptCraft, step forward Neophyte. 😛
If only there were a quick, easy and convenient way to take care of this tedious task. It is my pleasure to present to you Applied Research Centres Face Restoration. Upload your image… Check the results… Download your new image. It really is as simple as that.
LETS SEE WHAT YA GOT!
Even although the latest releases of MidJourney, Stable Diffusion, et all have vastly increased the quality and realism of portraits and facial features they sometimes still produce oddities. So lets see what ARC can do to fix this automatically?
First we open our browser, Brave in my case (affiliate link to follow lol), and visit ARC Face Restoration.
As you will see we have the option of v1.2 or v1.3. I have had a play around with both versions and it defaults as v1.3 for a reason. You get much more consistent results, so save yourself some time and just leave it at v1.3.
Now for our first upload and results. As it so happens one of my last projects was female portraits. Here are a few images I wanted to fix for that project, but I didn’t plan on getting around to this task any time soon.
I’ve got to say, I am really impressed with the results and ease of use. Let us have a look at a couple more to make sure this first one wasn’t a fluke.
This image here shows the classic MidJourney eye. Almost there, but glassy and sometimes duplicated. For me, this would probably be one of my most frequent types of fix in MidJourney. Adding ARC to your work pipeline could save so much time. I know I will be using this resource while it is freely available.
MidJourney was so close here. While the eye isn’t doubled or wonky, it has that glassy look to it again. The ARC fix here tackles that glassy look replacing it with way more realistic eyes. Also nice and restrained in the mouth region, cleaning the teeth, and really subtle lip line restructure. Got to say I am a fan.
This is the image that I was wanting to work on the most. I can see ARC Face Restoration as a definite part of my work pipeline from now on. Stable Diffusion/MidJourney –> ARC face Restoration (if needed) –> GIMP (in the process of going full Linux so no more Photoshop, I have to say as an alternative though, GIMP has me mostly covered). However I will soon be having a look at some other face restoration software to see how it compares.
What a quick way to work. The before ARC image on one layer, with the after ARC image on the other. Then just quick masking between the layers to keep the better highlights and details etc. Oh man, I need to get a stylus again.
HUMAN MATTING v1.2
While we are here it would be folly to not check out the Human Matting tab. Masking, matting, cutting out, grabbing, whatever you want to call it, achieving this manually is a job only masochists enjoy.
The images above have good plain backgrounds so lets upload a couple of our “After” ARC images to see what results we get back. Oh, and this is Human Matting v1.2 using the default transparent background, and I have added black and white backgrounds to better see the transition under different circumstances.
HUMAN MATTING v1.1
As you can see there is a slight halo around this image, with certain backgrounds this could really stand out. Just for comparison we shall have a look at the same images using v1.1 to see if there is any improvement.
As you can see v1.1 has a sharper transition with v1.2 showing more of a feathered transition between portrait and background. I can see both of these being useful in different circumstances with experimentation being the best way to figure out the strengths and weaknesses of each. Again, this could be a huge time saver for you.
I am excited to see what other ways ARC use this new emerging technology, as before this I have to admit that I had never heard of them before.
So if Applied Resource Centre is our baseline for post Text 2 Image auto improvement, then the competition better put their best foot forward as the standard has just been set pretty high.