Carrying on my from last post, it is time to complete "The High Priestess".
As a reminder to myself, what I am going to fix today:
1. The script on the left hand side
2. Increase the detail and refine the facial expression to be a bit more fierce and piercing
3. Fix the hand representation.
The first step is easy, and was executed with a simple, single word in the negative prompt:
watermark
The next step is to fix the hands, and that's likely to be more time consuming. I've used the prompt
hands cradling a floating orb
And decided to run a contact sheet of fifty images.
I'm pleased to say that none of these options worked with a good level of coherence. Therefore, I might try a different strategy, where I find an image of what I want, use clip to get an appropriate prompt, then run that through in-painting.
It is, after all, all about experimentation!
It gave me a sage piece of advice holding a crystal ball
I tried that prompt, and after about 150 different iterations, I got this:
The other results have thus far been horrendous, here is another sample of images using the same prompt, with a negative prompt to encourage correctly drawn hands:
I fear that I will be generating images for quite some time in order to get the hands to become perfect. It is so far from being close it isn't funny. It is more frustrating than anything else, but this is the process I've chosen to use. There will of course, be better ways, and improvements as the technology matures, but for now... I will suffer and keep generating new iterations.
So, what if I change my vision?
a glowing orb of light obscuring hands
The results are still not quite there...
Getting closer, but there's still too many fingers. Instead, what about if there's something with NO hands?
I suppose you could argue that there's hands underneath the veil, but it looks obvious that it is... a thinly veiled attempt to hide the fact that I am struggling to generate good and properly defined hands using a machine learning tool. Perhaps if I mask it with some mystical, floating energy or something...
That prompt certainly didn't work at all. That's a shame. Maybe I'll just have to go back over to Photoshop. Time to do the face enhancement and then create smoke that will blend in Photoshop. :)
This is the outcome of doing some Photoshop work, adding a path, rendering some flame, and essentially tweaking settings until it turned to a subtle sort of smoke. I think the blending is okay, and I ensured that I wasn't disrupting colour harmony.
There is also the need to fix the facial structure up, and finally, upscale the image and go back to Photoshop to tweak the direction of the gaze.
Here's a fun experiment, running through GFPGAN multiple times to enhance the face slowly, and incrementally.
Here is the progression. The first three images are purely GFPGAN, and then the tweaks below it are me, manually in Photoshop, using liquefy to polish the end result up.
This brings me to the final result of The High Priestess:
Next up on the agenda is "The Empress". This one felt good, once I got cracking into it. I'm sure The Empress will be even better as I familiarise myself with the process and enhance my workflow.
Want more content from me?
Witness my futile efforts to play my Steam Game collection in alphabetical order.
Are you aware that I love photography? Check out my work in a collection.
If you want to see my Splinterlands antics and rants live, Find me on Twitch
If you prefer sleeping in your designated time zone, go watch replays on YouTube.