The fish at the bottom of the sea
Ringo Star sang ' I'd like to be, under the sea ..'
ScuzzBlog: Diaries May 2024
Entry 25th May 2024: Post 1: Tensor Art - Final Thoughts Day 4.
Tensor Art - Final Thoughts Day 4.
My last galleries today for a while, I have more but I think I
have shown enough for now. And with it my final thoughts.
First up I think the whole AI process for creating art is a great
benefit to all those that want to dabble but in the past have been
restricted due to a lack of perceived skill. AI art creators such
as Tensor make it possible for anyone to create wonderful pictures.
So do I think there is any magical formula? In truth no. For the
most part pictures all have a massive RNG factor. One image can
be a true work of art and the next , using the same prompts, can
be a load of junk. I have tried myself to refine prompts to give
a more consistent result but have failed. I also sense that like
human biorhythm there is a logarithmic curve in force that gently
works its way through the processes guaranteeing at times a cycle
of truly dreadful pictures. When the crap cycle starts I generally
log out completely and clear my browser cache. Sometimes works.
So what are my tips.
If you want a one off pic and not a series of similar themed pics
then simply write a lyrical , poetic, descriptive load of twaddle
that includes all the subject matter you want to see.
If you want a theme that repeats you have to structure prompts
where clear descriptions are very simple in words and placed in
double or even triple brackets .. like this (((black coat))),
cus saying black coat doesn't necessarily mean your character
will wear a black coat. The brackets help.
What I do is structure the prompts into an opening base set of
descriptive words (prompts) that describe the main focus. then
I describe any activities like walking, standing, moving, flying
and which direction and viewpoint. Finally I add embellishments
which are about quality and these I have assembled from images
I have seen on other works, like masterpiece, best quality etc.
Sadly the generator is not very clever, and here is why ..
If you clearly state that the boat is gleaming, shiny red and
with a white strip .. you would think your boat is going to be
that colour. Unfortunately the software picks up key words and
if say later in the text you make some reference to say a blue
sky, don't be surprised that your boat is blue and also a car.
Any text or words you use in prompts can be interpreted to mean
the main context. So if you only want a red item then don't
specify any other colour anywhere else. Or after saying a red
boat just enter red boat again but this time in brackets (((red
boat))), which basically says ' hey moron the boat is bloody red '.
Also remember that a word may mean something to you but another
word may also mean the same thing... like rear view may cause the
AI to introduce an image of a sexy chick with a cute ass. The
generator does like inserting girls at a whim so any female
references avoid. I have had several images ruined cus I described
something in a girlie way.
Anyhoo you get the idea.
Should we be afraid of AI. So funny, hell no. In terms of art
generators they are little more than stepped morphing software.
The prompt interpreter pulls out keywords and depending on the
setting you attribute in the sliders and lower sections of the
interface will define it's sorting and morphing process on a
very crude set of perimeters. What it does then is go rush off
and half inch images from picture libraries and then cleverly
morph the whole lot into a stepped other image until it fits the
descriptive image of the final solution. The magic is in the
believing you can create a truly original masterpiece. Trust me
you do not. You simply morph a load of existing images.
I have tried to create original artwork but you cannot treat it
as a paint box. You can't open a canvass and then say I want a
rock face here, and a big gate there, and a column over here and
a big fat ogre here coming out of the sea, and mountains at the
back. The content of your image will be a morphed set of images
stored for that 'checkpoint' or 'Lora'. If you use a specific
Lora enough you soon get to notice the same stock walls, floors,
lighting methods, trees, and even characters. It really isn't
that clever.
There are many art generators on the market and a lot are free.
Another good one is Civitai and there also is SeaArtAI.
That is that for now. AI in truth is always a product of human
endeavour. I don't believe it ever really thinks for itself. It
just gives the impression it does.
Here are my last galleries. Enjoy.
PS My handle is Veiga Prime.
Tensor Art - Final Thoughts Day 4.
Robot Begins
Seacrawler Botfleet
Princess and the demon
The best picture
with a human aquarium ....
If you can only see this CONTENT window
then click the image above for the full site
Last updated 25th May 2024
Chandraise Kingdom
Keep the Faith
scuzzscink 2024