cornerstone

5-19-update start : this model does as instructed ::

-hmmm moment.. i have been randomly locking down any seed tuning the setting in and once i like the overall feel i unlock the seed give it a test gen if another great design pops up ... and so on...and then the rest of the gens are nearly perfect design one after the other .. i have sat in awe...

``````````````` if you see a black and white image you have left the building....hahahaha i'm sorry but it the first thing i think when it i see it appear... blasted outta the scene... it doesnt hurt to alway start super low... some lora are a strong trained team and dont need as much hand holding... either way it will be in the background doing the math... the best it can.. literally... still working out all the tricks we can do with it... i made a rubiks cube time to play...

-- in simple--its a bridge between loras, allowing each to do it best as its been defined and trained to do: w lora/sample without lora/sample

what interesting in that image is i think ai tried to faux a kitten wearing sunglasses (what i prompted) behind the glass the best it could with the information it had... most likely the banana car + hover car = just enough to construct a crude glass window : ) -- light reflections.. etc.. not bad considering it never seen that in the training

if your not using this then the layers of training in models are not getting the below additional instructions (tags/words/leftbrain) combined with a system (a mathematical image/rightbrain) that can deliver,receive, build..etc between lora and the base in the most effective manner: if one word is blocked by a model or lost it will pick the words it needs to finish : ) its task and pass the info along the system we are.... we are veins : ) we are part of every thing on the stage in the background inside the structures heart : ) no contamination just passing thru

first few words it will do::

0 : "developed"

1 : "directed"

2 : "clarified"

3 : "resolved"

4 : "restored"

5 : "adapted"

6 : "established"

7 : "initiated"

8 : "instituted"

9 : "originated"

10 : "planned"

11 : "arranged"

12 : "collaborated"

13 : "communicated"

14 : "explained"

15 : "incorporated"

16 : "marketed"

17 : "persuaded"

18 : "reconciled"

19 : "assessed"

20 : "encouraged"

21 : "guided"

22 : "motivated"

23 : "coordinated"

24 : "administered"

25 : "managed"

etc....

the point was the test was to take control with a mathematical design and add weigh to tags (words) associated with what i want to happen ::

so scene lacks something move up strength and reinforce modifiers without adding words to prompt...

and based on math, science, and art ( the organic trinity) = fun alchemy

this is were i am at ...

how to mix colors shapes words = balance ( feel free to tilt that balance)

effectively building modifiers (a different cornerstone block) as needed

-- the catch is we are building the scene once weighed and instructed with all available tools which is what took me a few days to figure out what my other lora was doing what...

-- meaning my other lora is my artwork is the supply depot for anything that need to be adjusted to scene

-- reason ai doesnt do what you ask is it didn't have all ingredients, we are fixing that ( i am creating compositions relying heavily on my creative lora) -- have to swap myself out and test building scenes with other artist loras any lora in theory could contain enough building material - imagine you trained it to learn ducks but their was always a power-plant in the background - ai learned way more then just ducks.. trust me : ) what you tell it to do it does.. it also does way more stuff.. the ai is underestimated (i have a trained eye on all disciplines and took months to understand it, granted their is no real manuals to read)

-- this lora likes long complex prompts

-- more testing.. testing.. testing

-- 5-13 -update end

--

if you got any questions just ask... kinda mad scientist trying to explain something i am inventing... hahahahaha

( right now its restoring composition pretty well) cornerstone : a non bias general noise to add depth to the difficult lora and some more little details going on - still working that out

-- helps this (trying something new)) difficult lora : training has actually detained some key artistic principals -- this cant be helped with current general training methods-- looking into this as well -- while i going thru the process for these two

-- compliments this (style ) mindblownart : reintroducing artistic principals - much more then style.. but to simplify -- right now its heavy influenced by my original artwork with is fun but a small cross section-- that needs work as well

compliments all styles- so pumped for this

however this is a CRAZY ENHANCER... IF IT NOT FIGHTING A DIFFICUT LORA..

ITS FUN : ).. STILL TESTING

THIS can compliment all loras: detailers, style, characters concepts.. think of it as a new dial to turn all the rest... but it might need to start at .09 to .85 -- goes off the end after that however again depends on the full weight of everything else

start at .25 or lower lol.. its strong

example with lora

everything same but without lora

i used another lora to add weight to flesh it out a bit.. online gens were making this lora strong... totally crap prompt but been at this all day...lol

this lora will focus on gaining control of everything.. where shooting big here !

1st test : a custom drawn image designed to act like a flash bomb and blind the ai long enough during the noise process to label and organize the elements as its been told ( less assuming from ai) and create a composition resembling what your prompt was, with everyone where it should be.

how does it blind the ai? i am a surreal abstract mathematical chaos machine.

how do you think it labels my artwork? totally lame ... lmao

using the Rorschach test for months (in my mind noting how everything i asked was represented)

a picture began to develop in my mind of the space we are working in and how the ai was using it

• so

the idea for the start as i continue to build a profile of the ai (think he is my client wanting a painting)

(lora training) is to see how what i create, and give to the ai,

who then in return paints a general picture based on what i taught it. (think when you ram the dial to max -- that's the painting i am seeing

• a picture is worth a thousand words

• so i am telling it 1000 words and its telling me 1000 words

but crap most of its words are beep bop beep....

but in tests i swear i heard a HI !

translation its like looking at something in the fog and seeing a hand wave to you for a sec

this is the crazy thinking that's gonna nudge us all into a little more control... hopeful

now after the flash (organized chaos within the noise focused on everything with intention of the perfect harmony of design, it knows it. its in there.

-i learned the dominated shape of information will burn thru all levels for each model

our goal is to keep each model focused by reminding it that it has a wealth of knowledge and jump starting its art mind...

with a work of art specifically designed

for the psyche of the client needs (ours needs to listen)

at the end of the day i am a fine artist throwing paint at a wall hoping for a masterpiece : )

super experimental!!!!!

• i liked the pencil work of these example images created during the training

• i see alot of intelligent design decisions

day one testing:: i made part of it too strong and its steering a curve....but the concept worked as intended.. not 100 % but come on right... we aint hittin that in one go

a one theories why it works... one is my hand sketching style is renaissance style.. the masters of anatomy-- this triggers its memory hub i am thinking while it builds to my request with much more complicated math of new artists like M.C. Escher for layout choices intertwined-- and its kinda weak kinda strong depending how much its needed from other models

-and it works because who we are teaming it with are more photo based at this point

the drawing i made looks like it becomes a 3d mesh over the scene helping where it can at the early stages of noise

--notes of day one