Monday, July 31, 2023

DEZGO : IMAGE to IMAGE : The other side of color

Sometimes I hit a mental block , I doodle but I am unable to form anything. I try different colors , still nothing. Maybe I can see how AI can form substances base on my colors.  

The previous blog https://prataverse.blogspot.com/2023/07/dezgo-elephant-man.html shows that if I let AI freedom to deviate from the image, it is able to generate interesting images. 

Lets try it for Image to Image because I want to use the colors of the image to let the AI have the freedom to create. Using 69-80% Strength ( drastic change) and 1 Guidance ( more freedom ).


 

                                                PROMPT : " "

MODEL : Dreamshaper 7

GUIDANCE : 1 

STRENGTH : 69


STRENGTH : 72 


STRENGTH : 75

STRENGTH : 75


STRENGTH : 75

STRENGTH : 75


STRENGTH : 80


STRENGTH : 80


STRENGTH : 80 
 

The next blog I try to classify my doodling and see what sort of AI setting is suitable for that general class. https://prataverse.blogspot.com/2023/08/dezgo-sculptural-doodle-class.html

Sunday, July 30, 2023

Dezgo : Elephant Man

 I want to test the other side , is the grass greener on the other side? This means  to give the AI the freedom to deviate from my image. It has generated a great looking Roman Warrior from https://prataverse.blogspot.com/2023/07/dezgo-bad-drawings-of-motley-crew-of.html

This is my 3rd testing of Controlled text to image.

 

Doodle of elephant

 

 PROMPT : Elephant shaped like a musical note

 PROMPT : Elephant shaped like a musical note

I use the regular setting of Control scale near 100%  and Guidance about 14 so what I get is basically what I doodle.


 

PROMPT : Elephant shaped like a musical note
CONTROL MODEL : Canny Edges
MODEL : Dreamshaper 7

CONTROL SCALE : 25%

GUIDANCE : 1

 







 
This last one looks like the elephant is dreaming. I get much more interesting images from my initial image to AI that I have bargained for. 
 
It is mind blowing!

 

 

Wednesday, July 26, 2023

Dezgo : Bad drawings of Motley Crew of Wannabe Adventurers

 Refer to my first testing of Controlled Text to Image  https://prataverse.blogspot.com/2023/07/dezgo-controlled-text-to-image.html

My second testing is to see if AI can refine my bad drawings.

Definition

process image intentionally = drawing ( with aesthetic )
process image intentionally =  bad drawing ( without aesthetic )
unintentionally = doodle

Context

I am creating bad drawings for Motley Crew of Wannabe level 1 Adventurers banding up to fight monsters in the dungeon.

 

RANGER

 Bad drawing #1 : elf archer
 

 Controlled Text to Image 

PROMPT : Elf archer 

All of the images generated are women elf archers. There is an association of gender female for elf archers instead of male. I would need to think of another prompt. 


Controlled Text to Image
 
PROMPT : Robin hood archer cartoon rendering 

There are extra lines in the bow that is generated into arrow the bowstring is too close to the neck line and causes generation of a branch and leaf ?

Bad drawing Robin Hood archer cleanup
 

Controlled Text to Image

PROMPT : Robin hood archer cartoon rendering 

 

Inpainting from Text
 
 MASK PROMPT : Bow
PROMPT : Change to thinner bow design
 

The ranger type is eco friendly. The forest is his natural habitat, his world. Skilled in hunting , prefer to kill from a distance with a bow and arrow while camouflaging and observing the prey approaching the time of death. When to die is decided by his hands. He is vain, clothing and weapons are of fashionable and exquisite in taste. Due to his thievery skills and questionable values he is not the first choice for party leadership.    

DWARF FIGHTER

Bad drawing #2 : Dwarf Fighter


Controlled Text to Image

PROMPT : Dwarf with axe and shield

The dwarf race typically works in the mines. They are tough and hardworking lot. They are skilled in mining tools like a hammer which making hammer like swinging motions such as maces and axes their choice of weapon. They are experience in mining and craft work but not skilled in weaponry nor fighting. They have a instinct to jump to a safe cover during cave in, wielding a shield on the non weapon hand can provide the protection function, to go against the instinct of fleeing for safe cover when overwhelming situations arises. 

 

WARRIOR

 

Bad drawing #3 : Warrior

 I think I let the AI more freedom by decreasing the control scale and guidance . As you can see there is no spear and shield but overall I got a better image of a strong warrior. This setting has inspired my next blog : https://prataverse.blogspot.com/2023/07/dezgo-elephant-man.html


Controlled Text to Image

PROMPT : Roman warrior full body shot realistic detail rendering

One look at the confident full of body armor warrior you know he is the leader of the motley crew of adventurers. He directs the party in decision making , who do what and what to do.

MAGE

Bad drawing #4 : Mage

 

Controlled Text to Image

PROMPT : Mage wearing cloak and holding magical staff realistic rendering

This is the most typical mage or wizard. Nothing interesting to say about him. His job function is being duplicated by the Magician in the party. However the team members trust him because he is human.  No ones know what the Magician really is , what is his intent or when he will leave the party to put trust in the Magician. Duplicate roles are important too, if the member gets killed in the dungeon, there is immediate ready replacement without training. 


MAGICIAN 

When I am producing the series of bad drawings, I realized that the team consist of standard , cookie cutter adventurer team and deviating from the term "motley crew". I decided to doodle instead and this image came up.

 Doodle #1 : Strange

This is a weird type , and does not exist in any RPG or fantasy anime. It has a serpent tail and a crescent moon shape face and always smoking a pipe. It can conjure illusions of itself to confuse the enemy. No adventurers in the team has ever touch it and therefore they cannot verify it is real in substance . However they all unanimously agreed of its presence as they all see the same thing. They are more focus on making more money than to debate on Metaphysics. They do not mind it following them being a "member" of the team.

      Controlled Text to Image

PROMPT : Crescent moon faced man with serpent tail smoking a pipe

 

ZEN MONK

Doodle #2 : Zen Monk

This is one of my series of doodles and I don't classify this as a bad drawing. The proportion of the monk poses an issue . The monk has a big head and no body, it seems the legs are connected to the head. The prompt Zen monk would pose a problem for the AI as it matches the Monk figure with the human form. 

 

 Controlled Text to Image


PROMPT : Nendoroid old zen monk


Inpainting from Text


MASK PROMPT : head
PROMPT : bald head monk


 Still it looks like a toy, I don't think the rest of the band would accept the toy monk to join the group.

 

Controlled Text to Image

PROMPT : Old zen monk

 As predicted , the position of the hand with staff looks wrong and the legs of the doodling has turn into the white skirting on the clothes. The face has multiple noses that relates to the doodle.  The image overall gives one a mysterious feel.  

Everyone like to tease the Zen monk 

" Why do you have an extra nose ? " the better to breathe he said. 

" Why do you have an extra nose ? " Zen practice focused on deep breathing he would say in another occasion. 

They asked the same questions again to see if the Zen monk gets agitated and do something unZenly. 

One member however irritate the Zen monk by doing nothing, by merely existing. 

 

UNDEAD KNIGHT

Doodle #3 : Undead Knight
 


Controlled Text to Image

PROMPT : Knight in armor holding two swords

The Undead Knight is causing a headache for the Zen monk. Part of the Zen monk role is to sent the undead back to the dead. This particular undead is different, he used to be a living knight fighting for a just cause , fighting for the king. Due to the fluke in nature , the cause was gone , the king was gone , he the flesh gone too but his spirit and his armor still lives on passionately. He is a talkative one, sharing stories of the good old days with the party.  

He being Undead and alive means the Zen monk cannot judge, cannot classify, cannot decide what to do with this particular Undead. It is a serious problem that could become a obstacle to his path to the Holy Land of Light.

" Why the headache ?" the Undead asked.

" Your existence , you should be dead!" said the Zen monk.

" Yet I am alive as Undead ! the One up there let me live for a Purpose , let the higher up decide , you don't have to be burden by the responsibility. Live and Let Live " said the undead.

" But you are an abomination of Nature ! " replied the monk.

" The universe changes , so does Nature , who are we mere mortals and my undead self tell Nature what to do , to stay still and fixed itself for us so that we can understand it ? ... Loosen up old man. " said the undead as he tried to massage the Zen monk stiff shoulders.

The Undead gets to live another day through the art of sweet talk and persuasion ... :)    

 

See the detail Instructional Doodling process and more bad drawings https://prataverse.blogspot.com/2023/08/dezgo-detail-instructional-doodling.html

Wednesday, July 19, 2023

Dezgo : Controlled Text to Image

If you want to have the image generation follow your sketch image the new Controlled Text to Image option is better than Image to Image option. Refer to https://prataverse.blogspot.com/2023/07/dezgo-ai-cubist-to-realist.html Image to Image post.


The image of the cat woman , Image to Image rendering.

 

PROMPT : Realistic rendering of cat woman face.

Controlled Model : Scribble



The image of the cat woman , Controlled Text to Image rendering.

The result is closer to the my doddle line art while totally ignoring the purple stroke color.

INSTRUCTION : Draw an alien that comes to mind.

PROMPT : Alien with a weapon in a bright environment.

Controlled Model : Canny Edges


 The result looks great. It makes my 1 minute doodle into a professional concept artwork , adding detailing to the alien and the environment.

PROMPT : Piranha head man.

Controlled Model : Scribble

 

 

This is another great result that I cannot do before using Image to Image . See the rest of the Fish head characters in https://prataverse.blogspot.com/2023/07/dezgo-ai-blowfishman-and-friends.html


 PROMPT : Ostrich man with long neck cartoon look

Controlled Model : Normal map


It is not accurate to my doodle but I do like the big eyes and the "M" shape recess overall it looks kind of funny.

I am motivated to continue to draw the rest of the body. I end up drawing a Goose Man instead. 


 

I upload another of my doodle that does not translate well using Image to Image generation. It is a doodle of a man's back and buttock. I do not want to directly prompt the obvious but use the form of sandbag to generate a 3D realistic image. 

PROMPT : Realistic fiber sandbag sagging.

MODEL : Line Art

 


I think the AI translate the thicker strokes to brown straps on the sandbag. I would prefer only the sandbag shape and the lines translate to either outlines or creases on the sandbag.

PROMPT : Realistic fiber sandbag like a body.

MODEL : Normal Map


 The straps are less obvious here and there are more creases but unfortunately the overall shape is totally lost. Maybe I reduced the percentage of the control parameters too much here. 


I trace my doodle using Adobe Animate ( formally known as Adobe Flash ) into a uniform vector line. 

Retracing is a drawing process that is very intentional. It is not my preferred method.

PROMPT : Realistic fiber sandbag sagging.

MODEL : Canny Edges

 


It worked well. I need to take note of the way I doodle next time if I intend to have the final image to be generated in Controlled Text To Image.
 
I continued to test the Controlled Text to Image feature with a theme in mind , that is to create a series of bad drawings for AI to process. I don't think the AI can differentiate my doodle from my bad drawing after all it is my method of processing the image , only I know how I process it .