CthulhuTECH

In the depths of a forgotten library, amidst ancient tomes and whispered secrets, I discovered the Instrumentum insaniae—an infernal device capable of summoning Lovecraftian nightmares. Crafted from melded metals, its intricate gears and pulsating crystalline orb exuded an eerie luminescence, hinting at the forbidden knowledge it held.

With caution, I approached the artifact noticing that each dial and lever seemed poised to unlock the gateway to madness, filling me with a chilling anticipation of the cosmic horrors it could unleash. Guided by forgotten incantations, I dared to traverse the boundaries of mortal perception.

As the Instrumentum insaniae sprang to life, its eldritch radiance cast macabre shadows, transforming ordinary objects into abominable forms. Statues twisted into tentacled monstrosities, while the architecture itself shifted in alignment with the chaotic madness of the cosmos. Through its unholy power, I became a conduit, beholding the terrifying beauty of Lovecraft's cosmic pantheon.

Yet, a whispered caution echoed within my mind, warning of the perils that accompany such forbidden knowledge. With a heavy heart, I chose to part ways with the Instrumentum insaniae, leaving it behind in the hopes that others would take up its mantle, for the delicate balance that separates our world from the abyss must be preserved, and the artifact's dormant secrets may hold the key to understanding the mysteries that lie beyond.

CthulhuTECH prompt notes and fun facts

• You do NOT need to use the mythostech keyword, the lora is just that powerful, we're talking elder gods up in your GPU, some concepts will works a lot better without using it

• Check out my samples and prompts (there’s like 500 something sample images, or will be once they’re uploaded minus anything the filter rejects

• Yes, you can make porn with it, yes, some of it is confusingly hot

• If you are upscaling, keep your denoise low, you don't want to correct away some coolthulhu, With REV 1.22 and coll remacri upscaler i use 0.32 as my baseline

• Generally works fine at a strength of 1, you may want to adjust with some models

• Its ok to put 'human' in the negative, helps a lot with hybrids and deepones and if you want to do some crazy mutants

Wildcards

https://civitai.com/models/63834?modelVersionId=68415

PureChaos edition

• Purechaos edition is a lot more random, if you want to make a lovecraftian nightmare and don't care a lot about control, go forth, forsake the elder sign, invoke nyarlathotep… and adjust the strength of the lora or keyword at your peril. I’ve only included a few purechaos examples because you really don’t need them.

Glowingrunes

• glowingrunes is OP, i suggest putting it in a negative 1.0 then calling positive glowingrunes when you want them, depending on what you’re prompting a glowingrunes of 0.2 will still show over a negative 1.0 glowingrunes, especially if you’re also using cyberpunk.

So much work

• Trained on just over 3300 AI generated lovecraftian images, Generated nearly 2600 image examples curated down to just over 500.

• Coffee Maker included because this started as a konyconi style LORA, but it isn't, not anymore :D. Still they make excellent stuff and you should check em out.

• All sample images are purely this lora on top of REV 1.22, no hidden loras, TIs, tricksy editing, everything is raw prompt output.

Known Issues

• There are still a couple of weak areas that will me more cartoony, industrial equipment for example, books, and for some reason mobile phones.

• Crouching people often have 3 feet or random shoes.... I don't know why, i suspect its maybe bleed from the crouching people in robes confusing the data

• Bioluminscent doesn’t work well on anything, more of a general glow…. A lot of the training images were beautiful so i think maybe I need to train this as a general concept and not tie it to specific keywords

Leave a review

If you like it, please review it

For the love of Ithaqua;

If you like it, please review it or;

squeeze me some carrot juice - https://ko-fi.com/bunnyviking

Prompt words

The best reference is this spreadsheet which has all the major prompts and illustrates how they fit into the hierarchy. You don’t need to use the whole hierarchy at all, but it may help to understand how things fit together (especially when working with humans, deep ones or artificial lifeforms)

The spreadsheet is currently a little incomplete because of an error I made but hopefully it’ll be completed in the next 24 hours… For now you can always get the complete list of every keyword ever by looking at the metadata with a text editor, or clicking the little ‘i’ symbol on the lora in additional networks (or via many extensions and tools)

Quickstart - if you just wanna jump in:

Human is the basic people keyword, I use man and woman for gender (you can kinda prompt futanari and easily prompt androgyny … whats the equivalent of futa for girl parts on dudes, thats like ezmode)

Mutant,Cyborg,Cultist,Demonic,Tattoo,Mask,Sorceror,Soldier,Manyeye,Dieselpunk,Fashion,Encountersuit,Cyberpunk,Scifi,Flamethrower

Monsters use Monster as their basic keyword and unless you want to be truly random I suggest actually putting in a type of monster (ghoul,chimera,manticore,unicorn etc)

Furry,Winged,Bioluminescent,Scaly,Octopoid,Humanoid,Dragon,cthulhu

Building creates buildings, I suggest being more specific e.g. a house, an apartment block, etc. You can get interiors but theyre a bit inconsistent except what ive specifically trained (e.g. cafe)

Industrial,Dieselpunk,Cyberpunk,Tendrils,Skyscraper,scaly

Technology is a catch-all for tech, its a bit mixed and needs work but if you prompt a particular thing (engine, server, air compressor) it’ll generally work ok.

Containment:Sphere,Hexahedral,pyramid - these make different shapes of containment for power plants, mystic thingoes or looking like cool AI hubs (adding eyes is cool)

Coffeemaker,pot,Mobilephone,Screen,Server,computer

Landscape c’mon ;)

Obelisks,Temple,Ruins,Pyramid,Statues,maesoleum make structures, you may want to negative them as necessary

Cyberpunk - particularly for urban or structures, dieselpunk will work a bit but it’s not trained

Jungle,Desert,bushland,Urban,Volcanic,hotspring,lava,Arctic,Underwater,abyss,forest,Park,mountain,Swamp,Hills,beach,fleshland

Hovercraft you can try some modifiers like large, small, transport but it’s inconsistent.

Cyberpunk, dieselpunk

Animal - same as monster, best to prompt a specific animal but you don’t have to.

Aquatic,Bioluminescent,Insect,serpent,

Balloon - makes zeppelins, can’t prompt much else

Cyberpunk,dieselpunk

Artifact - makes statues and small objects and things to put them in, definitely use some of the other prompts to guide it.

Statue,Obelisk,Relic,Curiosity,glowingrunes,Container,Statuette,token

Artificial - robots n such

Woman,Man,Cyborg,Android,Cyborg,Robot,Cyberpunk

Mech - Makes mechs or all types depending on how you prompt it, quads and crawlers are significantly harder to prompt

bipedal,Quad,Crawler,Biomechanical,Behemoth,Dieselpunk,cyberpunk

Deepone - lovecrafts fish-people. Seriously, check my prompts, these are hard.

Hybrid,Woman,cyberfish,Abomination,Man,pureblood,mutant,Lanternfish,goldfish,Withspawn,Angler,Bioluminescent,shark,cyborg,Encountersuit

Industrial - machinery n junk

Machine,Mining,Generator,Vehicle

Portal - y’know, portals

Doorway,rift,Primitive,Square,Circle,Hex,watcher,Tri,Scifi

Vehicle - from sports cars, to sports cars with guns, to tanks

Battlecar,Warmachine,battletank,Armed,Spikes,Spines,legs,cyberpunk,Closecombat,dieselpunk

Spaceship - sigh you understand. Can be a bit inconsistent across types of ship, prone to white background

Leviathan,Naval,Ufo,angular,Cyberpunk,Dieselpunk

Aircraft - … suggest fighter, modern fighter, bomber, etc

Antique,Cyberpunk,dieselpunk,drone

Book - On its own makes pretty cool books, doesn’t play nice with making different kinds of books.

Future plans

• Assuming people like it enough, review it, etc. I will release future versions will will include some or all of the following (if people dont like it, i'll just keep it to myself :D ) I have currently put approximately a full week of work days into this project, that's a lot of time. Making some of these changes will take considerably MORE time.
  

• Separate the major keywords into full concepts to remove some of the mixing (e.g. deepone would become a keyword not just a prompt) This is a major project in its own as i'd need to move a lot of things around (e.g. compiling all the dieselpunk and cyberpunk stuff into a keyword where it's currently in a sub sub folders

• Make separate Loras for those keywords

• Even if i don't do full concepts I will be totally reorganising the data as the way I have done it is super flawed and makes you need to prompt/negative hard if you want a very specific result

  
  

• Add more training data to some weak areas (such as industrial equipment and adding more types of books to books, aircraft, major deepone overhaul,etc,etc)

• Make a 'lite' version that is more about style than being comprehensive

• Make a 'cute' version

• Train a model (people would need to Looooooove this for me to train a model as it takes forever, at least I can run dreambooth on runpod even though that costs a few bucks)

  
  

I have a YouTube channel where I neglect to review Loras, checkpoints etc.

https://www.youtube.com/channel/UCjXlRo9kik1YJYONdjz4kiw

But i do plan to start doing it again now this ridiculous project is released.

p.s. If you 'cant get my results with my prompts that's a you problem and I can't help you.

p.p.s Cthulhu loves you anyway, dream sweetly of it.