Follow me on X/Twitter for exclusive content: https://x.com/Citron_Legacy 😁
Please checkout my other SciFi Loras: https://civarchive.com/collections/11067257
Description
TechJelly,
,<lora:TechJelly_IXL:0.7>,
FAQ
Comments (6)
This is a really cool concept so far!
but someone took my Galena Cat checkpoint away my preciousss and the other checkpoints aren't responding to my prompts the way I like them, so doing this Lora justice is just gonna have to wait a bit
Lol, I'm laughing at myself now. Apparently I couldn't resist.
Judging by the last couple days, I'll be spending around 3 days of my civitai time focused on this model.
(Yep, I couldn't stay strong and wait until Galena comes back...
self-control, thy name is not civitai's seductive appeal!
also, thy name is not ravem's willpower! also, not citron's magnificently absorbing Loras!)
Anyway, I can report this concept Lora is great. Plumbing its depths is a multi-day process, and every bit is worth it.
It's 1: a really cool concept and look, 2: gorgeous, 3: highly versatile, creatively meshing and synergizing with all sorts of concepts and other Loras, and 4: great at making flying warships in space (which along with dragons is basically the terminal value of AI art anyway, right?)
I'll start uploading my favorite results soon, but probably take a day or three, civit's uploading is hiccupping a lot lately.
BTW @CitronLegacy , I usually start my fun with a versatile lora like this by genning cool hybrids, but you had me beat here -- the showcase pics are great, and really display the lora's versatility,
like the robot version, and I especially lol'd at the TechJelly Imp Midna pic.
(only a true Zelda fan can accidentally(?) create such a cool gen!)
I don't know if it's the Lora training method, the bio-tech concept fusion, or just innate Illustrious, (or just innate CitronLegacy...), but this Lora is just really versatile and creative.
So, thanks for the work creating and sharing it here!
Sorry for the late reply! Work has been a beat down this week D:
So you cheated on Galena CAT... I see how it is! LOL just kidding. I think the model you used did a perfect job at lighting details especially with translucent parts.
Self-control is for the weak! ...oh actually I guess its the other way around lol
Thanks for the high praise but a lot of my creative fusions are mostly luck. Glad you liked the Imp! It was a random output but really seemed like a interesting concept that I've not see in any scifi content before.
I think the versatility comes from the fact that "the AI" didn't know what it was looking at when I trained it so its forced to be creative.
(This is a very rough theory so it probably needs some more brainstorming and evidence)
Theory: This lora is telling the checkpoint "I want these shapes, these colors, make it work!" and the checkpoint is trying to rationalize how a scene would make sense. Sometimes the tentacles look like tentacles but sometimes the AI thinks they are something else. This comes from the AI not understanding what "its looking at" so it has to be creative like SD1.5 was. This strategy would likely not work on Flux.
So anyway there is nothing special about the training settings but the training images are the key. For this I made 100 images of "semi-robotic jellyfish" on Flux and then selected a small number of them (maybe 30%) that really got my idea correct to use as the dataset.
As for the thanks, you are welcome! As I said on the SharkWing, I made this set of Scifi Loras specifically because I thought they would inspire you to make a Lora! (Well I also need to make scifi things for my personal growth but its a win-win!)
I think you could use some of the dragon or ship images you generated with this Lora to make a Lora.
Everything in this scifi series is "this + that" so "techjelly + warship" is a viable Lora idea! Just come up with a cool concept that explains the look. For example "Aerogel IronSide" but a better name lol
.
As a side note, using GPT-image-1 to generate images for a dataset would probably work well. Its more creative and flexible than Flux. This was trained entirely on images I made with GPT-image-1 https://civitai.com/models/1581572
@ravemry9 Re: dragons is basically the terminal value of AI art anyway, right?
Yes. Yes dragons really are the reason for AI lol
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.