Well, it does something, that's for sure. A lot of known mecha junk floating in the internet combined with my hardcore Photoshop skills and stuff i've made and rendered in 3D ( the non AI generated and not circulating in the internetz kind of shit )
Trained on NoobAI
No fucking clue what is considered Illustrious or NoobAI
Combination of natural language and tags
Good luck
Extremely depending of the checkpoint used, 'cause every checkpoint that isn't overtrained to produce only the most high quality images will most likely make you cry.
Just as i side note. Every model, be it Illustrious, NoobAi or Pony is based on SDXL 1.0. Just because something is trained on a dedicated model doesn't mean it has no influence to a certain degree.
Flux.1 D
Ehrm, i'll give up. Way to much time spend on it and Flux just takes to long in every regard, be it generating or training. The good thing though, was able to change a few settings in Kohya and i hope the weird horizontal "streaks" that my other Flux LoRas produced sometimes are gone.
Maybe someone that is using Flux more often and isn't so out of the loop as i'm has more luck.
v3.0
Added few more pictures with a different design i couldn't finish last time. Should be done for now, no more point in doing more.
Description
More Pictures, more training
FAQ
Comments (9)
This is the best Lora to mechanizing characters for now. This beats CyborgMechExosuit lora (for me). But there's one thing common in them: they generating unwanted backpacks or accessories at the back. Any idea how can I off them?
Hmm, hard to say. It's possible that, if you ask for a backside view of something that it will add a backpack by itself, 'cause "back" and "backpack" are pretty close to each other. There is only one picture that could cause this behaviour, but that could be enough to make it gravitate towards pictures with a subject and a backpack.
Also, military equipment. I try to add as much variety as possible, but sometimes depending on the model it adds stuff by itself. The most prominent problem i have myself with this model that it adds a few things that aren't wanted.
There are a lot of stuff that could be interpreted as something else. Wires for example could also be "roots" or "tentrils" which adds a lot of clutter to the image. It's always hard to determine how every individual person prompts for something or what you personally wanna see.
For the most part, models based on SDXL are kinda random on what they do in the end, which is a good thing for me, but maybe, or most likely, not for someone else.
If you see something that is unwanted, try to figure out which words are so similar to each other that they would cause problems and putt it in the negative prompt. A lot of times a single word will add or remove a lot of yankyness.
tl:dr - use my models and you're in a world of hurt :D
very cool concepts that works
Hmm, all these IL and NAI images look like treated with sandpaper or as if a filter was applied, meh.
Hi Bro thank you. I really like this lora. but when i generate images using flux1d lora's it will create a scan line effect from middle to bottom of the picture (sample : https://imgur.com/zwaSFgM ). did i do wrong? i change strength from 0.5 to 1.0 its all same result.
Still not sure what is causing this problem. Thought i got rid of of by changing a few settings in Kohya. Someone mentioned that if u upscale the picture it fixes this problem, but, can't confirm since the change in settings i didn't have this problem anymore. Seems to be dependent on the resolution, 'cause it happened most of the the with high-res pictures.
great lora. actually prefer v2 for nai over v3. with i2i i can draw pretty much any geometrical shape i want and this thing like grows industrial parts over it. its really sick thank you.
I'm using ComfyUI east install, and I'm just getting an error.
I'm using the model "flux1-dev-Q6_L.gguf" and the error is "Lora Key not loaded."
Could we bother you for this edgy mech data set trained for a Z Image Base lora? It's shaping up to be the most capable model outside of Chroma. Thanks.
Details
Files
Available On (3 platforms)
Same model published on other platforms. May have additional downloads or version variants.