Type | |
Stats | 3,961 64,985 |
Reviews | (483) |
Published | Apr 21, 2024 |
Base Model | |
Trigger Words | reelmech, mechanical parts, cable, wires, machinery, joints, body suit |
Hash | AutoV2 4B32CDCDB1 |
About PONY XL - Real Mechanical Parts version!!!!
It is important that you should know, it is on alpha stage and has lots of room to improve.
I've released it because it is giving good results on this stage.
You should put "animal" tag / word / token on the negative prompts to NOT GET animal results even you didn't write /call for it.
On positive prompt, you must add "cable, wires, reelmech and mechanical parts" to get what the lora offers. Otherwise, you will not get what you expect on this stage of LORA.
,I've spent almost 6 days for making proper Dataset, editing tags and 3 days for local training and testing. If you can't or don't want to support but using this LORA, please at least send your outputs, follow my profile and comment what you like or what you didn't.
To make this LORA into SDXL, Cascade i need at least RTX 4090 OR A6000 GPU because of VRAM limitation. So i need your support to make it. In a week or 2, i will set funding or similar to look for your support to make it better LORA, Checkpoints.
( why don't i use cloud training ? - It is because i don't trust cloud-training for my "Training DATA-SET images" and most of cloud training servers are NOT allowing to train NSFW or some different concept of content or restrictive about it.)
After learning many tricks-things about training and many training with failures-success, i made v1.3 of this LORA,
v1.3;
Added more variations into image dataset. v1.2 has 148, v1.3 has around 2400 image.
It is more responsive to text, more sharper in details.
trained on 768x768
-- still working on "ONLY" 1 arm, 1 leg OR partial of body. But it is really hard and not obvious except handmade-photoshop dataset which i don't want to get involved because of frequency of updates about this LORA or other LORA that i'm working on.
v1.2;
+Added more variations into mechanical parts.
+Hopefuly fixed issues where collides some of other loras.
+learning rate changed so it will snap into object more natural than the older version.
use reelmech tag and if you want almost full mechanical parts use it with 1.5 ratio such as this (reelmech:1.5)
v.1.0;
This is my first Lora model that is aimed for mechanical parts for organic bodies. Mostly humanoids.
I've tried it on more than 5 popular checkpoints and it worked fine but with 1 flaw. If you don't upscale at least x2 ( two times ) you will most likely get blurry faces. If you do upscale, you won't get any problems with faces.
Use lora tag <lora:reelmech:1>
also add a method, (mechanical parts:1.x) where x is equal to a number to increase number of mechanical parts on your model. I recommend it to set that number between 1.1 to 1.5.
If you leave a comment, i can make it better models of me and your liking!
I appreciate it if you support and buy me coffee.