LoRA model for Jeep Wrangler. Works best on revAnimated or Protogen
768x768, 40 images, 2000 steps
Use strength between 0.5 - 0.8
Description
FAQ
Comments (9)
Amazing, thank you for this!
I really super like 2018JEEP , hope it has Versions from different eras， (AND 2 DOOR ....) Please forgive my greed，Thank you VERY MUCH~！！！
Thanks - I'll try over the weekend. Gathering data and captioning takes all the time. I am extremely busy with work and family during the weekdays - this is just a hobby
WOW! This looks amazing! Thank you.
How does a person train a LORA with images for making cars? Do you cut-out the backgrounds and just use the car image? ... Or do you have to write some special words in the "Blip" captions?
I have 70 beautiful HQ-HD images of the Ford Falcon "MAD MAX" car ready to make a Lora, but don´t know where to start when making a "Car Lora". Do you have any tips or tutorial?
I can tell you, but I'll have to kill you :)
Kidding, i's all in the captioning. Make sure you describe every single thing in each picture. You don't need 75 pics, just 40ish HQ is fine. Upscale/resize to 768x768 and then give descriptions. e.g. if your LoRA object name is frdflcn, type "frdflcn, bright orange body, black grill, front right view, low angle, parked on a dirt road, snow on the ground, next to a winding river, stones in river, trees in background, mountains in background, natural lighting, sunset scene, pink and orange skies, white clouds"... and so on. Must start with the LoRA trigger word. Use Kohya UI as usual. I use batch size=6, epoch=10, bf16, LR = 0.0001, text encoder = 5e-5, constant, network dim/alpha = 128/128
You are a great friend, you have been very helpful. I have created a Lora a few days back, but nothing was showing properly on any images. You have opened my eyes to a diferent way to Label the descriptions on the "Blip" files. This will be very useful. I hope you like the "Mad Max" Pursuit Special Muscle car. I will be creating 2 Loras for them. One Lora with a brand new car and another Lora for the Mad-Mad-2 film version car with all body-parts ready for all Road-Warrior to enjoy.
BTW - Your Jeep model is most EXCELLENT ! Love all the details in it. My late father used to drive one and he would have loved to see this.
Cheers
@maxhitman You are welcome! Rest in peace, maxhitman Sr
I want to make some car LORAs, but I can only find widescreen images. When training, will the images be cropped/squeezed?
I use this colab for making LORAs,:
https://civitai.com/models/22530?modelVersionId=26903
Because my laptop has a 1060 Max-Q , it can generate fine (takes 22 seconds for a 512x512 pic to generate), and I doubt it can train quickly enough while having everything closed.
Can this do images with the roof off?