Good car
try to add suspension to the prompt to maybe get the car with some lowrider vibes
side view, angled view (frontal angled), rear view
white top
epoch 2 and 3 make more realistic cars but i feel that they use too much closeup compared to e1
Description
FAQ
Comments (7)
Nice work, Any way you could make some more Chevy Loras? Would love a Caprice Station wagon for scenes!
yes, looks amazing, right now i'm kinda occupied irl and have little time playing around SD but will try to make whenever i have time (not in the coming month though)
I've been trying to train some cars of my own and my results so far have been not good.
Could you give some tips or possibly a full workflow on how to get the results you do?
sure, try to find high resolution images for dataset, if images you found are low quality uou can try to use gigapixel ai to upscale them, manually remove any logos if there are any with photopea, open cmd window in image folder and do ren *.jpg *.png this will make all your jpg files into png, rename all files to 1 (ctrl a), use blip captioning to have a base captioning done, use visual studio code to manually caption each image with details you might want to keep or not depending on what you trying to achieve.
almost all my loras are made with a custom model, my latest one however is made with base sd 1.5, use base 1.5 so that your lora can work better with a lot of different other models
@truefalse Sounds pretty much exactly what I'm doing.
An approximate number of images? I have tried from 10 up to 100 of some cars.
I started out resizing everything to 512 on the long side. Then saw others saying it wasn't really needed. The last few batches were 1024 minimum. Good quality photos from multiple angles. Blip captioned with manual edits. Training with the SD 1.5 base.
Started out with up to 100 iterations per image. Now, I've gone as low as 4 with more epochs.
With all the changes and tweaks since I started, the quality overall is better. But still not what I would call good.
@100feet i do 20-30 images, now i do less than 50 steps per image, depends on how many images i have, 10 epochs or 9 minimum
@100feet also i do 768,768
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.