Stable-Diffusion v1.5 fine-tuned for 10k steps upon a blip captioned dataset of vintage images, then it underwent further fine tuning with dreambooth using the same dataset but rather then having it blip-captioned, it was split into "Vintage style", "Vintage face" and "Pulp cover" concepts
The model card and link to the dataset are available on Huggingface
Description
This version was first fined-tuned on the whole dataset and only then it was fine-tuned again to learn each individual concept, so you can use it even without Trigger-Words and still get a subtle "Vintage" touch
FAQ
Comments (2)
Very cool idea, I'm enjoying playing around with it so far. Any tips on recommended image sizes? I'm getting a lot of the images cropped at the top with 768 and 512. Also i tried the hires fix and it didn't turn out to well with that option on. any recommendations would be appreciated . Thanks!
This model does indeed suffer when out-of 512 resolutions are used. I did manage to get some decent ones on 768x1024: https://civitai.com/gallery/17241?modelId=1952&modelVersionId=2099&infinite=false&returnUrl=%2Fmodels%2F1952 Interestingly, I don't recall having an issue with crops while generating in 512x512 :(
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.