Instructions for best results:

1. Use at full strength (1.0) with SDXL Base

2. Generate at 1024 x 1024

3. Start your prompt with "Polaroid of"

4. Use FreeU_V2 if you want to add more saturation and contrast, rather than another fine tuned model (the node is built into ComfyUI, and can be installed in Auto1111). Most other models refuse to generate the border correctly.

About:

This LoRA was trained from scans of 35 of my favorite Polaroids, hence the white background. My goal was to be able to generate convincing Polaroids with the correct border dimensions. Someday I may may modify the training images to have a less stark background, but I wanted to put this out there for now in case it's useful to someone as-is.

Limitations:

1. Getting the correct border dimensions, combined with a square image inside is somewhat rare (maybe 1 out of every 7 gens), and seems to be highly dependent on the seed. All my examples are cherry-picked.

2. As with many LoRAs that deviate far from the standard base model, short prompts work best.

3. This LoRA falls apart when used with most other models besides SDXL Base. If you find one that works well, let me know!