Type | |
Stats | 230 42 |
Reviews | (30) |
Published | Oct 12, 2024 |
Base Model | |
Training | Steps: 10 Epochs: 130 |
Usage Tips | Clip Skip: 1 Strength: 0.7 |
Trigger Words | 360-degree panoramic photography |
Hash | AutoV2 FFD1CFB523 |
This Lora is an experimental model still in developed to integrate characters into 360-degree panoramic photos. While there are many Loras focused on landscapes, I wanted to push the boundaries by including characters within these scenes.
As this is still in the experimental stage, you may encounter some glitches in the generations, but these can often be fixed with inpainting and upscaling. I’m continually expanding my dataset of panoramic images, but the results so far have been promising at this stage.
The training involved 130 epochs and 10k steps using 340 images, though I believe adding more images to the dataset will further stabilize the model.
I've tested this Lora with several other models, and you may need to adjust the strength for each combination to get optimal results. I recommend experimenting with different settings to find what works best for your project.
A prompt structure that works well includes:
"A 360-degree panoramic photo taken by a VR headset" + "character description" + "background."
Resolution: 2.1
Better results at 1920x960, but also tested with 1536 x 768, below that, glitches are unbearable.
Support My Work
If you enjoy this project and would like to support my future developments, consider making a donation or becoming a patron! Your support helps me continue experimenting and creating new tools for the community.