This year, let AI hide easter eggs for you.
This SDXL LoRA experiment will place a small number of tiny easter eggs somewhere in the generated image. Apply. Generate. Have fun searching!
RealMessyEaster is trained on 75 labelled images of single small plastic eggs in a messy surrounding – mostly at the edges. With two goals:
Integrating an easter egg to look for.
Adding "messyness" to the background, to make eggs a little harder to spot.
Triggers: 1easteregg, messy
Anyway, have fun playing. Have a Messy Easter
🥙 New for Easter 2025: I pulled up a Huggingface Space to play with the model directly. Head to https://huggingface.co/spaces/5to9/easteregg 🥙
Description
This stage is far from perfect and needs further balancing. Known issues:
Long detailed prompts lower the reliability of easter egg generation. Level up the Lora's weight to 1, crank up the weight of the trigger e.g. (1easteregg:1.2).
Surroundings where you might expect colourful round objects tend to generate too many eggs. Reduce the weight of the Lora to 0.8 lower. Add "ball" to negative prompt.
Eggs are rarely hidden behind other objects. This will take a different set of training images.
Eggs are mostly yellow or orange and pop out. But give other colours a try, too.