Enhancement bundle for Lumina
All cover images are raw output from the model, 1MP resolution, no upscale, no hands/faces inpainting fixes, even no negative prompt.
Make Lumina Image 2 look better.
Currently support:
NetaYume Lumina (NTYM)
Trained with total ~7k images.
Photographs, digital arts, anime images, space images ... everything I can come up with. Many specialized sub datasets, such as close-up clothing, hands, complex ambient lighting ...
Only high resolution images with finest details. The whole dataset avg pixels is 3.37 MP, ~1800x1800. Every image is hand-picked by me.
Comprehensive natural language captions from Google LLM.
Anime characters are tagged by wd tagger v3 first and then refined to natural language by Google LLM.
Effect:
Better backgrounds, natural textures, lighting, and less noise.
Maybe slightly better creativity and prompt following
Why a bundle, rather separated LoRAs for different aspects?
This a mathematic problem. You can ask an AI for a quick answer. "When using an image model, what's the problem if I stacked too many LoRAs on my image model? In mathematic aspect."
If you train all your data and concepts in one go, then there will be no conflicts.
Furthermore, larger datasets can prevent overfitting. Even if you only use this model for anime, the knowledge from photographs can still teach the model how to add more details to anime. This is why we all say that "bigger is better" when it comes to datasets.
How to use
Strength 0.8~1 is recommended.
For anime image in the dataset, I used "Digital anime illustration." in prefix as trigger words. You can use this as well.
License
This model is released under Apache License 2.0.
Additional terms: Selling or monetizing models that merged this LoRA is prohibited.
Update logs
Note: version number means nothing, just my training id.
v0.27: Better and stronger effects.
v0.11: Trained on NetaYume v3.5. Full run. Trained with all my dataset.
v0.8: init version. Trained on NetaYume v3. Testing run. Total training data ~1k.