This is a LoRA trained with images by the artist Mejikara Scene. Recommended strength is between 1-0.6. Trigger word is 'mejikara scene'.
The training data had lots of speech bubbles, so I recommend putting 'speech bubble' in negative and if they still appear, try lowering LoRA weight gradually until they are gone. It also has a strong bias for white pupils, so if you don't want that put it in negative as well.
How I generate images:
Local generation using reForge
Sampler: Euler Ancestral CFG++
Schedule type: Normal
Sampling steps: 30
CFG Scale: 1.571428
Img2Img Upscale or Hires. fix: 1.5x at 0.428571 Denoising strength
ADetailer to improve faces
If you use on-site generation or otherwise don't have CFG++ samplers, you can use CFG Scale 7 and achieve similar results.
If you like what this can do, please take a moment to post an image to this model, it helps me out a lot. Thank you!