Innocent Girl - Li Qi

About Version 1.5: This version is a temporary iterative version that integrates a small amount of KoreanDollLikeness based on version 1.0 to correct the issue of focusing too much on the face when generating images. Overall, it is better than version 1.0. When I have some free time later, I will reorganize more full-body photo datasets to correct this issue.

Summoning word: liqi  This is also the name I gave to this LORA model (Li Qi), not including this summoning word has no significant impact. 

LORA recommended weight: (0.7-1.0)

VAE: vae-ft-mse-84000-ema

Base model suggestions: CNrealistic_MIX_V40, majicMIX realistic V7 version, Moyou artificial humans, and other realistic large models, tested models include CNrealistic_MIX_V40 large model output, enhancing the sense of girlishness, therefore highly recommended!!

Live image parameters: DPM++ SDE Karras  Iterations: 35  Size: 512*768

Recommend enabling high-definition restoration

High-definition restoration settings: Scaling algorithm: 4X-UltraSharp Redrawing magnitude 0.3-0.35   High-definition restoration sampling : 30 Scaling factor: 1.5 times or more

Can enable ADetailer facial optimization plugin to effectively avoid face collapse issues

Recommended starting words for positive prompts: (masterpiece:1.2), best quality, high resolution, (realistic:1.2), (photorealistic:1.4), (solo:1.2), (1girl:1.2), beautiful detailed eyes and face, (detailed skin texture:1), 8k, RAW photo Note: Due to more frequent face enhancement training, when writing prompt words, please add some clothing descriptions based on the starting prompt words!!

Releasing the trained model for the first time actually makes me a bit nervous. If there are any issues, I hope the experts will point them out.