Chinese Animation: Qianhuan | Legend of Zu - Jin Ping'er

Model Introduction: This model is the single-character SDXL Lora from the Chinese animated series "Zhu Xian," featuring extreme realism and transformation capabilities.

Basic Trigger Word: Gillian

Realistic Image: A Gillian-type woman with black long straight hair wearing a golden Hanfu of JPE type, where the color of the clothing can be changed to [golden] by replacing other words.

Parameter Recommendations

① For SDXL image generation, do not enable VAE, the image resolution should be larger than 768*1024

② Weights: Text-image 0.9, image-image redraw can be 1.0, better transformation 0.9

③ Sampling: DPM++ 2M SDE Karras (results in finer images), iterations 25-45. Euler a sampling can also be used to improve the accuracy of text-image, but the generated images may not be detailed enough, especially the background.

④ CFG: 6.5-7.0, CLIP: 1

⑤ High-Definition Restoration: Algorithm R-ESRGAN 4x+ Anime6B, around 25 iterations, high-definition redraw about 0.4, increase redraw if the image is blurry, decrease if deformed.

⑥ AD Face Enhancement: Algorithm face_yolov8s.pt, about 0.35 redraw intensity, adjustment techniques similar to high-definition restoration, not recommended to enhance face in close-up situations.

⑦ If AD face enhancement still cannot restore the facial features, it is recommended to send the image for image redraw or use upscale scripts for enlargement, redraw intensity 0.3-0.4, use Euler a sampling.

Other Recommendations

Model Suggestions for Image Generation: For 3D, recommend using Zhumei Industry V5, V4, 2D image generation not tested yet.

Transformation Aesthetic: Can include positive descriptors (eye makeup: 0.7), NSFW, cleavage, shiny skin, and other aesthetic terms.

Universal Positive Terms: highly detailed, ultra-high resolutions, 32K UHD, best quality, masterpiece

Universal Negative Terms: bad hand, blurry