Dreamy Spring Night | God Inverse - Li Muwan & **(Newly built model - available for single/double occupancy)
V1.0




The farthest distance in the world is not life and death, but forgetting. - 【Renegade Immortal】
Model introduction: This model is the new version of Chinese anime 【Renegade Immortal】 ** and Limuwan, which can stably produce both single and double images, achieve extreme restoration, and also support transformation and costume changes.
Limuwan trigger word: Limuwan
Limuwan image restoration: a girl with long black hair wearing white tulle and blue long skirt, the clothing color can be customized for transformation
** trigger word: *****
** image restoration: a man with a black high ponytail and parted bangs wearing blue traditional Chinese clothing (Hanfu), the clothing color can be customized for transformation
First meeting scene between new ** Limuwan: outside the heavenly palace
Peach blossom flying effect: Peach Blossoms Flying
Outside Yun Tian Zong Palace: outside the palace
Parameters for double images:
Lora weight 0.9
Sampling DPM++ 2M SDE Karras, iterations around 25-45
CFG 6.5 or 6
Keywords to write: 【interactive action】+【a girl with long black hair wearing white tulle and blue long skirt】+【a man with a black high ponytail and parted bangs wearing blue traditional Chinese clothing (Hanfu)】+pose+background environment, please refer to my display images and reverse images for specifics
Note: The double images displayed are the parameters and keywords for redrawing. For a direct output of double images, please refer to the above, and a similar image is needed for simple redraw and restoration
Recommended parameters
① For SDXL image output, do not use VAE, the image should be larger than 768*1024
② Weight: text-to-image 0.9, redrawn image 1.0 is better for transformation
③ Sampling: DPM++ 2M SDE Karras (produces finer images), iterations 25-45. Euler a sampling can also be used to improve the accuracy of text-to-image, but the images are not as detailed, especially the background
④ CFG: 6.5. In special cases, try 6.0 or 7.0, CLIP: 1
⑤ High-resolution restoration: algorithm R-ESRGAN 4x+ Anime6B, iterations around 25, high-resolution redraw about 0.4, increase redraw if blurry, decrease if deformed
⑥ AD face retouching: algorithm face_yolov8s.pt, redraw intensity around 0.35, adjustment technique similar to high-resolution restoration, but not recommended for close-ups
⑦ If AD face retouching cannot restore the face shape, it is recommended to send the image generation image for redraw or use the upscale script to enlarge, redraw intensity 0.3-0.4, use Euler a for sampling
Other recommendations
Model recommendation for image output: For 3D, it is recommended to use Dream Building Industry V5 and V4, 2D image output has not been tested yet
Limuwan transformation aesthetics: Can be positively described (eye makeup: 0.7), NSFW, cleavage, shiny skin, and other aesthetic terms
Common positive words: highly detailed, ultra-high resolutions, 32K UHD, best quality, masterpiece
Common negative words: bad hand, blurry
