Character Fuli (LoRa)

Versions 1.1-1.2:

LoRa was trained on the basis of Checkpoint Midnight Doll_XL, with the basic Illustrious model.

The following parameters were used for training:

Repeat: 13

Epoch: 15

With 53 images of the combined or separate themes described above. Total number of steps: 5168.

Versions 2.0:

LoRa was trained on the basis of Checkpoint RealCartoon-Pony, with the basic Pony model. It fits well Midnight Doll_XL, as well as many others.

RealCartoon-Pony — With this LoRa, it gives pleasant matte tones and high detail;

Midnight Doll_XL — With this LoRa makes glossy, clear and stylish images.

The following parameters were used for training:

Repeat: 17

Epoch: 10

With 52 images of the combined or separate themes described above. Total number of steps: 8840.

Recommended parameters for generating:

To activate LoRA character, use tags:

Character_Fuli

Checkpoint: Midnight Doll_XL (v1.1-v2.0), RealCartoon-Pony (v2.0)

Strength: ~0.7-1.0

Sampling method: Euler a, DPM++ 2M Karras

Sampling Steps: 20-40

CFG Scale: 2.5-5.5

Clip Skip: 1

*for Midnight Doll_XL, you can try using Upscale R-ESRGAN 4x+ Anime6B

Promt example: Character_Fuli, Sitting on a carpet in a spacious room. The warm sunset light rays from the windows. Masterpiece, best quality, ultra-detailed, cinematic emotional static frame, 8K UHD, extreme detail, physically based rendering, HDR. Unrealistic image, fantasy, dramatic lighting. Highly detailed style of digital art, (SFW:1.3).

Negative prompt: (NSFW:1.2),(bad quality:1.2),(worst quality:1.2), watermark, (blurry), (fullness), (belly), low quality, normal quality, text, signature, jpeg artifacts, bad ???????, copyright name, artist name, signature, verybadimagenegative_v1.3, ng_deepnegative_v1_75t,  cross-eyed, sketches, bad antomy, DeepNegative, {Multiple people}.