God descending upon the virtual star | Devouring the sky - Featuring Tianchen & Minyu

Devouring darkness, making the stars pale; breaking the rules, causing the sea to tremble. ——[Devouring the Stars]

Model Introduction

This model is the character Lora from the anime [Devouring the Stars], which is the main character Tianchen and the domain lord Mingyu, can be restored separately through different basic trigger words, including faces, mechas, character decorations, and even original anime effects! It is a character Lora that contains multiple concepts, representing various decorations, effects, and backgrounds that can be singular or mixed, and supports character transformation, mecha transformation!

Model Usage Instructions

Set Lora weight as 0.9 for image restoration, only retaining the face is recommended to set the image weight at 0.8, for 2D images, a weight of 1 is allowed

Tianchen Sampling: DPM++ 2M SDE Karras Iteration around 45

Tianchen Basic Trigger Word: Jatich. With this trigger word placed at the beginning, follow the instructions below to restore the character's other accessories

Original Mecha: wearing white and gold armor

Original Wings: white mechanical wings

Mecha Helmet: helmet, with a face mask it's a helmet with a white mask

Original Back Accessory: gold metal ring in the background

Virtual Background: dark blue background with lightning

Mingyu Sampling recommendation: For a full-body image, recommend using DPM++ 2M SDE Exponential with Iteration around 45, basic trigger words and clothing descriptions are advised to set weight at 1.1, for non-full-body images recommend using DPM++ 2M SDE Karras

Mingyu Basic Trigger Word: MinYu. With this trigger word placed at the beginning, select SDXL's 3D style sampling, then follow the instructions below to restore the character's other accessories

Original Mecha: a mechanical cyborg man wearing a silver mechanical suit

Holographic Screen: looking at a futuristic screen

Holographic Scene: holographic special effects background

Image Detail Attention (applicable to both characters):

① High-definition restoration algorithm R-ESRGAN 4x+ Anime6B, around 20 iterations, redrawn 0.3-0.4, the more distant the full-body image, the higher the redrawn. The same applies to facial redrawn

② Close-up images do not need high resolution, otherwise they may appear plasticky. Directly outputting a large image is acceptable

③ AD facial retouching algorithm, face_yolov8s.pt, around 0.4 for redrawn. Close-up images also allow AD facial retouching, to achieve an anime-like skin texture

④ The basic trigger words for the two characters cannot appear simultaneously unless you want to create a two-character image (not tested)

⑤ If the colors of the character-related parts are incorrect due to descriptions of other elements, you can send the flawed image for partial redraw, then remove the affected trigger words and add the features that need to be enhanced to the output prompt words, start partial redraw or UP scale enlargement, if it's a facial flaw, remember to turn on AD

Easter Egg!! The following trigger words restore some anime scenes, feel free to try them out (effects applicable to both characters), go and try your luck with the card draw

Background ①: red giant star in the background

Background ②: golden soft light, golden special effects, adding a dark sky creates a cosmic star background effect

Divine Descent Scene: gold metal ring in the background, golden soft light, these words are most effective for full-body images

Set Lora weight as 0.9 for image restoration, retaining only the face (transformation) is suggested to set the image weight at 0.8, for 2D images, a weight of 1 is allowed