Devour Technology | 3D Future Science Fiction - Stylized

Model Introduction: This model is trained to create a 3D sci-fi style character Lora with reference to the anime "Swallowed Starry Sky".

Recommendation: weight 1. Sampling DPM++ 2M SDE Karras iteration around 45, close-ups can reduce the number of iterations

① It is suggested that the output image size should not be less than 1280*768, using the high-definition restoration algorithm R-ESRGAN 4x+ Anime6B, around 20 iterations, redraw 0.3-0.4, the higher the full-body and distant perspective, the better. The same applies to facial redraw

② Close-up shots do not need to be high-resolution, otherwise they will appear plastic-like. Directly output a large image

③ AD facial correction algorithm, face_yolov8s.pt, around 0.4 of redraw. Close-ups also allow AD facial corrections, to restore the anime-like skin texture

④ The basic trigger words of the two characters should not appear at the same time unless you want to create a two-person image (untested)

⑤ If the description of other elements causes the colors of the characters' relevant parts to be incorrect, you can send the flawed image for partial redraw, then remove the affected trigger words, add the feature trigger words to be enhanced, activate the partial redraw or UP scale enlargement; if there are flaws on the face, remember to activate AD simultaneously