Dan Dan daydreaming: CG style handsome male lead | Tencent's DiT character poster wallpaper cover

0
0
0
0
CGI styleAnime StyleBoyStyle Boost
最近更新:初回公開:
CGI style,Anime Style,Boy,LoRA,HunyuanDiT v1.2Image info
CGI style,Anime Style,Boy,LoRA,HunyuanDiT v1.2Image info
CGI style,Anime Style,Boy,LoRA,HunyuanDiT v1.2Image info
CGI style,Anime Style,Boy,LoRA,HunyuanDiT v1.2Image info
CGI style,Anime Style,Boy,LoRA,HunyuanDiT v1.2Image info
CGI style,Anime Style,Boy,LoRA,HunyuanDiT v1.2Image info
CGI style,Anime Style,Boy,LoRA,HunyuanDiT v1.2Image info

DanDan is daydreaming: CG wind desire male protagonist with national style | Character cover | Realistic portrayal | Poster illustration

Character enhancement | Style reinforcement | CG rendering enhancement | Tencent DiT_V1.2 hybrid model

Model introduction: CG style, realistic anime portrait style, and dreamy extreme desire style.

Creative overview: The creator has recently been obsessed with the creative design of male protagonist characters. The more I look at posters of Chinese men and CG game IPs, the more flavorful they seem.

Model usage instructions: The checkpoint base film is trained based on the Tencent hybrid DiT_v1.2 large model.

When testing this Lora small model, other two-dimensional large models, three-dimensional large models, and realistic portrayal large models all produced good test images with ink effects.

Prompt keyword usage precautions:

1 Chinese boy, quality, masterpiece, original, high-resolution, extremely detailed wallpaper, perfect lighting, extremely detailed CG,

However, the visual effects of scenes where male characters are the main focus have not been sufficiently tested.

Other related parameter settings recommendations:

1. Sampling method: Eula A recommended for the Tencent hybrid DiT_v1.2 large model standard settings, iteration steps: 28.

Other sampling methods, DPM++2M Karras, Eula, iteration steps: 20-30 should suffice. SDE, can be between 30-60, or try LCM, iteration steps can be ≤15, etc.

2. Enable high-resolution repair: For realistic portrayal large models, recommend using 8x-NMKD-Superscale, 4x-UltraSharp enlargement algorithm; for two-dimensional, 3D large models, can use R-ESRGAN_4x+ Anime6B, R-ESRGAN_4x+, and other enlargement algorithms.

3. Test image resolution dimensions: Square and portrait images have more stable quality.

1:1 or 9:16 or 16:9 or 3:4 or 4:3, 512*512, 1024*1024; or 512*768; or 768*1024, 600*800, etc. Landscape image results depend on luck.

If using Tencent's hybrid model, it's not recommended to draw resolutions exceeding 1024*1024.

4. It is recommended to enable: After Detailer facial correction plugin, face_yolov8m, face_yolov8n, face_yolov8n_v2, etc.

Welcome all creators to use, return more illustrations, and have friendly exchanges.

ディスカッション

最も人気
|
最新
送信
もうすぐです
ダウンロード
(0.00KB)
詳細
タイプ
オンライン生成回数
0
ダウンロード
0
推奨パラメータ
推奨チェックポイントの組み合わせCheckpoint
なし
お勧めの重さ
0.8
CFG
7
VAE
なし

ギャラリー

最も人気
|
最新