StableDef9 - PukiWiki

[ トップ ] [ 一覧 | 検索 | 履歴 | ログイン ]

最新の20件

2024-04-19

RevYOLOv5_2

2024-04-18

私的AI研究会

2024-04-16

RevYOLOv5

2024-04-14

ミーティング履歴

2024-04-10

YOLOv7_Colab3

2024-03-18

PyLearn

2024-03-17

2024-03-15

2024-03-05

RecentDeleted

2024-03-02

OpenVINOv2

2024-03-01

Anaconda1

2024-02-16

ProjectEnv3

2024-02-15

2024-02-14

SendMail

2024-01-21

GanFOMM

2024-01-18

ハードウェアTips

2024-01-17

進捗メモ-mizutu

私的AI研究会 > StableDef9

Stable-Diffusion9「人物画像を生成する」 †

　人物の画像生成は違和感を生じやすい、自然表現に近くなる手法を検討する。

Stable-Diffusion9「人物画像を生成する」
- 人物画像の生成方法
- 更新履歴
参考資料

※ 最終更新:2022/12/31　

人物画像の生成方法 †

引用ページ → Stable Diffusion プロンプトの組み立て方
サイトに従い、架空の人物 Mary Smith (アメリカ合衆国で一番女性の名前 “Mary”、苗字が “Smith”) を想定

違和感のない画像を生成するシード値を探す †

「メアリー・スミス」
Mary Smith

・シード値

0434215135	0922651506	0990636901	1117413489	1131199578	1139557532	1222445929	1705698508	1710439177	1823389825
1955217204	1985442815	2475085059	3132950872	3176422667	3230509125	3521458713	3805326286	4015905281	4229243702

・一番自然な最初のシード値「0434215135」をこれ以降は採用する

出力画像の再現

(py38) PS > python stable_diffusion_make2.py -i input_person1.csv -s n

▼「input_person1.csv」

-Seed-,-Prompt-,-InitImage-,-MaskImage-,-Step-,-Scale-,-Strengh-,-Device-,-Time-,-Output-,-Loop-
0434215135,Mary Smith,,,32,7.5,0.5,CPU,00:04:08,result_person/out000_0434215135.png,1
0922651506,Mary Smith,,,32,7.5,0.5,CPU,00:04:07,result_person/output_0922651506.png,1
0990636901,Mary Smith,,,32,7.5,0.5,CPU,00:04:25,result_person/output_0990636901.png,1
1117413489,Mary Smith,,,32,7.5,0.5,CPU,00:04:11,result_person/output_1117413489.png,1
1131199578,Mary Smith,,,32,7.5,0.5,CPU,00:04:08,result_person/output_1131199578.png,1
1139557532,Mary Smith,,,32,7.5,0.5,CPU,00:04:07,result_person/output_1139557532.png,1
1222445929,Mary Smith,,,32,7.5,0.5,CPU,00:04:08,result_person/output_1222445929.png,1
1705698508,Mary Smith,,,32,7.5,0.5,CPU,00:04:07,result_person/output_1705698508.png,1
1710439177,Mary Smith,,,32,7.5,0.5,CPU,00:04:06,result_person/output_1710439177.png,1
1823389825,Mary Smith,,,32,7.5,0.5,CPU,00:04:07,result_person/output_1823389825.png,1
1955217204,Mary Smith,,,32,7.5,0.5,CPU,00:04:55,result_person/output_1955217204.png,1
1985442815,Mary Smith,,,32,7.5,0.5,CPU,00:04:09,result_person/output_1985442815.png,1
2475085059,Mary Smith,,,32,7.5,0.5,CPU,00:04:07,result_person/output_2475085059.png,1
3132950872,Mary Smith,,,32,7.5,0.5,CPU,00:04:08,result_person/output_3132950872.png,1
3176422667,Mary Smith,,,32,7.5,0.5,CPU,00:04:09,result_person/output_3176422667.png,1
3230509125,Mary Smith,,,32,7.5,0.5,CPU,00:04:09,result_person/output_3230509125.png,1
3521458713,Mary Smith,,,32,7.5,0.5,CPU,00:04:12,result_person/output_3521458713.png,1
3805326286,Mary Smith,,,32,7.5,0.5,CPU,00:04:08,result_person/output_3805326286.png,1
4015905281,Mary Smith,,,32,7.5,0.5,CPU,00:04:27,result_person/output_4015905281.png,1
4229243702,Mary Smith,,,32,7.5,0.5,CPU,00:04:06,result_person/output_4229243702.png,1

写真と絵画の違いを出す †

「メアリー・スミス」指定なし / 写真 / 油絵 / 鉛筆画 / 水彩画
Mary Smith / ,photo / ,oil painting / ,pencil draw / ,watercolor painting

出力画像の再現

(py38) PS > python stable_diffusion_make2.py -i input_person2.csv -s n

▼「input_person2.csv」

-Seed-,-Prompt-,-InitImage-,-MaskImage-,-Step-,-Scale-,-Strengh-,-Device-,-Time-,-Output-,-Loop-
0434215135,"Mary Smith",,,32,7.5,0.5,CPU,00:04:08,result_person/out000_0434215135.png,1
0434215135,"Mary Smith, photo",,,32,7.5,0.5,CPU,00:04:15,result_person/out001_0434215135.png,1
0434215135,"Mary Smith, oil painting",,,32,7.5,0.5,CPU,00:04:16,result_person/out002_0434215135.png,1
0434215135,"Mary Smith, pencil draw",,,32,7.5,0.5,CPU,00:04:15,result_person/out003_0434215135.png,1
0434215135,"Mary Smith, watercolor painting",,,32,7.5,0.5,CPU,00:04:12,result_person/out004_0434215135.png,1

色合いを指定する †

「メアリー・スミス」写真 / 油絵 / 鉛筆画 / 水彩画 , < カラー / モノクロ >
Mary Smith ,photo / ,oil painting / ,pencil draw / ,watercolor painting ,< RGB / monotone >

出力画像の再現

(py38) PS > python stable_diffusion_make2.py -i input_person3.csv -s n

▼「input_person3.csv」

-Seed-,-Prompt-,-InitImage-,-MaskImage-,-Step-,-Scale-,-Strengh-,-Device-,-Time-,-Output-,-Loop-
0434215135,"Mary Smith, photo, RGB",,,32,7.5,0.5,CPU,00:04:16,result_person/out005_0434215135.png,1
0434215135,"Mary Smith, oil painting, RGB",,,32,7.5,0.5,CPU,00:04:15,result_person/out006_0434215135.png,1
0434215135,"Mary Smith, pencil draw, RGB",,,32,7.5,0.5,CPU,00:04:11,result_person/out007_0434215135.png,1
0434215135,"Mary Smith, watercolor painting, RGB",,,32,7.5,0.5,CPU,00:04:09,result_person/out008_0434215135.png,1
0434215135,"Mary Smith, photo, monotone",,,32,7.5,0.5,CPU,00:04:09,result_person/out009_0434215135.png,1
0434215135,"Mary Smith, oil painting, monotone",,,32,7.5,0.5,CPU,00:04:10,result_person/out010_0434215135.png,1
0434215135,"Mary Smith, pencil draw, monotone",,,32,7.5,0.5,CPU,00:04:10,result_person/out011_0434215135.png,1
0434215135,"Mary Smith, watercolor painting, monotone",,,32,7.5,0.5,CPU,00:04:10,result_person/out012_0434215135.png,1

作風を指定する †

「メアリー・スミス」ムンク / ピカソ / ゴッホ / モネ
Mary Smith , art by Edvard Munch / , art by Pablo Picasso / , art by Vincent van Gogh / , art by Claude Monet

出力画像の再現

(py38) PS > python stable_diffusion_make2.py -i input_person4.csv -s n

▼「input_person4.csv」

-Seed-,-Prompt-,-InitImage-,-MaskImage-,-Step-,-Scale-,-Strengh-,-Device-,-Time-,-Output-,-Loop-
0434215135,"Mary Smith, photo, art by Edvard Munch",,,32,7.5,0.5,CPU,00:04:13,result_person/out013_0434215135.png,1
0434215135,"Mary Smith, oil painting, art by Pablo Picasso",,,32,7.5,0.5,CPU,00:04:22,result_person/out014_0434215135.png,1
0434215135,"Mary Smith, pencil draw, art by Vincent van Gogh",,,32,7.5,0.5,CPU,00:04:10,result_person/out015_0434215135.png,1
0434215135,"Mary Smith, watercolor painting, art by Claude Monet",,,32,7.5,0.5,CPU,00:04:09,result_person/out016_0434215135.png,1

人物の特徴指定 †

写真向けのキーワード Mary Smith, portrait, Canon EOS 5D Mark II, HDR, trending on 500px

キーワード＋年齢的特徴（young, middle age, old）と beautiful (なしとあり)
<Keyword> , young / < > / < > , middle age / < > , old
<KeyWord> , young / < > , beautiful / < > , beautiful / < > ,middle age, beautiful / < > , old, beautiful

出力画像の再現

(py38) PS > python stable_diffusion_make2.py -i input_person5.csv -s n

▼「input_person5.csv」

-Seed-,-Prompt-,-InitImage-,-MaskImage-,-Step-,-Scale-,-Strengh-,-Device-,-Time-,-Output-,-Loop-
0434215135,"Mary Smith, portrait, Canon EOS 5D Mark II, HDR, trending on 500px",,,32,7.5,0.5,CPU,00:04:13,result_person/out020_0434215135.png,1
0434215135,"Mary Smith , young, portrait, Canon EOS 5D Mark II, HDR, trending on 500px",,,32,7.5,0.5,CPU,00:04:13,result_person/out021_0434215135.png,1
0434215135,"Mary Smith, middle age, portrait, Canon EOS 5D Mark II, HDR, trending on 500px",,,32,7.5,0.5,CPU,00:04:12,result_person/out022_0434215135.png,1
0434215135,"Mary Smith, portrait, old, Canon EOS 5D Mark II, HDR, trending on 500px",,,32,7.5,0.5,CPU,00:04:09,result_person/out023_0434215135.png,1
0434215135,"Mary Smith, beautiful, portrait, Canon EOS 5D Mark II, HDR, trending on 500px",,,32,7.5,0.5,CPU,00:04:10,result_person/out024_0434215135.png,1
0434215135,"Mary Smith , young, beautiful, portrait, Canon EOS 5D Mark II, HDR, trending on 500px",,,32,7.5,0.5,CPU,00:04:12,result_person/out025_0434215135.png,1
0434215135,"Mary Smith, middle age, beautiful, portrait, Canon EOS 5D Mark II, HDR, trending on 500px",,,32,7.5,0.5,CPU,00:04:10,result_person/out026_0434215135.png,1
0434215135,"Mary Smith, portrait, old, beautiful, Canon EOS 5D Mark II, HDR, trending on 500px",,,32,7.5,0.5,CPU,00:04:11,result_person/out027_0434215135.png,1

男性でやってみる †

写真向けのキーワード James Smith, portrait, Canon EOS 5D Mark II, HDR, trending on 500px

キーワード＋年齢的特徴（young, middle age, old）と handsome (なしとあり)
<Keyword> , young / < > / < > , middle age / < > , old
<KeyWord> , young / < > , handsome / < > , handsome / < > ,middle age, handsome / < > , old, handsome

出力画像の再現

(py38) PS > python stable_diffusion_make2.py -i input_person6.csv -s n

▼「input_person6.csv」

-Seed-,-Prompt-,-InitImage-,-MaskImage-,-Step-,-Scale-,-Strengh-,-Device-,-Time-,-Output-,-Loop-
0434215135,"James Smith , portrait, Canon EOS 5D Mark II, HDR, trending on 500px",,,32,7.5,0.5,CPU,00:04:11,result_person/out028_0434215135.png,1
0434215135,"James Smith , young, portrait, Canon EOS 5D Mark II, HDR, trending on 500px",,,32,7.5,0.5,CPU,00:04:12,result_person/out029_0434215135.png,1
0434215135,"James Smith, middle age, portrait, Canon EOS 5D Mark II, HDR, trending on 500px",,,32,7.5,0.5,CPU,00:04:10,result_person/out030_0434215135.png,1
0434215135,"James Smith, portrait, old, Canon EOS 5D Mark II, HDR, trending on 500px",,,32,7.5,0.5,CPU,00:04:10,result_person/out031_0434215135.png,1
0434215135,"James Smith, handsome, portrait, Canon EOS 5D Mark II, HDR, trending on 500px",,,32,7.5,0.5,CPU,00:04:10,result_person/out032_0434215135.png,1
0434215135,"James Smith , young, handsome, portrait, Canon EOS 5D Mark II, HDR, trending on 500px",,,32,7.5,0.5,CPU,00:04:10,result_person/out033_0434215135.png,1
0434215135,"James Smith, middle age, handsome, portrait, Canon EOS 5D Mark II, HDR, trending on 500px",,,32,7.5,0.5,CPU,00:04:10,result_person/out034_0434215135.png,1
0434215135,"James Smith, portrait, old, handsome, Canon EOS 5D Mark II, HDR, trending on 500px",,,32,7.5,0.5,CPU,00:04:09,result_person/out035_0434215135.png,1

更新履歴 †

2022/12/28 初版

参考資料 †

「Stable Diffusion」実用性

「Stable Diffusion」プロンプトについて
- Stable Diffusion プロンプトの組み立て方
- Stable Diffusion でプロンプトに複数の作家名を並べる理由について

「Stable Diffusion」の次へ
- Stable diffusionから派生した画像生成AI「stablediffusion-infinity」と「Waifu Diffusion」違いはどこ？　使い方も紹介
- lkwq007/stablediffusion-infinity

Last-modified: 2022-12-31 (土) 09:40:11