【StableDiffusion】同じ人物で違う服・ポーズにしたい！Seed（シード）を使ってみた

Seika_USA

2024年4月11日 04:16

こんにちは。アメリカ・シリコンバレー在住のSeikaです。

現在、StableDiffusionを使って、同じ人物で違う服・ポーズにすることを試みています。
調べているとSeed（シード）という言葉が出てきました。
そこで今回はSeed値を使って生成した画像を比較していきたいと思います！

失敗作も多いのですが過程を残すのも大事だと思いますので、noteにまとめています。ご了承ください。

（2024年4月10日公開）

Seed(シード)値の固定とは？

Stable Diffusionでは、Seed値と呼ばれるランダムな数値を使用して画像を生成します。
Seed値を固定することで、同じ人物の画像を複数生成することができます。

Seed値の固定方法

Stable DiffusionのWebUIで、"Generation"タブを開く。
"Seed"欄を確認
デフォルトはランダムになるよう-1になっています。
ここに任意のSeed値を入力。
"Generate"ボタンをクリックして、画像を生成。
先ほどの任意で指定したSeed値を入力、別のプロンプトで生成。

Seed値1

まずSeed値1を入力してみます。

Top quality, 1 beautiful woman, long Hair, brown hair, wearing Camisole & skirt, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

Seed値5

今度は適当に５。

Top quality, 1 beautiful woman, long Hair, brown hair, wearing Camisole & skirt, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 5, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

ここでファイル名にいつもとの違いを発見。
Seed値が反映されています。

Seed値0

Seed値0ではどうなるのでしょうか。

Top quality, 1 beautiful woman, long Hair, brown hair, wearing Camisole & skirt, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 0, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

Seed値-1

今度はランダムに出る−１。

Top quality, 1 beautiful woman, long Hair, brown hair, wearing Camisole & skirt, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: -1, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

Seed値−2

-2ではどうでしょう。

Top quality, 1 beautiful woman, long Hair, brown hair, wearing Camisole & skirt, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: -2, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

Seed値−4

マイナスの値を大きくします。

Top quality, 1 beautiful woman, long Hair, brown hair, wearing Camisole & skirt, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: -4, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

Seedはランダムになる−１以外は、同じ画像になることが確認できました。

服装を変える

今度はSeed値を固定して、服装を変えてみたいと思います。
Seed値１の時に生成したこちらの画像を使います。

プロンプトでは、キャミソールとスカートとなっていた服をトレーナーに変えてみます。

Top quality, 1 beautiful woman, long Hair, brown hair, wearing sweatshirt, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

今度はシャツにします。

Top quality, 1 beautiful woman, long Hair, brown hair, wearing shirt, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

チュニックトップという通常のトップスより長い洋服にしてみます。

Top quality, 1 beautiful woman, long Hair, brown hair, wearing tunic top, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

構図・ポーズを変える

「下から」という単語を入れてみます。

Top quality, 1 beautiful woman, long Hair, brown hair, wearing sweatshirt, shy-smile, from below, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) lora:ip-adapter-faceid-plusv2_sd15_lora:1

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

グッドボタンポーズ👍

グッドボタンのポーズをしてもらいたい👍

Top quality, 1 beautiful woman, long Hair, brown hair, wearing sweatshirt, shy-smile, thumbs up, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) lora:ip-adapter-faceid-plusv2_sd15_lora:1

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

親指を立てるポーズを優先的に反映するために、プロンプトの前の方に記述します。
3つ目にthumbs upを書きました。

Top quality, 1 beautiful woman, thumbs up, long Hair, brown hair, wearing sweatshirt, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

今度は先頭にします。

thumbs up, Top quality, 1 beautiful woman, long Hair, brown hair, wearing sweatshirt, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

腕を上げる

腕は上げられるのでしょうか？

arm up, Top quality, 1 beautiful woman, long Hair, brown hair, wearing sweatshirt, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

ガオ〜のポーズ🦁

claw pose, Top quality, 1 beautiful woman, long Hair, brown hair, wearing sweatshirt, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

腕組のポーズ

腕組のポーズではどうでしょう。

crossed arms, Top quality, 1 beautiful woman, long Hair, brown hair, wearing sweatshirt, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

全身

もっとカメラを引いて、全身を確認したい。
standingという単語を入れてみました。

standing, Top quality, 1 beautiful woman, long Hair, brown hair, wearing sweatshirt, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

fullbody（全身）にチェンジ。

fullbody, top quality, 1 beautiful woman, long Hair, brown hair, wearing shirt, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

Seedが良くない気がしたので、ランダム値の-1にしてみます。

fullbody, top quality, 1 beautiful woman, long Hair, brown hair, wearing shirt, shy-smile, Sunlight, at street, 35mm Lens, /1, (white Background: 1.3) <lora:ip-adapter-faceid-plusv2_sd15_lora:1>

Negative prompt: sfw ng:deepnegatives:v.1.7.5t, ((badhandv4)), (worst quality:2), (low quality:2), (normal quality:2), lowres, bad anatomy, bad hands, normal quality, ((monochrome)), watermark

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: -1, Size: 512x512, Model hash: 1a17bcd93d, Model: beautifulRealistic_v7, ControlNet 0: "Module: none, Model: None, Weight: 2, Resize Mode: Crop and Resize, Low Vram: False, Guidance Start: 0, Guidance End: 1, Pixel Perfect: True, Control Mode: Balanced, Hr Option: Both, Save Detected Map: True", Lora hashes: "ip-adapter-faceid-plusv2_sd15_lora: a95a0f4bdcb9", Version: v1.8.0

同一人物って難しい・・・
今回はここまでで断念。

Seed(シード)値の調べ方

最後に、過去に生成した画像のSeed(シード)値の調べ方を記述します。

「PNG info」タブをクリック。
画像をドロップ。
「parameters」で値を確認。

今回は、Seed値を使って同じ人物で違う服・ポーズを試みました。
Seed値を使った方が同一人物に近づける気がします。
上手く作れる人はもっと上手いので、引き続き研究していいプロンプトを発見していきたいと思います。
もし記事が役に立ったと思ったら、ぜひスキボタンを押してくださいね😊
それでは、次回の記事でお会いしましょう！

Follow me here😊

参照

この記事が参加している募集

#AIとやってみた

36,471件

よろしければサポートお願いいたします。いただいたサポートはクリエイターとしての活動費に使わせていただきます。