七夢
sengoku war chronicle poster style, large side profile portrait, armored horse in the center, mounted heroine, burning castle, samurai army, blue sky, red flames, gold border, painterly gouache texture,
１．はじめに
本スタイルLoRAは、信長の野望ゲームボーイ版のパッケージイラストの再現を狙い作成に着手しましたが、あの独特の渋みのあるかっこいい感じにはなっておりません。
また、パッケージイラストともだいぶかけ離れています。言われれば、似てるところもあるかも・・ぐらいです。
下記の構成でイラストを生成するスタイルLoRAです。
・騎馬にまたがる人物 ・人物の横顔 ・炎上する背景とお城 ・兵士
もし興味があって遊んでみようかな？という場合には、以下の説明をご一読いただけると、良いのかなと思います。
クレジット的に考えて(。-`ω-)
２．超重要事項
LoRA資料はまったく問題ないのですが、「な・ぜ・か」生まれたままの無垢なお姿を披露するケースが、ごくマレに存在します。プロンプトも問題はないので、まったく謎です。AI様の趣味趣向としか思えません。
確率は結果から推測すると、１／５０生成ぐらいです。
基本的にはありませんが、どうしても回避したい場合には、キャラクターの服装をプロンプトで指定をする・ネガティブで禁止するなどの対応をお願いします。
本当にAIさんは謎だらけです(´ω｀)
３．ご説明
・LoRA資料は「PixAI Edit Pro」を使用し、統一されたもので作成されています。が、変幻自在です。一応、はじめの部分で記した内容の方向ではありますが、とにかく安定しません。
・サイズは１：１の正方形サイズが推奨です。
・モードはウルトラが推奨です。スタンダードでも遊べなくはありません。ただし、以下に示すCまで対応させてしまうことで、「１はじめに」に記載した、イラストが生成しやすくなるように感じています。
・他のLoRAと合わせる場合、合わせるLoRA次第で、さらに、大きく変化します。
・はじめに記載した内容と違う場合には、LoRAの強さの調整が重要な鍵を握ります。
A・LoRA作成時の初期トリガーは以下です。
「sengoku war chronicle poster style, large side profile portrait, armored horse in the center, mounted heroine, burning castle, samurai army, blue sky, red flames, gold border, painterly gouache texture,」
B・初期トリガーで上手くいかない場合には、初期トリガーは削除し、下記のようにプロンプトを変更してみてください。
「sengoku war chronicle poster style, painterly gouache texture, gold border, large side profile portrait on the left, mounted heroine, samurai army, burning Japan's castle, blue sky and red flames, central horse with a natural unarmored head, simple bridle, simple reins, plain horse face, no head armor, with a female rider seated astride on its back, fully seated in the saddle, legs on both sides of the horse, feet in the stirrups, hands holding the reins,」
C・Bまで対応しても上手くいかない場合は、Bのプロンプトに追加する形で、本プロンプトを追加してください。
「horse and rider positioned behind the wall of fire, partially obscured by flames and smoke, horse and rider set farther back, not dominating the foreground,
retro Japanese historical game cover illustration style, dramatic sengoku-era painted poster art, rough hand-painted look, analog gouache and poster color texture, visible brushstrokes, bold painterly rendering, coarse print reproduction texture, slightly grainy paper surface, strong red-orange and deep blue color contrast, dark heavy shadows, simplified but expressive facial rendering, rugged masculine historical illustration feel, flat yet rich poster-like shading, limited-color vintage print look, slightly rough edges, old console game package art atmosphere, hand-painted war chronicle illustration, matte finish, unified painterly treatment across characters and background, no modern glossy anime rendering, no clean digital line art,
rough painterly historical poster style, visible brushstrokes, gouache and oil paint texture, textured paper, dry brush texture, slightly rough edges, hand-painted look, illustrated war chronicle poster, unpolished painterly finish, matte color finish, muted facial rendering, less glossy anime face, integrated painterly characters and background, the characters must be rendered in the same rough painterly style as the background,」
D・Cまで試しても、上手くいかない場合は、ネガティブを1度削除し、下記にネガティブを変更してみてください。 「clean anime face, polished anime rendering, smooth skin, glossy skin, shiny hair, clean line art, crisp lineart, cel shading, soft airbrush shading, modern anime illustration, overly pretty face, ultra clean character, plastic skin, high gloss, studio polish, separated character layer,clean anime face, modern anime style, glossy skin, shiny hair, clean line art, crisp digital rendering, cel shading, soft airbrush shading, ultra detailed modern illustration, polished character art, plastic texture, photorealistic, 3d render, smooth gradient, vector art, logo, typography, text, letters, watermark」
・とにかく筆舌に尽くしがたいので、どのようなイラストになるか？プロンプト構成はどのような感じか？については、イラストを投稿し、そちらで補完いたします。
・予想外の変化が楽しいと感じる場合には、面白いLoRAなのかもしれません。
・スタイルＬｏＲＡは安定した画質や構図などの生成を目的とするものであるため、その目的から、かけ離れたスタイルＬｏＲＡです。 要するに、失敗なのですが、得られた結果の知見を記録にとどめる・せっかくのクレジットがもったえない(´Д⊂ヽ。ため、公開ということにしました。
(Please read the explanation/Special conditions) Nobunaga's Ambition Package Art Style
I set out to create this style LoRA to replicate the package illustration from the Game Boy version of Nobunaga's Ambition, but it didn't quite capture that unique, rugged, and cool aesthetic.
It also differs significantly from the actual package art; at best, you might see a resemblance if someone points it out.
This style LoRA generates illustrations based on the following composition:
・A figure astride a horse ・A profile view of the figure ・A burning background and a castle ・Soldiers
If you are interested and thinking of giving it a try, I recommend reading the explanation below.
Just for the sake of proper credit (and avoiding surprises) (。-`ω-)
While the training data for the LoRA is perfectly fine, on very rare occasions—for some unknown reason—the character appears in their birthday suit (completely naked). There are no issues with the prompts, so it remains a total mystery. I can only attribute it to the AI's own peculiar tastes.
Judging by the results, the probability seems to be around 1 in 50 generations.
It doesn't happen often, but if you absolutely want to avoid it, please specify the character's clothing in the prompt or use negative prompts to forbid nudity.
The AI really is full of mysteries (´ω｀)
・The LoRA was created using "PixAI Edit Pro" with a unified dataset, yet the results are highly unpredictable. While it generally aims for the style described in the introduction, it lacks consistency.
・A 1:1 square aspect ratio is recommended.
・"Ultra" mode is recommended. It is possible to use "Standard" mode, but incorporating the settings shown in "C" below seems to make it easier to generate the type of illustration described in the "Introduction" section.
・When combining this with other LoRAs, the results can change drastically depending on the other LoRA used.
・If the output differs from the description in the introduction, adjusting the LoRA strength is key.
A. The initial trigger words used during LoRA creation are as follows: "sengoku war chronicle poster style, large side profile portrait, armored horse in the center, mounted heroine, burning castle, samurai army, blue sky, red flames, gold border, painterly gouache texture,"
B. If the initial trigger does not yield good results, remove it and try modifying the prompt as follows:
"sengoku war chronicle poster style, painterly gouache texture, gold border, large side profile portrait on the left, mounted heroine, samurai army, burning Japan's castle, blue sky and red flames, central horse with a natural unarmored head, simple bridle, simple reins, plain horse face, no head armor, with a female rider seated astride on its back, fully seated in the saddle, legs on both sides of the horse, feet in the stirrups, hands holding the reins,"
C. If the issue persists even after trying step B, please add the following prompt to the one used in step B. "Horse and rider positioned behind a wall of fire, partially obscured by flames and smoke; horse and rider set further back rather than dominating the foreground,
Retro Japanese historical game cover illustration style, dramatic Sengoku-era painted poster art, rough hand-painted look, analog gouache and poster color texture, visible brushstrokes, bold painterly rendering, coarse print reproduction texture, slightly grainy paper surface, strong red-orange and deep blue color contrast, dark heavy shadows, simplified yet expressive facial rendering, rugged masculine historical illustration feel, flat yet rich poster-like shading, limited-color vintage print look, slightly rough edges, old console game package art atmosphere, hand-painted war chronicle illustration, matte finish, unified painterly treatment across characters and background; no modern glossy anime rendering, no clean digital line art,
Rough painterly historical poster style, Visible brushstrokes, Gouache and oil paint texture, Textured paper, Dry brush texture, Slightly rough edges, Hand-painted look, Illustrated war chronicle poster, Unpolished painterly finish, Matte color finish, Muted facial rendering, Less glossy anime face, Integrated painterly characters and background, Characters rendered in the same rough painterly style as the background,"
If things still don't work out after trying steps D through C, please delete the current negative prompt and try using the following one instead: "Clean anime face, polished anime rendering, smooth skin, glossy skin, shiny hair, clean line art, crisp lineart, cel shading, soft airbrush shading, modern anime illustration, overly pretty face, ultra-clean character, plastic skin, high gloss, studio polish, separated character layer, clean anime face, modern anime style, glossy skin, shiny hair, clean line art, crisp digital rendering, cel shading, soft airbrush shading, ultra-detailed modern illustration, polished character art, plastic texture, photorealistic, 3D render, smooth gradient, vector art, logo, typography, text, letters, watermark"
・It is difficult to put into words, so I will post the actual illustration to demonstrate the resulting style and the structure of the prompt used.・If you enjoy unexpected variations, you might find this LoRA interesting.
・Style LoRAs are typically designed to ensure consistent image quality and composition; however, this LoRA deviates significantly from that purpose. In short, it is a failure. Nevertheless, I decided to release it to document the insights gained and because I didn't want to waste the compute credits I had already spent.
