生成式AI近年發展快速，其中對寫小說，卻沒辦法自己畫插圖的伍德來說，影響最大的莫過於AI生圖了。儘管在內容庫上的產權爭議仍待釐清，但若擱置爭議，這幾年的進步確實是肉眼可見。例如初期動不動手指數目不對的問題出現率在各家引擎中就有顯著下降。而各家引擎擅場及風格不同，成品的筆觸也有點差異。今天就來跟各位一起稍稍看看伍德現在兩個比較常用的引擎：Bing及ImageFX的成品差異。
AI生圖剛發展時，一度因為調參數、套濾鏡需要各種技巧，甚至想產出張能看的圖片還要一些程式語言的概念，而讓「詠唱師」一詞一時蔚為風潮，市面上也有不少書籍。不過隨著技術進步，儘管詠唱師和各種技術仍舊發展中，對一般人來說，支援「講人話」(自然語言)，又能生出不錯圖的AI生圖系統已經不少了*1。
免費的引擎中，微軟的Bing是發展相對早的(點我)。他搭載的是Dall E-3，原則上說人話就能產出不錯的圖片。而ImageFX則是Google旗下的團隊，目前並未開放台港IP使用，如果需要還是得先用VPN翻牆，而且對英文也比較敏感。那麼接下來就以拙作《魔都妖探》幾個畫過插圖的場景來比對一下兩者的表現*2。前者為Bing、後者則為ImageFX。
1. Japanese Anime Style. A man and a woman walking along the beach at night under starry sky. The woman has long black hair. She is singing and clasping her hands. The man has black short hair. The man is wearing a blue jacket, white t-shirt, black trousers. The man is smiling, looking at and walking after the girl. The girl is wearing a pink t-shirt, beige coat, green long skirt and white high heel sandals.
(Case 4-9的經典(?)海灘場面)
(這張Bing記得是沒有指定衣服到原文這麼細生成的，所以衣服有差請見諒。)
2.
Bing: An anime couple in their 20s and a white cat. The background is a coffee shop. Both of them are smiling with confidence. They are enjoying coffee and cake around the table. The man has black short hair, while the girl has brown hair with ponytail. The man is wearing a casual shirt, jacket and slim-fit jeans.
ImageFX: Japanese Anime Style. One man and one woman seating in a coffee shop. They are sitting next to the window. There are two cups of coffee on the table. The handsome man has black layered short hair and turning his head, looking at the camera. He is wearing blue jacket, a belt, black trousers and black sneakers. The woman has brown hair with ponytail. She is playing with a white cat. She is wearing a white cardigan sweater, pink t-shirt and blue jeans. They are happy and grinning.
(Case 2-10咖啡廳景)
3.
Bing：An anime couple in their 20s. They are shy, standing and looking at each other in a shark tunnel. The man has blonde short hair. The man wears a red casual shirt and black slim-fit long pants. The girl has white long hair. The girl is wearing a beige coat. A dolphin is looking at them in the water.
ImageFX：Japanese Anime Style. One man and one woman talking happily in a tunnel in an aquarium. The man has blonde layered short hair. He is wearing red jacket, a belt, black trousers and red sneakers. The woman has silver long hair. She is wearing an apricot blouse and blue jeans. They are smiling. There are lots of tropical fish in the tunnel.
(Break 4-4的水族館景)
4.
Bing：An anime elegant thief in his 20s. The background is an open window in an art gallery. He is smiling with confidence and stepping on the window frame. Outside the window is night sky. The man has silver short hair and red eyes. The man is wearing a silver suit, red tie, white gloves and long pants.
ImageFX：Japanese Anime Style. An elegant thief in his 20s. The background is an open window in an art gallery. He is smiling mysteriously and stepping on the window frame. Outside the window is night sky. The man has silver layered short hair and crystal blue eyes. The man is wearing a silver suit, red tie, white gloves and long pants.
(Case 3-5)
5.
An anime couple in their 20s in a library in the afternoon. Both of them are standing. The man is sitting at a desk and focus on reading big newspaper. There are lots of newspaper on the desk. The man has short blonde hair. The man wears bright red jacket and black pants. He looks confused. The girl has long silver hair. The girl is behind the man. She wears a shirt and skirt. She is smiling and looks interested in what the man is reading.
(Case 8-10金毛和音奈在圖書館支援查案件資料)
(這張的錦懋的確有種「這寫什麼文字，我怎麼看不懂」的感覺(X))
6.
Bing： A slim anime university student with short blonde hair fighting with another slim man with black short hair in the garden of a hospital. The man with blonde hair is angry and wearing a red sports jacket and black slim-fit long pants. The man with black hair is wearing a blue jacket, slim-fit jeans, black socks and sneakers.
ImageFX：Japanese Anime Style. A university student with short blonde hair fighting with another young man with black short hair in the garden of a hospital. Both of them are angry and arguing. The man with blonde hair is wearing a red sports jacket and black slim-fit long pants. The man with black hair is wearing a blue jacket, slim-fit jeans and sneakers.
(Break 6-3；賀輔和錦懋的互毆)
7.
Japanese Anime Style. A couple in their 20s talking face-to-face before a campus building at a sunny noon. The sky is blue. The man has blonde short straight hair, and is wearing a t-shirt, green jacket and slim-fit black pants. The girl has brown curly long hair. The man is feeling awkward and reaching his hands out. The girl is wearing peach cardigan sweater and gray short skirt. The girl is sobbing helplessly.
(Case 9-1，音奈視角的世界名畫(X))
基本上可以看到Bing的風格還是沒有完全統一、比較多樣。當然也可能跟伍德使用Bing的時間比較早，在下指令時還不是那麼精準。不過其實Bing在構圖上是比較有創意的(這裡都是伍德給過而使用或在微修的，沒有很那些品質不行、但啟發伍德改進指令的成品)。
另一方面，ImageFX的日系動畫風格非常有2005年附近、進入電繪時代一陣子的動畫作品的味道，和Bing相比細緻非常多。有幾張甚至伍德都覺得印在L夾上拿去文具店便宜賣會有人買單(欸)。不過ImageFX在構圖上比較完全遵照文字，需要使用者自己本來就有想法。而且ImageFX最讓伍德頭痛的是，在髮型上不分男女蠻沒層次的。特別黑短髮男性會有少年化的傾向(儘管伍德需要的是青年)。
兩者基本上各有擅場，由於生成圖數沒有限制，甚至可以稍微和AI摸索一下，找出最適合的衣著和構圖，再下更精確的指令。
最後用伍德用ImageFX骰出一張蠻滿意的圖片結尾。
感覺是金毛和音奈昨晚同居後，因為太累而差點趕不上課堂的感覺。你說在累什麼？伍德不知道(欸)
那麼這次就聊到這裡，以後伍德的小說應該也會時不時放些AI插圖。當然，伍德的終極目標還是有位能互相信任的繪師朋友來做最為細緻的調整呢。
*1. 當然，免費系統在尺度上的限制就比較多。不過伍德也只是生一般向插圖，倒是沒什麼影響。
*2. 目前(2025/3)在小說中的圖部分經過PixAI及Stable Diffusion再加工，不過伍德現在覺得直接用產出的圖說不定比較好一些。
