一篇 AI 虛擬採訪的協作流程分享，使用 Nano Banana Pro 產出照片得到的清醒夢閱讀體驗 Sharing an AI Virtual Interview Workflow: The Lucid Dream Reading Experience with Nano Banana Pro Photos
自從去年 Google 發表 Nano Banana Pro 之後，我跟 萌朧動漫情報網 的朋友合作撰寫了一篇虛構的採訪報導。這篇採訪文主要是基於過往的採訪經驗，並將活動細節與人物設定重新揉合撰寫而成。相關的成果可以參考這裡：
Since Google announced Nano Banana Pro last year, I collaborated with friends from Moelong News to write a fictional interview report. This article was primarily written by blending past interview experiences with reimagined details of events and character settings. You can check out the result here:
【漫博2277】特派現場：燃燒的靈魂與 QR Code 的階級戰：2277 國際動漫祭首日實錄【Nano Banana Pro 獨家產出】
從文字到真實的「明晰夢」
From Text to Realistic “Lucid Dreams”
其實在看到 Nano Banana Pro 能夠產生出非常逼真的照片之後，我就開始構思還能有什麼樣的應用方式。尤其是它能特別準確地呈現 中文字，這點非常實用，你可以做出一張又一張看似真實的「清醒夢」（Lucid Dream）般的場景。
Actually, after seeing that Nano Banana Pro could generate extremely realistic photos, I started conceiving what other application methods there could be. Especially its ability to accurately render Chinese text, which is very practical. You can create scene after scene that looks like a realistic “Lucid Dream”.
全語音輸入的協作體驗
Full Voice Input Collaboration Experience
這篇文章也是我第一次全程使用麥克風，透過語音輸入突破打字速度的限制，讓 AI 能跟上我的思緒產出，這個方法非常有效。後來也看到許多 AI 語音輸入的工具，我相信未來純語音介面與 AI 互動的準確度應該會提升到驚人的地步，也許到了那一天，我們就不再需要虛擬鍵盤了。
This article also marks my first time using a microphone exclusively for the entire process, utilizing voice input to break through typing speed limits so that the AI could keep up with my train of thought. This method proved to be very effective. I have since seen many AI voice input tools, and I believe the accuracy of pure voice interfaces interacting with AI will reach amazing levels in the future. Perhaps one day, we will no longer need virtual keyboards.
上下文與真實感的構建
Building Context and Realism
要產出更為真實的照片，就必須提供更多的 Context（情境脈絡）。因此這次我採用的方法是以口述方式描述我想要的故事與情境，再請 AI 幫我生成這個角色應有的人物設定。只要提供職業、性別、大致年齡、人物風格以及場景，它就能產出與描述十分貼切的人物形象。
To produce more realistic photos, you must provide more Context. Therefore, the method I adopted this time was to describe the story and scenario verbally, and then ask the AI to help me generate the character settings that this role should have. As long as you provide the profession, gender, approximate age, character style, and scene, it can produce a character image that matches the description very closely.
這次體驗讓我深刻感受到大模型在文字聯想上的強大實力，彷彿只要不斷地進行連結，真的連猴子也能敲出整本莎士比亞。
This experience gave me a profound sense of the powerful capabilities of large models in text association. It seems that as long as you keep making connections, even a monkey could eventually type out the entire works of Shakespeare.
開發之 WordPress 寫稿 AI 工具與應用實務
Development of WordPress AI Writing Tool and Practical Application
也因為想要開發工具加速寫稿與產稿流程，於 WordPress 系統中，透過專用寫稿外掛（Plugin）進行文章協作。 該外掛主要功能為輔助撰稿。在提供足夠的文字素材與照片的前提下，能有效縮短初期編寫時間並協助生成文章初稿。
Also, because I wanted to develop tools to speed up the writing and production process, I used a dedicated writing plugin in the WordPress system to collaborate on articles. The plugin’s primary function is to assist in drafting. Given sufficient text materials and photos, it can effectively shorten the initial writing time and help generate the first draft of the article.
結語
Conclusion
當然，並非所有細節都能完美預測，因為目前生成的照片仍帶有一種過於工整、匠氣的風格，或者說像是舞台劇或產品照的感覺。不過，利用這些照片來創造令人信服的場景已不再遙不可及。產出這整篇文章與配圖，大約只花了四、五個小時。老實說，我相信如果準備了更多背景資訊以及預先設定好的場景，要建立一條量產流水線絕非不可能，這也是在熟悉如何與 AI 協作的過程當中，特別有意思的體悟。
Of course, not every detail can be perfectly predicted, as the photos currently generated still have a style that is too neat and “manufactured,” or feel like stage plays or product shots. However, using these photos to create convincing scenes is no longer out of reach. Producing this entire article and its images took only about four to five hours. Honestly, I believe that if one prepared more background information and pre-set scenes, establishing a mass production line is definitely possible. This is a particularly interesting insight gained while learning how to collaborate with AI.