2023/10/25 by 李宏偉  Spoon H.-W. Lee

我在數位人文網站裡的某篇文章，提到只要一張照片就可以利用 Headshot 生成臉孔模型。目前 Headshot 已經進化到第二代，微調的功能更多、展現出來的效果也更擬真。最近因為要拍「數位人文」課程的微影片，有感於自己身材走樣離譜，深怕危及觀眾的視力健康，因此突發奇想要加碼利用 Character Creator (CC) 來創造出幾個我的「阿凡達」( avatar 替身)，讓他們在微影片中替我拋頭露面。
In one of my articles on a digital humanities website, I mentioned that a single photo is all it takes to generate a facial model using Headshot. Currently, Headshot has evolved into its second generation, offering even more refined features and significantly more realistic results. Recently, while preparing short videos for my "Digital Humanities" course, I became increasingly aware of how out-of-shape I have become, seriously worrying that my appearance might harm the visual health of my audience. This prompted me to come up with a whimsical idea: to go a step further and use Character Creator (CC) to create several "avatars" of myself, letting them take my place and appear on camera in the videos.

其實，我們實驗室使用甲尚公司的產品已經好有幾年時間，但是我向來秉持要和小龜兩個人專業知識和技能應該稍有重疊但需彼此互補為宜的原則，所以人物和場景的創建以及故事和動畫的設計，向來都是她的業務與專長。因此，不管是 iClone 也好、CC 也好，對我來說都是很新鮮的嘗試。
Actually, our lab has been using Reallusion's products for several years now. However, I've always believed that my professional expertise and skills should partially overlap yet complement those of Little Turtle. Therefore, tasks like creating characters and scenes, as well as designing stories and animations, have always been her responsibility and area of specialization. Consequently, whether it’s iClone or CC, it remains a fresh and novel experience for me. 

前幾天原本只想小小暖身一下，沒想到越玩越上癮，做出了10個左右的我的替身，有原民風、台客風、嬉皮風、運動風、歐巴風、牛仔風、復古風和英倫風，相當有趣。在製作的當下，不僅要思考如何搭配穿著 (與我熟識的人都知道，現實生活中我對這件事情是最不計較的)，也逐漸能夠順暢地使用各種功能。以後在相關的教學活動上，或許可以讓學生從設計自己的替身開始，作為學習這套 APP 的有趣開端。
A few days ago, I initially intended just to warm up a little bit, but I quickly found myself completely addicted. Before I knew it, I had created around ten different avatars of myself—with styles ranging from Indigenous, Taike (local Taiwanese), Hippie, Sports, Korean Oppa, Cowboy, Vintage, to British. It turned out to be quite entertaining. During the creation process, I had to think about outfit combinations (which, as people familiar with me know, is something I pay the least attention to in real life). Gradually, I became more comfortable and fluent with the various features in the software. Perhaps in future teaching activities, I could have my students start by designing their own avatars, providing a fun entry point for learning how to use this app.

接下來，就等著讓這些在多重宇宙中出現的我(們)，在數位人文課程的微影片中活靈活現起來囉！
Now, all that's left is to let these multiverse versions of myself come alive and shine in the short videos for the Digital Humanities course! 

2023/7/23 by 李宏偉 Spoon H.-W. Lee

在原住民族知識與文化的保存工作中，除了語言與文字之外，視覺化材料的呈現也是我們積極努力的目標。因此，除了利用 3D 掃描或建模將古物、器皿或甚至建築保留在元宇宙之外，黑白照片的修復和著色也是值得努力的方向。
In our work to preserve indigenous knowledge and culture, apart from language and written texts, we are also actively striving toward the presentation of visual materials. Therefore, in addition to using 3D scanning or modeling to digitally preserve artifacts, utensils, and even architecture in the metaverse, the restoration and colorization of historical black-and-white photographs is another worthwhile direction for our efforts. 

自上學期起我便嘗試帶領學生利用 Photoshop 的 Neural Filter 來做原住民老照片的 AI 著色，但是所得結果總覺得僅僅差強人意。昨晚失眠，一時興起利用 DeOldify 這隻自動著色引擎寫了個 Python 程式，然後又 Google 了一下看看有沒有很厲害的 AI 著色引擎，結果讓我找到 HitPaw 的 Photo Enhancer，不過它不必自己寫 code，只要下載安裝即可，但是得花錢才能匯出結果。
Since last semester, I've been guiding my students to use Photoshop's Neural Filter for AI-based colorization of old indigenous photos, but the results have always been merely satisfactory at best. Last night, being unable to sleep, I spontaneously wrote a Python script using DeOldify, an automatic colorization engine. Then, I Googled around for more powerful AI colorization tools, which led me to HitPaw's Photo Enhancer. Unlike DeOldify, HitPaw doesn't require any coding—just download and install—but exporting results requires a paid subscription.

比較這三個 AI 引擎的效果，就著色方面的表現我覺得DeOldify 和 HitPaw  差不多，但是後者有個「人臉專用模型」，既可處理著色問題又能柔和臉孔質地，所以整體效果會更好一些。至於 Photoshop 的表現確實略遜一籌，不過它的另一個優勢是可以手動上色，所以最右邊那張圖看起來顏色會更豐富，但那是自己再加顏色的結果。
Comparing the results from these three AI engines, I feel that DeOldify and HitPaw are similar in terms of colorization performance. However, HitPaw features a "face-specific model" that not only handles the coloring but also softens facial textures, resulting in a slightly better overall effect. As for Photoshop, its performance is indeed somewhat inferior; however, its advantage lies in allowing manual coloring. Thus, the image on the far right appears richer in color, but that’s the result of additional manual adjustments.

以當前情勢來看，在全面 AI 化之前，應該還會有一段時間處於「半自動武器」優於「全自動武器」的階段。但是我想，針對 AI 自動著色對台灣原住民族老照片表現總是較差的問題，原因可能在於它們的預訓練模型沒有見過幾張泰雅族或賽夏族的「彩色照片」。話說回來，這會有雞生蛋、蛋生雞的問題：如果沒有彩色照片給 AI 看，AI 就生不出彩色照片！這時候，依照現有實體物件（譬如服飾）進行人工著色，或者請教耆老憑其記憶修復照片，或許都是可行解法，但是也都刻不容緩。
From the current situation, it appears that before full AI automation is realized, we'll likely remain for some time at a stage where "semi-automatic weapons" outperform "fully automatic weapons." Regarding the consistently poor performance of AI colorization on old photos of Taiwanese indigenous peoples, I suspect the reason might be that their pre-trained models have hardly ever seen color photographs of Atayal or Saisiyat individuals. This brings up a classic chicken-and-egg dilemma: Without existing color photographs for AI training, the AI simply can't produce colorized photos! Under these circumstances, manually colorizing based on existing physical artifacts (such as traditional clothing), or consulting elders to restore images based on their memories, might both be viable solutions—solutions that are urgent and can't afford delay.

這方面的 AIGC 和純粹創作取向的 AIGC 不同，因為我們要做的是保存，所以需要考究，而這也正是我和學生們必須加碼厚植原住民族文化底蘊的根本原因。
This kind of AIGC differs from purely creative-oriented AIGC, because what we're doing is preservation—it requires accuracy and careful research. And that is precisely why my students and I must further deepen our understanding of indigenous cultures as a foundational effort.

2023/7/18 by 李宏偉  Spoon H.-W. Lee

我原本在「玄奘大學數位人文計畫」網站裡的「新知分享」就已經貼過一則說明 Adobe Firefly 的文章，而且最近也持續在玩這隻 AIGC。其實 AIGC 被翻譯成「生成式 AI」稍有偏差，因為這個中文字眼跟 Generative AI 比較貼近，而 AIGC 是 AI generated contents 的縮寫，所以應該是「AI 所生成的內容」才對。也就是說 AIGC 指的是這張圖，而不是長出這張圖的 AI 引擎。
I had already posted an article explaining Adobe Firefly in the “Knowledge Sharing” section of the “Digital Humanities Project” website at Hsuan Chuang University, and I’ve continued experimenting with this AIGC tool recently. In fact, translating AIGC as “Generative AI” is somewhat inaccurate, since that term corresponds more closely to “Generative AI” itself. AIGC is short for “AI-generated contents,” so it should be understood as “content generated by AI.” In other words, AIGC refers to the image itself, not the AI engine that created it.

不過，就算不這麼錙銖必較也無所謂，因為手邊有工具能用、好用就好，所以最近幾天也持續在嘗試開發 Adobe Firefly 的潛能。我發現，簡單地改掉關鍵字就會得到非常不同的結果。譬如我把 abstract style of aboriginal people (V+ing) in a forest 當中的 abstract style 改成 Cubism style，真的就長出很有畢卡索風味的立體派畫作。
That said, it’s not really a big deal even if we don’t get too caught up in the terminology—what matters is having a tool that works and is easy to use. So over the past few days, I’ve continued exploring the potential of Adobe Firefly. I discovered that simply changing a keyword can produce drastically different results. For example, when I replaced “abstract style” with “Cubism style” in the prompt abstract style of aboriginal people (V+ing) in a forest, the result turned out to be a Cubist painting with a distinct Picasso-like flavor.

真的很神奇，很好玩！於是我就又動手建置了另一個「玄奘大學 USR 泰雅爾巴來」的網站。而且相同地，一切都在 24 小時之內完成。
It’s truly amazing and a lot of fun! So, I went ahead and built another website—this time for the “Hsuan Chuang University USR Atayal Balay” project. And once again, everything was completed within 24 hours.

2023/7/15 by 張鈺潔  Julia Y.-C. Chang

在 Adobe Firefly (Beta) 裡面有許多不同創作模組，可供生成完整圖片，我選擇「生成填色」來繪製與修改圖片。這張照片想要表達的是「運動健康美少女保護森林」的一種意象，很像日劇「哪啊哪啊~神去村」招募實習生的宣傳照片。
In Adobe Firefly (Beta), there are various creative modules available for generating complete images. I chose the "Generative Fill" feature to create and modify the image. This photo is meant to convey the concept of a "sporty and healthy young woman protecting the forest"—an image reminiscent of the promotional photos used to recruit interns in the Japanese drama Wood Job! (What a Wonderful Forest!).

我先把原本圖片去背，使用 text to image 功能輸入 forest，這支 AIGC 就自動把主角送至森林裡。想說也許可以幫這個女生別個胸針，我就用筆刷刷掉一塊範圍，一樣使用 text to image 功能輸入 butterfly pin，這支 AIGC 就自動在主角的衣服上面別上蝴蝶別針。這女孩兒的肩膀上應該可以帶一隻小動物，一樣用筆刷刷掉一塊區域，使用 text to image 功能輸入 rabbit，這支 AIGC 就自動放個小動物在女主角肩膀上。弄著弄著，一張還算不錯的宣傳照片就出來了。
First, I removed the background from the original image and used the Text to Image feature to input “forest.” The AIGC automatically placed the main character into a forest setting. I thought it might be nice to add a brooch to the girl’s outfit, so I brushed out a small area and typed in “butterfly pin”—the AIGC instantly added a butterfly brooch to her clothes. Then I figured a small animal on her shoulder might be cute, so I brushed out another area and entered “rabbit”—sure enough, the AIGC placed a little rabbit on her shoulder. And just like that, I ended up with a pretty decent promotional image.

看著這位美少女，心想... 會有人看到這張照片，便想要像「哪啊哪啊~神去村」裡面的「勇氣 (劇中男主角)」一樣走進山裡，守護著這片山林嗎？
Looking at this beautiful young woman, I couldn’t help but wonder…  Will someone see this photo and, like Yūki—the protagonist in *Wood Job!*—be inspired to head into the mountains and protect the forest?

2023/7/13 by 張鈺潔  Julia Y.-C. Chang

這個 Spark AR 濾鏡的設計概念是體驗泰雅族的紋面文化。透過自身臉孔的紋面體驗，像是親身經歷一般，喚起人們對於逐漸失落文化的一個角落的關注；更期待的是，喚起對於原民族的認同與尊重。
The design concept behind this Spark AR filter is to offer an experience of Atayal facial tattoo culture. By allowing users to virtually wear traditional facial tattoos on their own faces—as if experiencing it firsthand—the filter aims to raise awareness of a fading cultural tradition. More importantly, it seeks to inspire recognition of and respect for indigenous identity.

這個濾鏡使用的技術是 face tracking，只需要開啟 Facebook或者是 instagram 相機，並且拍向自己，就可以體驗紋面那種原始自然的感受。
This filter uses face tracking technology. By simply opening the Facebook or Instagram camera and pointing it at yourself, you can experience the raw, natural feeling of traditional facial tattoos.

看著照片中紋上泰雅女性紋面的自己，總覺得像是在提醒著自己，成年後該有的成熟與穩重，好像如此才配得這份紋面。
Looking at myself in the photo, adorned with the traditional Atayal women’s facial tattoo, I can't help but feel as if it's a gentle reminder of the maturity and steadiness expected in adulthood—as if only with such qualities can one be truly worthy of bearing this tattoo.

濾鏡連結網址 (已失效)
Filter Link URL (Expired)

• https://www.instagram.com/ar/642505654596985/

• https://www.facebook.com/fbcameraeffects/tryit/642505654596985/

• https://www.instagram.com/ar/642505654596985/?src=vc

• https://www.messenger.com/ar/642505654596985/

2023/7/13 by 李宏偉 Spoon H.-W. Lee

這項計畫是以玄奘大學應用認知科技實驗室 (Applied Cognitive Technology Laboratory, ACT Lab) 做為基本班底，而我們 ACT Lab 向來有個傳統，就是著重在「應用」。因此，點子生成和實現的重要性，遠遠高於技術面的追求和鑽研。譬如以這篇短文的主題「生成 3D 擬真頭像」為例，如果找得到快速又合用的工具，就不必浪費時間慢慢拉模或 coding。
This project is built upon the foundation of the Applied Cognitive Technology Laboratory (ACT Lab) at Hsuan Chuang University. At ACT Lab, we have a long-standing tradition of emphasizing "application". As such, the importance of idea generation and practical implementation far outweighs the pursuit of technical depth for its own sake. Take the topic of this short article—generating realistic 3D head models—as an example. If we can find tools that are both fast and effective, there’s no need to waste time manually sculpting models or writing code from scratch.

針對某個真人生出擬真 3D 頭像或全身模型的應用需求很多，除了可以讓標的對象從真實世界走入虛擬空間之外，也可以嘗試展現各種在現實生活中可能或不可能做到的變化，譬如讓年輕小夥子長出滿臉皺紋、變胖變瘦、貼上刀疤刺青，或是變身成全身藍色的阿凡達 (或藍色小精靈)。
There are many practical applications for generating realistic 3D head or full-body models of real individuals. Not only can it bring the subject from the physical world into virtual space, but it also opens up possibilities for visualizing various transformations—whether achievable in real life or not. For instance, a young man can be aged with wrinkles, made thinner or heavier, given scars or tattoos, or even transformed into a blue-skinned avatar (or a Smurf).

在這個例子裡，只要簡單兩個步驟就可以生出一個堪用的 3D 全身模型。首先，在 Character Creator 裡打開 Headshot 這個 plug-in，並且上傳一張大頭照；接著，依照指示做好兩三項設定就可以合著標的對象的臉孔自動產生一個光頭模型。如果你選用的是 Auto 模式，更可以一鍵搞定。要是看著光頭覺得怪怪的，不妨在 content 頁籤中選擇想要的髮型、衣服、配件或甚至妝容等；如果還覺得不過癮，也可以調整身型體態、臉部細節或是膚色膚質等，直到滿意為止。然後，就可以把這的 avatar 丟回 iClone，讓他在虛擬世界裡活靈活現。
In this example, generating a usable 3D full-body model takes just two simple steps. First, open the Headshot plug-in in Character Creator and upload a portrait photo. Then, follow the prompts to complete a couple of basic settings, and the software will automatically generate a bald 3D model with the subject’s facial features. If you choose the Auto mode, you can even finish the process with a single click. If the bald head looks a bit strange to you, feel free to explore the Content tab, where you can choose hairstyles, clothing, accessories, or even makeup. Still not satisfied? You can further fine-tune the body shape, facial details, skin tone, and texture until you get the result you want. After that, simply export the avatar back to iClone, where it can come to life in the virtual world.

瞧，是不是快速又簡單呢？這正好符合我們 ACT Lab 的精神：動起來，即知即行、快又有效！
See? Isn’t it quick and easy? This perfectly reflects the spirit of our ACT Lab: take action, respond instantly, and get things done—fast and effectively!

2023/7/13 by 張鈺潔  Julia Y.-C. Chang

新機器 HoloLens 購入後，帶著它出門展示是件偶爾得去做的份內事兒。那日，是個沁涼悠閒的午後，我們帶著HoloLens去教室做展示。如同先前常走的流程，短演講後，就是讓學生體驗幾個 HoloLens 既有的免費 APP。
After purchasing the new HoloLens device, taking it out for demonstrations became an occasional but expected part of the job. That day, on a cool and leisurely afternoon, we brought the HoloLens to a classroom for a demo session. Following the usual routine, we began with a brief talk, then invited the students to try out several of the built-in free apps available on the HoloLens.

因為 HoloLens 是個一次只能讓一個人體驗沉浸式感受的混合實境頭盔，所以頂多只能讓他們一個一個體驗。人少少的，這倒讓容易緊張和怕生的我放鬆許多，可以在陪同講解中觀察到一些有趣的事兒。
Since the HoloLens is a mixed reality headset that only allows one person at a time to experience immersive content, we could only let the students try it out one by one. With just a small group, it actually helped me feel more at ease—especially as someone who tends to get nervous and shy around crowds. This slower pace gave me the chance to observe some interesting things while guiding each student through the experience.

那天，我用「小松鼠吃硬幣」給他們玩。小松鼠會在現地的桌面、地板、四周，跑來竄去，玩家要完成的任務，就是想辦法讓小松鼠吃硬幣。這是一個混合實境的極致應用，因為必須辨認現地的障礙物，讓虛擬小松鼠融入在現地，與現地互動。使用者也可以 hand-free 與小松鼠互動。基本上，這是個非常困難的技術，程式撰寫上是非常耗費心力的。 其中一位學生，戴上頭盔後，馬上融入那樣的混合現實，啟發想像力，對著她的閨蜜說，想在她的肩膀上放上一個 Hologram，想在背景放上星空，跟她一起看夜空。兩人，就這樣一搭一唱玩了起來。無限想像，瀰漫在空氣中，可能性不斷被自然激發。細想，彷彿能看到他們心中的可能未來。
That day, I let them play with the “Squirrel Eats Coins” app. The little squirrel runs around the real-world environment—on desks, floors, and all around—and the player’s task is to find a way to get the squirrel to eat the coins. This is a prime example of mixed reality at its finest, as it requires the system to recognize real-world obstacles and seamlessly blend the virtual squirrel into the physical space, allowing it to interact with the surroundings. Users can also interact with the squirrel hands-free. Technically, it’s a highly complex experience—extremely demanding in terms of programming and system design. One student, upon putting on the headset, immediately immersed herself in the mixed reality. Her imagination sparked, and she turned to her best friend and said she wanted to place a hologram on her friend’s shoulder, and add a starry sky in the background so they could watch it together. The two of them started playing off each other’s ideas, co-creating a moment of magic. Imagination filled the air, and endless possibilities emerged effortlessly. As I reflected on the scene, it felt as if I could glimpse the future they envisioned in their hearts.

他們倆接觸到 HoloLens 的燦爛笑容，熱衷與人分享的那一瞬間，縈繞心頭。這樣的展示，往往得到更多的，不是這些學生，受鼓勵的反而是自己。
The radiant smiles on their faces as they experienced HoloLens for the first time—and that instant desire to share their excitement with others—lingered in my mind. In moments like these, it often feels like the one who gains the most from the demonstration isn’t the students, but me. I’m the one who ends up feeling encouraged.

(原文載於 心科技部落格 https://ipsytec.blogspot.com/)
(Originally published on the iPsy Tech Blog. https://ipsytec.blogspot.com/ )