當AI遇見創作,會碰撞出怎樣的火花?

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"你一定見到過抖音有趣的互動特效,比如曾經的“甩狗頭”、“控雨術”,如今的地標 AR 打卡等。貼合人體輪廓,準確識別特徵,即時響應動作……這些技術能力你可能不會注意,但你一定對絲滑的體驗和豐富有趣的玩法有所感受。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"基於廣泛的影響力,抖音背後的技術能力吸引了不少關注。AI 算法,在抖音背後的智能互動特效和智能視頻編輯中起到了非常重要的作用。如何擁有抖音同款能力?如何能夠讓音視頻產品能力促進業務增長?"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"抖音背後的技術支持平臺火山引擎,在金秋 9 月舉辦系列增長沙龍,先後落地上海、深圳和北京,從技術、產品、應用和體驗四大維度,剖析“智能互動特效和智能視頻編輯”推動泛娛樂行業業務增長的邏輯。在北京站沙龍現場,我們看到了抖音產品邏輯、生態建設和技術能力的冰山一角。"}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"不是再造一個“抖音”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2021 年 6 月,"},{"type":"link","attrs":{"href":"https:\/\/www.infoq.cn\/article\/x16KkP5p1pepKk7tkAlB","title":"xxx","type":null},"content":[{"type":"text","text":"火山引擎"}]},{"type":"text","text":"在首次品牌發佈會上,宣佈將字節跳動積累的推薦算法、數據分析和人工智能等核心技術,通過火山引擎開放給企業客戶。抖音同款技術能力的吸引力,引發業內熱議。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/aa\/aa41ed592384871cb9339a1a053dfccc.webp","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在火山引擎增長沙龍北京站,火山引擎 AI 解決方案負責人駱怡航表示:火山引擎已經開放的支撐抖音的技術能力,並不是幫企業再造一個抖音,而是希望讓企業通過運用抖音同款技術能力,在自己的企業應用中搭建互動場景。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"目前,火山引擎作爲字節跳動的企業級技術服務平臺,AI 產品線以 AI 中臺爲底座,提供包括視覺、語音等八項基礎能力,支撐上層智能體驗套件,在產品層提出面向各行業的音視頻解決方案,已經對外推出了 AI 大數據和視頻雲等產品,服務於金融證券、手機、汽車等行業。在內容創作方面,抖音和剪映已經成爲上述技術能力名副其實的“代言人”。直播和短視頻顛覆了內容創作的輸入和輸出方式,不止在泛娛樂行業,提高內容生產者和消費者身份轉換頻率,爲行業提供優質的內容呈現方式。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在分享中,駱怡航多次提到了生態建設,他認爲,提供技術產品僅僅搭建了當前場景,一個強大的生態可以持續激發產品創新。建立生態,滋養創意孵化。相比再“造”一個抖音,根據不同行業和具體的場景,建設因地制宜的內容生態,更有利於抖音同款能力展現最大化。"}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"“抖音式”增長"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"那麼如何擁有抖音同款能力?火山引擎提供了工具——智能創作套件。火山引擎智能互動特效總監範青談道:“我們目前看到的兩個最重要的視頻生產方式就是直播和短視頻,讓消費者更容易進行開播,更容易進行內容的生產,是我們現在的產品需要做的事情,我們叫做智能創作的套件。” "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/e4\/e4fc6949d6992c182acb89a950276d7c.webp","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"產品迭代跟着用戶需求走,音視頻的玩法隨之多樣。以美顏產品爲例,審美趨勢的改變讓美顏產品的功能越來越細,超過 40 個調整維度反而讓用戶的使用門檻提高。爲了讓用戶更容易上手,抖音產品開始調整產品方向爲自動化優化,包括部分特徵的保留和 AI 算法下的畫質清晰穩定。如今,這些功能在智能創作套件上都有所體現。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"火山引擎智能互動特效總監範青提出了深耕場景、數據驅動和落地爲先,這意味着抖音互動特效需要做好場景適配、工程優化和特殊場景的落地,代表着抖音背後的技術積累,如近、中、遠距離場景的算法靈敏度、低中高芯片的適配等。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在 AI 算法上,火山引擎做到了算法適配、工程優化和場景落地,其中,算法方面會針對近中遠距離、全身半身、橫豎屏、室內室外光照、實時非實時分別適配;工程優化方面會適配中低端芯片,提升機型覆蓋率,移動端大屏主機端,平臺模型差異化;場景落地方面,會基於場景解決垂直問題,如電商試穿試戴穩定性的關聯等。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在特效引擎方面,火山引擎技術能力可以實現對 App、小程序、瀏覽器等多平臺的支持,提供 CG 級特效玩法加速落地移動終端、GAN 類特效提升生產週期,以及持續發力攻克低端機型瓶頸。目前積累的智能特效包括"},{"type":"link","attrs":{"href":"https:\/\/www.infoq.cn\/article\/GCGIboPIfTpBe9dEqF3m","title":"xxx","type":null},"content":[{"type":"text","text":" GAN"}]},{"type":"text","text":"、美妝、美顏、美體、貼紙特效、圖像處理、虛擬形象等。在特效內容方面,火山引擎提供統一的特效素材平臺管理、線上爆款監控以及商業務拓展收入空間等服務,通過持續上新特效玩法和更新工具,幫助內容生產者提高生產效率,進而完善內容生態。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“一個生態的形成,生產者和消費者之間的平衡、生產者的活躍度是內容生態最重要的部分。”範青表示,推薦算法落地的土壤是足夠豐富的內容,只依賴 PGC 撐起內容豐富度有很大瓶頸。直播和短視頻相較於圖文,可以在單位時間內帶來更多信息量,也更容易提高信息的被吸收程度。無論是 UGC 用戶還是 PGC 用戶,都有可能隨時轉換身份。“抖音式”增長就在用戶身份在生產者和消費者之間轉化的時候發生。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/e2\/e25fd4296ec3690c3f333e8df2ca2dbb.webp","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"值得注意的是,版權問題是音視頻行業如今廣受關注的話題。在內容生產的鏈路中,內容形式有文字、圖片和視頻等,要求平臺解決相應的版權問題。火山引擎智能視頻編輯解決方案總監郭灃儀表示,AI 技術在內容層面提升創作效率的能力,火山引擎在這個層面解決了所有內容端的能力建設,也在持續解決版權問題。 "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/57\/575566eed80f6d7c6d3585ba76db53d1.webp","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"創維高級產品經理劉熙桐帶來了火山引擎技術能力在創維的實踐經驗。她談到,大屏電視行業已經步入了存量市場爭奪的時代,中國電視行業已經從早期的基於硬件做性能優化,過渡到以內容生態和用戶體驗爲核心的強交互時期。用戶需求的大屏電視不只是電視基礎功能,而是希望能夠給生活帶來更多改變。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在劉熙桐看來,當前大屏電視行業面臨硬件趨同性嚴重、場景少、玩法少、交互體驗有待進一步加強的現狀,創維的應對方式是:多模態的人機交互,支持體感、手感、語音等交互方式;更多樣的應用服務,藉助人工智能與雲計算等技術,實現大屏遊戲、互動健身、視頻通話、遠程會議等服務。“數據集是騙不了人的”。在她看來,火山引擎在內容層面和技術層面的積累上有較大優勢,工程優化效果和算法穩定性比較高。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“同樣是屏幕,手機上能實現的,電視大屏也能實現。”劉熙桐表示。"}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"非線形編輯器改變音視頻創作方式"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"音視頻智能編輯方式深入到生活中的方方面面,給人一種如今便捷的編輯方式的存在向來如此的錯覺。事實並非如此。在傳統的電影膠片存儲音視頻時代,當需要對內容進行編輯的時候,必須將膠片剪開再拼接上,才能完成剪輯工作。這種剪輯方式對原始膠片的改變是破壞性的,對使用者而言也十分不便。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"這種編輯方式被稱爲線性編輯。如今,火山引擎提供的智能視頻編輯方式爲"},{"type":"link","attrs":{"href":"https:\/\/www.infoq.cn\/article\/bYg6rh1uED3rV9Kd5RTd","title":"xxx","type":null},"content":[{"type":"text","text":"非線性編輯"}]},{"type":"text","text":",可以實現多軌道、多端、協同進行視頻編輯。所謂非線形編輯,指的是隨着技術的發展,數字媒體時代的數字化存儲音視頻方式,通過計算機或者 App 隨時可以對數字媒體素材進行剪輯工作,不會對原始資源產生破壞性改變。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/ee\/ee2fa0213bcc279e759e3fbd4585ac96.webp","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"字節跳動非線性編輯器(Non-linear edit)簡稱 NLE,是字節跳動團隊推出的音視頻編輯中間件,爲音視頻創作產品提供更加便捷的操作 API 和統一的草稿數據格式,在集成剪輯原子能力的基礎上提供操作記錄恢復等能力。據火山引擎智能視頻編輯高級研發經理 Heaven 介紹,基於 NLE,用戶可以方便地進行多端、多產品音視頻作品二次創作和協同創作。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"據他介紹,字節跳動非線形編輯器的優勢在於:"}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"核心引擎經過億級 DAU 產品驗證,性能、穩定性可靠"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"數據驅動模型設計,更輕量級接口調用,業務實現不費力"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"原生撤銷 \/ 重做 (redo\/undo) 能力支持,省去業務方實現麻煩"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"可擴展草稿協議,跨平臺存儲 \/ 恢復,輕鬆實現多端、多業務草稿打通和功能升級"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Heaven 稱,只要掌握 NLEModel、NLETrack、NLETrackSlot、NLESegment、NLEResourceNode 這 5 種數據模型,就可以通過不同組合,打造不同的複雜場景。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/1e\/1e52597b18cb7fe2ff446ab9713da7e9.webp","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“NLEModel 可以理解爲最外層的容器,我們在這個容器下進行一些模型的組合,在 Model 裏面可以添加不同的軌道,也就是 NLETrack。在 Track 之內又可以添加視頻片斷,這個片斷就是 TrackSlot。一條軌道是一個時間軸,不同的時間片斷之內可能會承載不同的內容。NLETrackSlot 定義這個時間軸上的時間片斷,比如 0 到 5 秒鐘,通過 NLEsegment 和 Resoun 定義承載的信息。”他進一步解釋。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"傳統的數據驅動模型,以事件或者命令驅動,通常先定好策略和接口功能的設定,調用方通過事件和命令進行調用,在現在來看就比較繁瑣,也不利於拓展。因爲一旦涉及增加功能,就會涉及很多改動。Heaven 談到,基於這種考慮,火山引擎制定了一種數據驅動模型範式,不再定義功能,而是通過定義五種數據模型來進行組合和配置的使用,通過業務方向進行排列組合進而實現不同的功能,不僅解決了拓展性的問題,還解決的存儲的問題,可以更方便的實現 redo\/undo 的功能。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Heaven 談到,對數據模型的封裝,最終目的是爲了簡化字節跳動非線性編輯器 SDK 的使用方式,儘管目前已經做了很大程度上的簡化,但真正實現複雜的編輯器,還是需要做非常多的工作。“我們下一步或者正在做的,是對更上層的業務組件進行封裝,比如軌道編輯器,你可以直接拿到組件集成到你的產品裏。我們會提供一些 UI 定製化的能力,針對不同客戶的訴求提供不同層級的接入能力,還會結合 AI 的雲服務和雲渲染等比較高級的能力,提供更加智能化的創作能力,以及精品內容的生產消費能力。”Heaven 稱,構建完整的服務體系,真正提供一站式解決方案,是火山引擎下一步想要實現的事情。"}]}]}
發表評論
所有評論
還沒有人評論,想成為第一個評論的人麼? 請在上方評論欄輸入並且點擊發布.
相關文章