Lolly
啟動應用程式

Lolly 的比較定位

Lolly 做到了今日創意工具做不到的事,以及它刻意留給那些工具去做的事。

若要看逐一工具的比較版本,針對 Canva、Adobe、Figma、渲染 API 與線上轉換工具各有一頁,請見Lolly 逐一工具比較。每一頁都說明了對方工具做得更好的地方,以及 Lolly 改以何種方式取代。

狀態: Lolly 已經發佈:1.0 版本於 2026 年 8 月 30 日以開源方式發佈。它的安全性尚未經過獨立稽核,採用情況的證據也只有幾個月的歷史。採用與治理頁面說明了目前的狀態。

今日的工具

下方每個環形圖評分的是某個產品類別以目前實際發佈的狀態(而非行銷宣傳)完整交付某項能力的程度 - 每個類別都以其最佳代表產品來評分。Lolly 也是用同一把尺來評分:它拿到了整張表中唯一的紅色環,原因是成熟度。點開列名稱可看到評分背後的理由。各欄依上方的整體完整度列排序 - 即已評分列的平均值,不含花費列。

Lolly includes lolly.work, the same project's separately deployed organisation service. Local creation needs no account; shared work and server rendering use the configured instance. Competitor scores: August 2026. Lolly coverage: 28 September 2026.

Capability Lolly+ lolly.work Penpotopen design CanvaAffinity
Cavalry
Adobedesktop pro Brand DAMFrontify
Bynder
Cloudinarymedia pipeline Figmaonline pro Render APIsBannerbear
Placid
Creatomate
Overall completenessunweighted mean - weight rows for your own context · columns are sorted by this row 84 70 57 48 41 41 41 34
Production maturity & track recordCanva, Adobe, Figma, Bynder/Frontify: a decade-plus at massive scale. Render APIs: years in production pipelines. Penpot: shipping, large community, younger at enterprise scale. Lolly: version 1.0 public since 30 August 2026 under MPL-2.0, and SUSE runs it as an internal deployment; no usage data is shared yet and there are no public case studies yet. Cloudinary: 2012, enterprise media infrastructure at global scale.
25 75 100 100 100 100 100 75
Mass generation from data (CSV / API)Lolly: batch grid and CLI, plus lolly.work's recoverable server jobs and batches with retries, cancellation and retained outputs. Canva: Bulk Create + Autofill API - real, but Enterprise-gated, async, text/image fields only; Affinity Publisher adds desktop data merge. Adobe: InDesign data merge and scripting. Figma: Buzz fills templates from CSV or XLSX, free in beta (Aug 2026) - cloud- and account-gated, 50. Penpot: open API/MCP, not purpose-built. Render APIs: this is their entire product - account- and cloud-gated, so 75 under the method rule. DAM: Studio-style batch create and resize. Cloudinary: URL-driven overlays and named transforms derive variants at scale - account-gated, 75.
100 50 50 50 75 75 50 75
Offline & air-gap operationLolly: standalone creation works offline once its assets are available; lolly.work's organisation services need a reachable instance, which can run on an internal network. The Canva column is scored on its best family member: Affinity (free since Oct 2025) runs offline as a desktop suite once a verified account activates it - 75, the same deduction Adobe takes. Canva's own Offline (2026) still edits pre-synced designs only, per device, in a 14-day window. Adobe: desktop apps run offline; licensing phones home. Figma: cached-file viewing and limited editing. Penpot: self-host on your infra behind a firewall (server required). Cloud APIs and DAM: none.
100 75 75 75 0 0 25 0
On-device rendering - data never leavesLolly: local rendering and conversion in-browser or CLI need no upload. Choosing lolly.work's shared storage, work collabs or server rendering sends the relevant data to the configured instance. Adobe: local desktop rendering, but cloud services and telemetry in the suite. Figma: canvas renders locally, files live in Figma's cloud. Penpot: 90 - rendering happens in the browser and the save target is a server that can be your own sovereign private cloud, even your own laptop, with private export throughout; only the server hop separates it from Lolly. Canva itself is server-side by design; the column's 75 is Affinity - local desktop rendering, account activation and telemetry, the same shape as Adobe. Render APIs, DAM: server-side by design.
100 90 75 75 0 0 25 0
Hard brand constraints (structural)Lolly: templates fix layout rules and expose permitted inputs. lolly.work adds group access, input policies, shared catalogues, approvals and audit records. Copy and usage rights can still need review. Canva: element locks - permission-based, admin-set and static: no template logic such as conditional logo switching or responsive recomposition at fill time, and Canva AI does not yet respect Brand Controls - 50. Render APIs: templates are fixed (only declared fields vary) but no governance layer. DAM: locked templates with restrictions. Figma: Buzz templates hold locked guidelines at fill time; the design surface stays review-enforced - 50. Penpot: tokens and systems enforced by review, not runtime. Adobe: libraries, little enforcement. Cloudinary: presets and Enterprise-gated transformation templates fix operations, not brand layout - 50.
100 50 50 25 75 50 50 75
C2PA provenance, signed at creationAdobe: the broadest shipped implementation (Photoshop, Lightroom, Premiere, Firefly) - signing happens locally in the desktop apps and in the cloud for the Content Authenticity web app, and nothing signs without an Adobe account and an Adobe-provisioned identity, so it is account-gated end to end: 75 under the method rule. Lolly: optional on-device signing without an account, plus lolly.work server signing with a configured organisation identity. A verifier must trust the signer's root to show a trusted identity. Version 1.0 is public and SUSE runs it internally, but no usage data is shared yet, so 75. Penpot: 50 via the official Lolly Export plugin - the same on-device engine signing Lolly uses, opt-in, as in Lolly (and disclosed plainly: it is Lolly's own plugin). No evidence of C2PA writing in Canva, Figma, render APIs or DAM as of Aug 2026. Cloudinary: 50 - signs images on delivery (fl_c2pa, CAI member since 2020), attesting delivery by Cloudinary rather than creation by you.
75 50 0 75 0 50 0 0
CLI, pipeline & AI-agent automationLolly: CLI, TUI, MCP server and URL mode, plus lolly.work's authenticated render API, recoverable batches and governed delivery. One engine across the local and server paths. Render APIs: API-first, webhooks, integrations - account- and cloud-gated, so 75 under the method rule. Penpot: open API, plugins, official MCP server. Canva: Connect API - Enterprise-gated, async, rate-limited. Adobe: UXP/ExtendScript plus Firefly APIs. Figma: strong REST/plugin API, no render pipeline. DAM: asset-management APIs. Cloudinary: API-first media pipeline - account-gated, 75.
100 75 50 50 50 75 50 75
Live collaboration (real-time co-editing)Figma: the scale benchmark - 200 simultaneous editors, up to 500 people in a file - account and cloud required. Canva: mature real-time editing across the product, same precondition. Penpot: real-time multiplayer with live cursors; self-hosting lifts the vendor-account precondition, but a server is still required. Lolly: direct two-device private collabs without an account, plus authenticated work collabs through lolly.work with shared edits, cursors and saved team sessions. Partial reflects the current single-node work deployment and limited operating history. Private collabs work without internet on a shared network; work collabs need the organisation's instance. Adobe: Live Co-Editing in beta, cloud documents only. Render APIs: no editing surface. DAM: comments and approvals, not canvas co-editing. Cloudinary: DAM comments and workflows, no canvas co-editing.
50 75 75 25 25 25 75 0
No design skill requiredCanva: the category benchmark for ease. DAM: fill a locked template. Lolly: fill in fields using an existing tool, or publish a tool from Design through Share with rules. lolly.work manages access and input policy for the team. Render APIs: swaps design skill for developer skill. Adobe, Figma, Penpot: professional tools. Cloudinary: the media library is easy; transformations are developer territory.
100 25 100 25 100 50 25 50
Open source, self-hosted, no lock-inPenpot: MPL-licensed, 51k+ stars, self-host via Docker/K8s - fully verifiable today. Lolly and lolly.work: MPL-2.0 open source in separate repositories, both self-hostable. lolly.work provides the organisation services as part of the project. The current 75 reflects the young project and limited deployment history. Everything else is proprietary SaaS or proprietary desktop.
75 100 0 0 0 0 0 0
Cost at production scaleLolly + lolly.work and Penpot: no per-seat software licence charge for the open-source self-hosted path; hardware, storage and operations remain costs. Render APIs: metered - roughly $0.005–0.049 per image, credits often expiring monthly. Canva/Figma: per-seat; Canva's automation needs Enterprise, though Affinity itself is free since Oct 2025. Adobe: premium per-seat suite. DAM: sales-led enterprise quotes, MAU or seat models. Cloudinary: credit-metered across transforms, storage and bandwidth - 25.
100 100 50 25 25 25 50 25
Annual medium enterprise spendIndicative list-price arithmetic for a medium enterprise - think 500-5000 people - August 2026: a guess range, not a quote; deals vary and the seat mix dominates. Ring = how much of the board's largest annual spend (US$500k) you keep; the number in the ring is each range's upper bound. Canva (US$75k-400k; Affinity free since Oct 2025 pulls the design-seat share down), Adobe (US$100k-500k) and Figma (US$30k-250k) assume the realistic creative-seat share of an organisation this size - tens to hundreds of seats, not a licence for every employee. Render APIs price by usage (US$5k-60k). DAM is quote-only, commonly US$50k-250k and up. Penpot is hand-set at 99: "your server" can be your own laptop - a small technical hurdle at well under 1% of an Adobe-scale spend. Lolly + lolly.work: US$0 in software licence charges for self-hosting. The figure excludes devices, organisation hosting, storage, support and administration; these need a separate budget. Standalone Lolly does not need the organisation service. Cloudinary: US$30k-150k, credit-based; enterprise contracts commonly sit near US$82k.
$0 $0 $400k $500k $250k+ $150k $250k $60k

現今各創意工具的功能完整度,調查於 2026 年 8 月。評分標準:0 分表示不具備,25 分表示需以變通方式達成,50 分表示有實際功能但受限或不完整,75 分表示強大但有但書,100 分表示核心能力。

signed by Lollypositioning-comparison.htmlHTMLpixels, not shapes40 KBAI generatedgenerated by GPT-6自己驗證取得已簽署檔案

評分說明。 功能是否具備與生產成熟度是兩回事。Lolly 的成熟度評分反映的是 2026 年 8 月的首次公開發佈,以及尚無獨立資安稽核;具備 SSO、核准或伺服器端算繪,並不能證明它有正式環境中的實際紀錄。以下的功能描述包含 lolly.work,而競品的評分仍保留其標註日期的 8 月調查結果。

Canva 的評分依每列以其家族中最佳的成員來計算,因為它擁有 Affinity 與 Cavalry(兩者皆於 2025 年 10 月起免費提供)。離線與裝置端渲染評 75 分,靠的是 Affinity - 一套仍需要已驗證帳號、也帶有遙測的桌面套件,這也是 Adobe 同樣被扣分的原因 - 而 Canva 自身的離線模式只能編輯先前已同步過的設計,僅限單一裝置,視窗也有限。自動填入評 50 分:功能屬實,但僅限企業版、非同步、只支援文字與圖片。當 Buzz 推出試算表填入功能後(免費測試版,2026 年 8 月),Figma 的大量生成評分從 25 分升至 50 分。

整張表遵循一條規則:凡涉及你的內容或身分的列,要拿到滿分(100)必須有一項無須帳號、也無雲端前提即可使用的能力;描述產品本身的列(成熟度、易用性)不受此規則限制。這條規則讓 Adobe 在來源溯源(provenance)一項上被扣分:目前發佈範圍最廣的 C2PA 支援(Photoshop、Lightroom、Premiere、Firefly)可在本機與雲端簽署,但一律需要 Adobe 帳號與身分,所以評 75 分。同一條規則也讓渲染 API 在大量生成與自動化上被封頂扣分。

Lolly 的來源溯源評 75 分,反映的是裝置端離線簽署:架構上更為穩固,但尚未經過稽核,而且裝置金鑰在標準驗證器中會顯示為未經驗證,除非有身分機構或組織自有的 CA 為其背書。Penpot 的 50 分是透過官方的 Lolly Export 外掛取得:採用同一套引擎簽署,選擇性加入,並揭露為 Lolly 自有的功能。Penpot 也拿下了整張表中唯一超出常規尺度的環 - 裝置端渲染 90 分 - 瀏覽器畫布、儲存到你自己主權可控的雲端(甚至是一台筆電)、私人匯出;與 Lolly 的差別只在於多了一道伺服器跳轉。Cloudinary 有自己獨立的一欄:它是一套媒體管線(數位資產管理、轉換 API、CDN),也是唯一一個發佈了 C2PA 支援的雲端欄位(50 分,因為 fl_c2pa 是在交付時簽署,證明的是「由 Cloudinary 交付」,而非「由你製作」)。

即時協作則反過來:Figma 訂下了規模基準(200 位編輯者同時在線),而 Lolly 一對一、氣隙隔離(air-gapped)的 P2P 協作評為「部分」。價格是一項標明為推測值的估計:以真實座位組合計算牌價,刻意估得寬鬆,用於呈現規模而非用於採購。渲染 API 在限制條件上評 75 分:範本是鎖定的,沒有品牌治理層。

差距在於:目前沒有任何已發佈的產品是「以限制條件優先」且能離線運作、無須帳號、渲染路徑中也沒有伺服器的 - 也沒有任何競品複製了這條「無帳號」的準則。Lolly 現在也推出了自有的開放式畫布 - Design,一個可直接操作的自由畫布 - 但其上的顏色、字型與資產都遵循品牌全域設定,所以即使是自由編排,依然維持限制條件優先。

Lolly 目前仍不是一套不受限制的設計套件;設計師仍會繼續使用 Illustrator 與 Figma 來進行客製化的工作 - 而當那份工作需要變成一項受治理、可重現的資產時,Design 工具的匯入設計功能會把完成的 Figma、Penpot、Illustrator、InDesign 或 PDF 檔案帶入畫布,成為可編輯、符合品牌規範的方塊。

Design 的自由畫布,其上提供的顏色、字型與資產皆來自品牌本身 signed by Lollyvector SVG71 paths~2.7k nodes59 groups75 KB自己驗證取得已簽署檔案Design 的自由畫布,其上提供的顏色、字型與資產皆來自品牌本身 signed by Lollyvector SVG64 paths~2.7k nodes59 groups74 KB自己驗證取得已簽署檔案

適合用它做的事

Deck Studio 很能衡量這裡的能力上限:一整份投影片組以資料形式宣告,在畫布上即時排版,再匯出為原生可編輯的 PowerPoint。

Deck Studio 的分割檢視 - 左側以區塊列出投影片組的每張投影片,右側則即時渲染排版完成的投影片組 signed by Lollyvector SVG135 paths~17k nodes252 groups1 image226 KB自己驗證取得已簽署檔案Deck Studio 的分割檢視 - 左側以區塊列出投影片組的每張投影片,右側則即時渲染排版完成的投影片組 signed by Lollyvector SVG128 paths~17k nodes252 groups1 image225 KB自己驗證取得已簽署檔案

不適合用它做的事

以機率的方式創新,以確定的方式規模化

多數「AI 創意」的推銷詞把模型放錯了那條古老界線的一側。抄寫員與泥金裝飾師早就確立了界線該落在哪裡:你在草稿階段自由發揮,那裡任何嘗試都可以進行,也沒有任何東西是定案的;然後你才走向印刷機,印刷機之所以令人卻步,正是因為它會定案。草稿才是藝術所在之處。印刷機則是藝術得以流傳的方式。兩種工具,兩種任務,各自以自己的方式富有創造力,而印刷出來的成品之所以值得信賴,是因為印刷機每一次印刷都信守承諾。

Lolly 是印刷機,不是草稿。你可以把任何東西帶進發想階段 - 一個模型、一位設計師、一張餐巾紙 - 但當一個構想必須變成許多資產時,它就要經過某個每次算繪結果都相同、且輸入內容任何人都能讀懂回溯的東西。這正是上面這場比較真正的重點:不在於誰有更好的生成器,而在於誰能讓那個定案的步驟可重現。

信任創意過程,以嚴謹擴展規模。

規則存在於工具,而不是檔案

市面上其他每一項工具產生的都是一個檔案,而檔案只能事後查看,一次一份。Lolly 把品牌規則往上游移動一步。精確的十六進位色碼、已授權的字型檔、出血邊界、間距 - 這一切都直接寫死在工具的 HTML 與 CSS 中,因此範本不可能產出偏離品牌的成品。真正負責強制執行的是版面配置本身。

所以真正值得留意的是工具本身,而不是它產出的每一個檔案。把工具做對,它日後產出的每一份資產都會因構造方式而符合品牌規範。你如何檢查自己的工作、由誰來檢查,依然由你決定;Lolly 只是讓被檢查的東西變小,產出做得更快。

這才是確定性引擎真正帶來的改變。對創意團隊來說,這是護欄,而非取代 - 你依然投出那顆球(資料、文案、圖片),而程式碼就是保齡球道兩側的護欄,讓每一次投擲都不會滾進溝裡。

製作人唯一的工作:輸入文字。字體、顏色與間距早在工具做好時就已經定案 signed by Lollyvector SVG121 paths~4.6k nodes197 groups2 images113 KB自己驗證取得已簽署檔案

這帶來的獨特價值

其中第一項是最常被低估的。只要一個下拉選單和兩個無法脫離品牌範圍的色彩欄位,就能畫出海報等級的城市地圖,道路與水域皆為真正的向量路徑:

阿姆斯特丹的運河環與道路網以品牌自身的墨色繪滿整張畫面,每一筆都由範本而非手工繪製 signed by Lollyvector SVG8 paths~25k nodes7 groups287 KB自己驗證取得已簽署檔案阿姆斯特丹的運河環與道路網以品牌自身的墨色繪滿整張畫面,每一筆都由範本而非手工繪製 signed by Lollyvector SVG8 paths~25k nodes7 groups287 KB自己驗證取得已簽署檔案

內容主權

前一節加總起來的結果有個名字:主權。你的媒體流程運行在你自己擁有的硬體上。你的品牌 - 代符(token)、字型、標誌,以及執行這些規範的工具 - 都存放在你握有的檔案裡、由你掌控的版本控制系統中,而非某家廠商附帶「匯出」按鈕的資料庫裡。算繪發生在你眼前的裝置上,因此成品的存在從不需要經過第三方,從輸入到完成檔案的整條路徑都是開放原始碼且可供檢視的。就算所有 SaaS 設計廠商明天全部消失,Lolly 的部署也不會受到任何影響。

這對任何作品理應比訂閱方案更長久的人都很重要:無論是相片書存放在那台筆電上的家長,還是品牌資料庫受採購規範約束的公家機關。對組織而言 - 公家機關、受監管產業,任何品牌是策略性資產而非裝飾品的單位 - 「我們的內容存放在哪裡、誰能關掉它」是治理問題,而非個人偏好。這裡所說的主權是架構本身的特性,而非為了合規才加上的代管功能,隱私權政策與自行驗證這兩個頁面的存在,就是為了讓你能親自查核這項主張,而非全憑信任。

這一切底下有一個承諾,以承諾而非功能的姿態陳述:只要是在你的裝置上算繪,就永遠免費。 引擎、殼層、工具、格式 - 整條裝置端的創作路徑都是開放原始碼,而且會一直維持如此。這項承諾有其機制作為後盾:已發布的版本所採用的授權方式使其無法被收回,也不存在任何貢獻者合約能在日後重新授權這份作品。整個界線可以濃縮成一句話:凡是在你的裝置上算繪的一切,永遠免費且開放原始碼;跨網路協調人與機器則是另一個獨立控制平面 lolly.work 的工作。