In-depth Report
-
Tencent Miora (Tencent Design Miora) is an AI creative agent studio fully launched by Tencent on July 22, 2026. It is positioned as the "creative design version of WorkBuddy". It uses multi-agent collaboration to break down the user's one-sentence requirements into images, videos, 3D, UI and other tasks, and delivers a complete set of visual solutions with a unified style on the same canvas. The biggest differentiation of the product lies in its "ability to think and remember" and its ability to condense the creative process into reusable Skills, allowing non-designers to run a complete set of brand materials.
-
Miora was built by Tencent's CodeBuddy/WorkBuddy team. It shares the same Agent underlying architecture with the latter two, but is deeply customized for the creative design field. It was first released on Cloud Day in Hong Kong in May 2026. It ranked first on Product Hunt’s daily product list on July 13. On July 22, the invitation limit for internal testing ended and it was open to global users (the international version was fully launched). New users will receive 1,000 points when they register. It is Tencent's third product after CodeBuddy and WorkBuddy, which extends the agent architecture to the vertical field. Officials also revealed that WorkBuddy will support direct calls to Miora capabilities in the future. Judging from Tencent's overall layout, Miora is an important part of its "intelligent agent track" for the C-end and creative people.
-
The core of Miora is an arrangement of "one master Agent plus multiple expert Buddy". The user throws in a brief, and the main Agent first understands the goal, breaks down the task, and then schedules Buddy, an expert in the corresponding field, to work together; the sub-Agents currently online include rendering, audio and video, 3D modeling, video rendering, and UI design. Different from traditional single-point tools that "you write prompts and it produces a picture", Miora can produce a complete set of brand visual solutions in one go. All deliverables naturally follow the same set of style standards, eliminating the need for manual alignment in the later stage. The interface that carries full-link operations is an editable multi-modal infinite canvas. Images, videos, 3D models, UI design drafts, and texts are all freely placed and combined in the form of nodes. The Agent can read the canvas context and continuously collaborate with the user, and supports operations such as marking and circling the modified area, frame selection editing, local fine-tuning, modifying angles, cropping, erasing, high-definition zooming and other operations - no need to overturn and start again. Many actual test users have mentioned that it is easier to use two strokes than to write hundreds of words using Prompt, and each step can be adjusted, instead of waiting for a long time for an unmodifiable result. Memory systems are another key capability. Miora not only remembers used materials, but also understands the user's aesthetic preferences, brand specifications, creative methods and project context, divided into two layers: "project memory" and "personal memory". It can be viewed and edited by users at any time, and the longer it is used, the closer it is to expectations. According to the official statement, by the tenth cooperation, the results will look like they were made by the users themselves. In terms of model scheduling, Agent inference base models are divided into three levels: Standard, Pro, and Max. The multi-modal generation model pool has built-in NanoBanana, GPT-image, Seedance, Keling, Vidu Q3, etc. Users can specify or automatically match; both interrogative and fully automatic generation modes are provided. The Skills platform encapsulates the entire workflow into reusable and shareable assets. After users run through a set of brand solutions, they can save it as an exclusive private skill by saying "precipitate into skill" in natural language. They can then reuse the same brand design with one click, or publish it to the shared market for others to call. This turns "the pits that have been traversed and the processes that have been passed through" into a creative methodology that can be accumulated.
-
Miora currently uses a points system. After the full launch, new users will receive 1,000 creative points for free upon registration, which can be used to offset the consumption of various advanced creative functions within the platform. They will be able to run through the entire process from brief to delivery before paying. Low-level reasoning is billed in Standard / Pro / Max brackets. The more complex the case (such as brand VI), the higher the consumption, the Max in-depth reasoning needs to be turned on. It should be pointed out that there is currently no complete disclosure of paid membership levels and unit prices in public information, and the specific points pricing system and enterprise version plan are still under observation. Judging from Tencent's consistent approach, using free points to attract new users and then monetizing high-value creations is the most likely path.
-
The positive feedback focused on a few things. The first is "one brief to get a complete set of materials", the style is unified from beginning to end, and there is no need to go back and forth; some users spent an afternoon from scratch to create "distributable" brand online materials, ranging from logos and visual specifications to e-commerce details pages and 30-second TVCs. The second is the memory function. After the user feeds in the brand tonality, color matching restrictions, and aesthetic preferences, subsequent projects will no longer hit the mark. It is repeatedly said that "the more you use it, the more you will understand." Third, local modifications are precise and will not cause damage to other places after changing one place, which is especially friendly to novices. The internal test data disclosed by Tencent also shows that more than half of the users invited to test are not professional designers, but come from R&D, market operations and other positions, confirming the trend of civilianization of creative tools. Negatives and shortcomings are also recorded honestly. The most commonly mentioned problem is the "last mile" problem: Miora does not currently import vector source files such as AI, PSD, SVG, etc., and manual vectorization is still required after the complete VI is output. Some actual measurements also pointed out that the model relies on external multi-modal engines (such as GPT-image), which is still subject to cost and stability; individual generated results will have detailed flaws that require rework. These are not fatal, but they mean that it is more suitable for the division of labor of "prototyping from 0 to 1, detailing from 1 to 10 and handing it over to people", rather than completely replacing professional designers.
-
Many Chinese technology media regard Miora as a representative of "turning AI tools into AI teams". Geek Park calls it an important move for Tencent in the intelligent agent track, and it is also the third product whose architecture extends vertically after CodeBuddy and WorkBuddy. Some reviewers pointed out from an engineering perspective that it and WorkBuddy have the same set of Agent orchestration ideas - one writes code to check data, and the other creates visuals. The difference is not how powerful the model is, but the arrangement of "main Agent plus Specialist" is truly incorporated into the design. The industry generally recognizes its combination of "multi-agent collaboration + memory + reusable skills" and believes that it addresses the three major pain points of traditional AI creative tools: "style drift, platform switching, and AI amnesia."
-
One is the dependence on external multimodal models. Miora has built-in domestic and overseas models such as NanoBanana, GPT-image, Keling, Vidu, etc. Once an underlying model adjusts its price or interface, its cost and availability will be directly transmitted. The second is data ownership and copyright. Users continuously feed brand specifications, aesthetic preferences, and project assets to the memory system. There is currently limited public explanation of the ownership of these data, whether it can be exported, and the copyright boundaries of the generated content. The third is the common implementation bottleneck of "AI design" - vector source files are incompatible, and the finishing of complex print-level output still requires manual labor, which means that it is still far from "fully automatic design delivery". Fourth, as a newly launched product, it still needs time to test whether the long-term stability and points pricing strategy will change.
-
There are three types of people who are most suitable for Miora. Designers and creative practitioners use it to put all categories on the same canvas in one stop, with a unified style and no need to go back and forth. Entrepreneurs and small teams can run a complete set of brand launch materials at a low cost by one person and quickly implement ideas. Operations and e-commerce merchants use dozens of detail pages and dozens of short videos during major promotions. Using Skill precipitation to template the running process is more valuable than a single surprise. Product managers and developers can also use it to quickly produce interactive prototypes and reduce communication costs. The people it is not suitable for are also clear: professional publishing and packaging scenarios that require print-level vector source files and have strict requirements for pixel-by-pixel accuracy still have to be finished manually; companies that have strong restrictions on data export and compliance also need to first evaluate their links to call overseas models. In terms of alternatives, single-point image generation can include Midjourney and Stable Diffusion, and comprehensive design platforms include Canva Magic and Adobe Firefly. Competing products in the similar "design agent" direction are also rapidly emerging.
-
Miora is not another drawing toy, but a creative partner that truly integrates Agent orchestration into the design workflow; it takes over repetitive, trivial, and time-consuming links and allows users to only be responsible for the core creative judgment. For those who are tired of tossing between multiple AI tools, it’s worth trying it for yourself before judging.
User Reviews
-
Russell.Parker369—刚注册领了 1000 积分,先玩为敬。 -
Johnny_Flores_Max—请问现在付费档位出来了吗？1000 积分用完之后怎么收费呀？ -
JacobCampbell_2021—偶尔还是会有风格不统一的时候,得手动介入调一下,但比全部自己来强太多。 -
eRIClOPEZ—支持导入自己的 AI 矢量文件吗？还是只能传图片？ -
梁婷_1—内测数据说一半以上用户不是设计师,我是做运营的,平时大促几十张详情页手工做要命,现在用 Skill 模板套,终于不用求人了。 -
枫叶635—分享个玩法:把跑通的工作流直接说「沉淀成 Skill」就能存成专属模板,下次同品牌换产品直接套,风格还不变,对大促批量出图太爽了。 -
Logan.Torres_7—我们小团队没有设计,以前做品牌物料要么外包要么凑合,沟通成本还高。这次拿 Miora 跑了一下午,从 Logo、视觉规范到电商详情页、短视频脚本全有了,虽然细节要人工收尾,但对刚起步的我们来说已经是真杠杆了,至少从 0 到能发出去这一步它帮了大忙。 -
purplecat355—一份 brief 直接出整套品牌物料,这效率真的离谱。 -
Donald_Vasquez_2023—做落地页那个场景我实测过。以前要么自己学 Figma 要么跟设计师磨,现在大白话告诉它「要一个 AI 写作工具的落地页,简单好上手」,它直接给三套结构和配色字体建议,虽然不是最终稿,但把我从空白推进到了 0.3,这个跳跃非常大,后面改到 1.0 就快多了。 -
Sandra_RogersK—不导入 PSD 和 AI 源文件这点还是难受,做 VI 最后还得回 AI 里矢量化。 -
GeraldReedZ—记忆功能是我没想到的好用。我把「品牌调性」和「配色禁区」喂进去之后,后面出的图风格基本稳了,不用每次重新讲一遍。 -
crazyleopard118—3D 生成我也试了,让它出一个科技感大脑图标,真给了个能在 blender 里打开的模型文件。做电商和广告素材的话,从 0 到毛坯压缩到分钟级,细节再找人打磨就行。 -
Mason.Cox—完整跑了一遍品牌全案,说说真实感受。多 Agent 协同确实不是噱头,主 Agent 会先拆解任务再派活给生图、视频、3D 那些 Buddy,出来的东西风格是统一的,省了后期对版的功夫。画布上框选局部重绘也很顺手,改哪指哪不用推翻重来。唯一卡我的点是它现在不吃 PSD 和 AI 矢量文件,品牌 VI 最后一步还得在 AI 里手动矢量化,这就是所谓的最后一公里吧。 -
Carol.Richardson_77—做公众号封面终于不用自己抠了。把最满意的一张丢进去让它「逆向」出我的风格规范,后面新封面直接套,一致性比手做还稳。 -
ÇetinKorol—说个真实案例。我们做智能宠物品牌,从 Logo、三款产品外观、包装、IP 形象到周边、45 平体验店、配套 App、电商详情页、30 秒 TVC,全让它跑了一遍。整体 Buddy 之间风格能保持一致,IP 形象和 Logo 从头到尾没翻车,这点很稳。中间每一步都有确认卡片,想全自动也能切自动模式。扣点是斜角角度偶尔画不精确,得人工修,但这本来也难为 AI。 -
汪妍—跟 Midjourney 那种单点生图不一样,Miora 更像一个「创意总监」,你跟它讲目标,它去协调各个专家出整套方案,这个思路才是对的。 -
MarilynBell_771—画两笔比写几百字 Prompt 好用。 -
NicoleWolf—小白也能上手,首页那个灵感精灵妙妙有点可爱。 -
SRichardson_2022—登上 PH 日榜第一确实有点东西。 -
AidenHolland—官方说新用户送 1000 积分,够先完整跑一遍流程体验了,付费之前就能看效果,这点挺良心。