{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-07-28T07:00:52.885Z","headline":"FLUX 3 x mimic：新一代视频动作模型","description":"Black Forest Labs 发布多模态基础模型 FLUX 3，联合训练图像、视频和音频，其中视频预测占训练算力的 95% 以上。该模型与机器人公司 mimic 合作推出 FLUX-mimic，已在奥迪生产线上测试部署。加入动作预测后，视频生成质量最初下降最多 10%，但经 3500 步训练后恢复原有水平。","url":"https://www.aioga.com/news/cmrywwsva05zgrolgrh7uc4ls/","mainEntityOfPage":"https://www.aioga.com/news/cmrywwsva05zgrolgrh7uc4ls/","datePublished":"2026-07-24T11:55:45.885Z","dateModified":"2026-07-24T11:55:45.885Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://bfl.ai/blog/flux-3-mimic","https://aihot.virxact.com/items/cmrywwsva05zgrolgrh7uc4ls"],"canonicalUrl":"https://www.aioga.com/news/cmrywwsva05zgrolgrh7uc4ls/","directAnswer":{"@type":"Answer","text":"Black Forest Labs 公开 FLUX 3 早期版本，该模型从训练之初联合处理图像、视频与音频，并与 mimic 合作形成视频动作模型 FLUX-mimic，相关机器人已在奥迪测试和部署。","url":"https://www.aioga.com/news/cmrywwsva05zgrolgrh7uc4ls/","dateCreated":"2026-07-24T11:55:45.885Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"bfl.ai source article","url":"https://bfl.ai/blog/flux-3-mimic","datePublished":"2026-07-24T11:55:45.885Z","provider":{"@type":"Organization","name":"bfl.ai","url":"https://bfl.ai/blog/flux-3-mimic"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.virxact.com/items/cmrywwsva05zgrolgrh7uc4ls","datePublished":"2026-07-24T11:55:45.885Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.virxact.com/items/cmrywwsva05zgrolgrh7uc4ls"}}],"aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","originalPublisher":{"name":"bfl.ai","url":"https://bfl.ai/blog/flux-3-mimic"},"article":{"id":"cmrywwsva05zgrolgrh7uc4ls","slug":"cmrywwsva05zgrolgrh7uc4ls","url":"https://www.aioga.com/news/cmrywwsva05zgrolgrh7uc4ls/","title":"FLUX 3 x mimic：新一代视频动作模型","title_en":"Flux 3 X Mimic：新一代视频动作模型","summary":"Black Forest Labs 发布多模态基础模型 FLUX 3，联合训练图像、视频和音频，其中视频预测占训练算力的 95% 以上。该模型与机器人公司 mimic 合作推出 FLUX-mimic，已在奥迪生产线上测试部署。加入动作预测后，视频生成质量最初下降最多 10%，但经 3500 步训练后恢复原有水平。","source":"Hacker News 热门（buzzing.cc 中文翻译）","sourceUrl":"https://bfl.ai/blog/flux-3-mimic","aiHotUrl":"https://aihot.virxact.com/items/cmrywwsva05zgrolgrh7uc4ls","publishedAt":"2026-07-24T11:55:45.885Z","category":"模型更新","score":75,"selected":true,"articleBody":["An early version of FLUX 3, our new multimodal foundation model：https://bfl.ai/blog/flux-3, is now running on robots. We gave mimic robotics：https://www.mimicrobotics.com/ early access to FLUX.3. Their strength in robot learning and deployment, combined with the model's world knowledge and BFL's foundation model expertise, produced FLUX-mimic: the next generation of video-action models.","FLUX 1 and FLUX 2 generate images. FLUX 3 expands into multimodality and generates audio-visual content jointly - and, at the same time, provides the foundation of FLUX-mimic: A video-action model, developed in collaboration with mimic, running robots that have been tested and deployed at Audi.","At first glance, producing convincing visual content and controlling robots seem to have little in common. One requires generating pixels, the other an understanding of how the physical world responds when you touch and manipulate it. If one model does both, it was never really only a content creation model. It is a model of how the world behaves, and content creation is one thing one can do with it.","FLUX 3 is one model, jointly trained across images, video and audio from the beginning. The most demanding part of that training - accounting for over 95% of the total compute costs - is video prediction. To generate realistic videos, a model has no choice but to learn contact, motion, weight, cause and effect; get any of them wrong and it looks wrong. Learning to render the world accurately means learning how the world behaves.","Relatively speaking, audio is the easy modality. Low dimensional and far less detailed than video, it makes up less than 0.5% of the tokens in a 720p video with audio. Once a model has done the hard work of learning video understanding, it will learn the causal relationship between video and audio to predict speech synchronized to lip movement and audio effects synchronized to the physical events causing them.","Actions follow the same shape: a low dimensional representation of a robot's state, tightly coupled to visual observations. Actions, audio and video frames are all partial representations of a single underlying physical reality. After the model has learned about the physical processes behind video and audio, action prediction is not a new departure - it is one more view of the reality it already models.","If that framing is correct, teaching FLUX 3 to predict actions should not incur lasting costs: we expect a brief phase of disturbance as the model has to learn the structure of the action space and align its internal representation of the world to it, before returning to full performance. That is exactly what we observe.","In a large-scale training run, we added action prediction to the curriculum and observed the effect on video generation quality. Human ratings on text-to-video and image-to-video initially fell by up to 10% as the model started to incorporate the new action modality. After 3500 steps, the model had regained its full previous quality on video generation tasks while now also predicting actions.","Each series is normalized to its own quality before action prediction was added. Higher is better.","The model had to integrate actions into its inputs and outputs - but doing so didn't cost it capacity permanently. It merely had to learn how this new modality relates to its existing model of the world. Once this was figured out, the performance penalty on its existing capabilities was gone. Video generation and action prediction don't need separate foundations. The same backbone carries both.","This makes Physical AI a natural extension of our roadmap at Black Forest Labs rather than a change in direction. Content creation is what our multimodal FLUX 3 backbone does with image, video and audio. Physical AI is what it does with actions. One foundation model, with visual intelligence at its core, enabling two families of applications. We didn't build a separate foundation model. We focused on the hard thing: building a model that understands the world. Acting in it is what that understanding makes possible.","What happens when we point the FLUX 3 backbone at real automation tasks on real production lines?","That's the question mimic and BFL created FLUX-mimic to answer. mimic builds their own robots and brings expertise in robot learning, dexterous manipulation and production deployment; BFL builds visual foundation models and brings multimodal training and modeling expertise. Together, we built a next-generation model for general-purpose manipulation - adapted to industry requirements and integrated into mimic's full-stack deployment system. FLUX-mimic is a video-action model built on the FLUX 3 backbone.","Our thesis is that FLUX 3 has to learn an internal representation of the world to be able to generate videos. FLUX-mimic follows through on this thesis and decodes actions from the learned world representation of the FLUX backbone. This approach, pioneered in mimic-video, trains a lightweight action decoder on top of intermediate features extracted from the video prediction path of FLUX.","Architecture overview how FLUX-mimic is built on top of FLUX 3","The success of this approach depends on two related but different aspects: the quality of the world model learned by FLUX and the quality of the feature representation of this world model. The quality of the world model is directly related to the generation quality: if a model does not understand how the world behaves, it cannot simulate it. However, even the best world model does not help an action decoder if it is inaccessible: if the feature space keeps the causal relationships between modalities entangled nonlinearly, understanding those relationships from the feature representation remains as difficult as understanding them from the raw inputs - representation quality matters.","Generation quality and representation quality have long been studied and approached in isolation from each other. Generative approaches result in high-quality world models that enable simulations and they exhibit scaling laws for predictable returns on compute investments. However, compared to more specialized approaches for representation learning they produce less disentangled representations, which puts a ceiling on their usefulness for tasks that require world understanding.","As generative models themselves rely on their own features, this divergence in their representation quality seems counter-intuitive. Improved representations within generative models should improve the quality of their world model and make them more usable for downstream tasks. In our work, Self-Flow：https://bfl.ai/research/self-flow, we demonstrated how to unify generation and representation learning in a single framework and observed exactly this reciprocal improvement: the world model improved - as measured by generation quality across video, image and audio - and its representation quality improved - as measured by success rate for robot control tasks in simulation.","Self-Flow vs. Flow Matching (FM). Left: generation error (Fréchet distance) per modality, each normalized to FM = 100 (lower is better). Right: success rate on manipulation tasks averaged over four task groups through finetuning (higher is better).","Scaling laws remain true with Self-Flow, and FLUX 3 is the application of that: the scaled-up version of Self-Flow. It is trained on tens of millions of hours of general video content to learn world dynamics as broadly as possible from day one, and on hundreds of thousands of hours of video content focused on human and robot manipulation tasks to be ready as a backbone for visual intelligence. This scaling is what translates the success of Self-Flow from the lab to reality. mimic deployed FLUX-mimic in real factory use cases spanning the daily reality of production and logistics work: kitting parts into structured trays, inserting electronic control units into tight-fitting fixtures, assembling components together, and handling soft, flexible materials like seals and cables that conventional automation has never been able to touch.","Benchmarks demonstrate that the action decoder outperforms previous vision-language-action models, even with a completely frozen FLUX backbone - a setting where previous vision-language-action models fail to succeed. This highlights how scaling gives our backbone strong knowledge of the world and how to act in it, and how Self-Flow makes this knowledge readily decodable from the backbone's feature representations. When finetuning the backbone together with the action decoder, FLUX-mimic achieves state-of-the-art success rates.","Dashed line marks each model's median success rate across 20 autonomous trials. Higher is better.","A backbone exposing world knowledge in decodable representations changes what it takes to teach a robot a new task. If the physics is already in the representation and readily accessible, adapting to a task is no longer a matter of teaching the model how the world works - it only has to learn how this particular task maps onto what it already knows. The expensive part is done before the robot ever moves.","This shows up directly in how much demonstration data a new task requires. In our Self-Flow experiments, action prediction reached a given success rate in half the training steps compared to a video model without Self-Flow - better representations make the world knowledge easier to extract, so less data is needed to reach the same capability. The mimic-video paper reports up to 10x sample efficiency for video-action models over vision-language-action models; FLUX-mimic combines both effects.","The same benefit shows up in behavior. FLUX-mimic naturally recovers from failure: a robot that misses a grasp corrects itself, grasps again, and completes the task. No demonstration set can cover every possible way a task can go wrong. Recovery that was never demonstrated has to come from somewhere else - from a model that already knows how the world behaves.","The backbone's predicted future (top), alongside the rollout the robot produced from the decoded actions (bottom).","Real-world deployment sets a hard constraint: the model has to act as fast as the world moves. The dominant compute cost for FLUX-mimic sits in the backbone. It is the largest component of the model and, in mimic's optimized deployment stack, its latency effectively sets the ceiling for the whole system.","This is where our methodology pays off a second time. Better representations mean more capability per parameter: a model that has learned a well-structured world model needs less capacity to reach a given level of performance than one that has not. For deployment, this translates directly into being able to run a smaller backbone - and a smaller backbone is a faster backbone.","CLIP score at 1.0M training steps; higher is better. Backbone depth is the dominant driver of deployment latency, so fewer layers means a faster model.","As a result, the backbone of FLUX-mimic can be optimized to run from input to world representation in less than 80ms on a single NVIDIA RTX 5090 GPU - which puts it on the same order of magnitude as human visual reaction time.","Real-world deployments require additional optimizations of the full deployment stack to avoid adding any additional latency: mimic's optimizations range from the action decoder, through cutting inter-process latency between sensors, the model and actuators, to real-time chunking such that prediction and execution overlap and keep the robots running smoothly without jitter. The end result is a self-contained robot system with reaction times of 101ms.","All of this leads back to the place where automation matters: the factory. Audi runs one of the most automated production networks in the automotive industry - which gives it a precise view of where conventional automation still stops. Despite decades of robotics investment, tasks with flexible parts and fine manipulation have stayed manual, largely for economic reasons: the variant diversity of premium production makes conventionally programmed robot cells too costly to re-engineer for each case. Learning-based systems change that math.","\"In partnership with mimic, Audi has been testing and deploying FLUX-mimic. We have seen these robots solve complex soft-body manipulation work that would have been simply impossible with conventional robotics. This can have a major impact in assisting our employees, increasing efficiency, and expanding flexible automation across production and logistics operations. For us, partnering with pioneering companies such as mimic and Black Forest Labs is essential in pushing the frontier of physical AI and validating these innovations in real-world production environments.\" — Christoph Schneider, Audi Production Lab","FLUX-mimic is a purpose-built robotics model - and a proof point for what's to come. Its sample efficiency and robustness come from the FLUX 3 backbone and the quality of the representations it exposes, not from task-specific engineering. That is what lets the approach transfer across tasks, industries, and hardware. Read more through mimic：https://www.mimicrobotics.com/blog/introducing-flux-mimic.","One model, with visual intelligence at its core, generating image, video and audio - and driving robots on a production line. Content creation and physical AI are two applications of the same foundation."],"articleImages":[{"sourceUrl":"https://bfl.ai/_next/image?url=https%3A%2F%2Fcdn.sanity.io%2Fimages%2F2gpum2i6%2Fproduction%2F630a011fdc057238266e03b1437ae9667a89295d-1920x1080.jpg&w=3840&q=75","alt":"FLUX 3 x mimic: The Next Generation of Video-Action Models","afterParagraph":0,"url":"/media/articles/cmrywwsva05zgrolgrh7uc4ls/dea54e7a1702562c.avif"},{"sourceUrl":"https://bfl.ai/_next/image?url=https%3A%2F%2Fcdn.sanity.io%2Fimages%2F2gpum2i6%2Fproduction%2F949f7f0a4c6229b2a6aadaa89187200f43ccfd77-2880x1840.png&w=3840&q=75","alt":"","afterParagraph":7,"url":"/media/articles/cmrywwsva05zgrolgrh7uc4ls/ae5da37f68bf49d0.avif"},{"sourceUrl":"https://bfl.ai/_next/image?url=https%3A%2F%2Fcdn.sanity.io%2Fimages%2F2gpum2i6%2Fproduction%2F0c904136368acfde7f49c6941c0037d05b7feeb9-2232x1452.png&w=3840&q=75","alt":"","afterParagraph":13,"url":"/media/articles/cmrywwsva05zgrolgrh7uc4ls/4505d254cda5163b.avif"},{"sourceUrl":"https://bfl.ai/_next/image?url=https%3A%2F%2Fcdn.sanity.io%2Fimages%2F2gpum2i6%2Fproduction%2F12f116224e4991c6d031a2f660829a1b8186fb55-2240x1152.png&w=3840&q=75","alt":"","afterParagraph":17,"url":"/media/articles/cmrywwsva05zgrolgrh7uc4ls/b0c6b147c01eacd8.avif"},{"sourceUrl":"https://bfl.ai/_next/image?url=https%3A%2F%2Fcdn.sanity.io%2Fimages%2F2gpum2i6%2Fproduction%2Fd1de2431871ca972fa647bc0a12af4bea6cefd0d-2360x1346.png&w=3840&q=75","alt":"","afterParagraph":20,"url":"/media/articles/cmrywwsva05zgrolgrh7uc4ls/9bb809c54b28211b.avif"},{"sourceUrl":"https://bfl.ai/_next/image?url=https%3A%2F%2Fcdn.sanity.io%2Fimages%2F2gpum2i6%2Fproduction%2F1b3771a0506ced1533053990f73340bfc397b5ad-2080x1340.png&w=3840&q=75","alt":"","afterParagraph":27,"url":"/media/articles/cmrywwsva05zgrolgrh7uc4ls/6ac82d2fdf1fc583.avif"}],"mediaStatus":"ok","articleBodyZh":["FLUX 3 的早期版本，我们的新型多模态基础模型：https://bfl.ai/blog/flux-3，现在已经在机器人上运行。我们给了 Mimic Robotics：https://www.mimicrobotics.com/ 早期访问 FLUX.3 的机会。他们在机器人学习和部署方面的优势，结合模型的世界知识和 BFL 的基础模型专业知识，产生了 FLUX-mimic：下一代视频动作模型。","FLUX 1 和 FLUX 2 生成功能是生成图像。FLUX 3 扩展到了多模态，并联合生成视听内容——同时也为 FLUX-mimic 提供了基础：这是一个视频动作模型，由 Mimic 协作开发，可在已经在奥迪测试和部署的机器人上运行。","乍一看，生成逼真的视觉内容和控制机器人似乎没有什么共同点。一个需要生成像素，另一个需要理解当你触碰和操作物理世界时它的响应方式。如果一个模型两者都能做到，它从来不只是一个内容生成模型。它是一个世界行为模型，而内容生成只是它可以做的一件事。","FLUX 3 是一个模型，从一开始就跨图像、视频和音频联合训练。训练中最苛刻的部分——占总计算成本的超过 95%——是视频预测。为了生成逼真的视频，模型别无选择，只能学习接触、运动、重量、因果关系；任何一个学错了，看起来就会不对。学习精准地呈现世界意味着学习世界如何运作。","相对而言，音频是容易的模态。它的维度低，细节远少于视频，在带音频的 720p 视频中占不到 0.5% 的 token。一旦模型完成了学习视频理解的艰难工作，它将学习视频和音频之间的因果关系，以预测与唇部动作同步的语音以及与引起物理事件同步的音效。","动作遵循相同的形式：机器人状态的低维表示，与视觉观测紧密关联。动作、音频和视频帧都是单一底层物理现实的部分表示。在模型学习了视频和音频背后的物理过程之后，动作预测并不是一个新的突破——它只是对模型已经建模的现实的另一种观察视角。","如果这种框架是正确的，教 FLUX 3 去预测动作不应产生长期的负面影响：我们预计在模型需要学习动作空间的结构并将其内部世界表示与之对齐的过程中，会出现短暂的扰动期，然后模型会恢复到全性能状态。这正是我们观察到的情况。","在一次大规模训练中，我们将动作预测加入训练计划，并观察其对视频生成质量的影响。在模型开始整合新的动作模态时，人类对文本到视频和图像到视频的评分最初下降了最多 10%。在 3500 步之后，模型在视频生成任务上恢复了之前的全部质量，同时现在也能预测动作。","每个系列在加入动作预测前都已归一化到自身的质量水平。数值越高越好。","模型必须将动作整合到其输入和输出中——但这样做并没有永久消耗其能力。模型只需学习这种新模态与其现有世界模型的关系。一旦弄明白这一点，原有能力的性能损失就消失了。视频生成与动作预测不需要独立的基础。相同的骨干支撑着两者。","这使得物理人工智能成为 Black Forest Labs 路线图的自然延伸，而不是方向的改变。内容创作是我们多模态 FLUX 3 骨干在图像、视频和音频上的功能。物理人工智能是其在动作上的应用。一个基础模型，以视觉智能为核心，支持两类应用。我们没有构建一个独立的基础模型。我们专注于难点：构建一个理解世界的模型。在其中行动，是这种理解所使能的。","当我们将 FLUX 3 骨干指向实际生产线上的真实自动化任务会发生什么？","这是mimic和BFL创建FLUX-mimic来回答的问题。mimic构建自己的机器人，并在机器人学习、灵巧操控和生产部署方面拥有专业知识；BFL构建视觉基础模型，并提供多模态训练和建模专业知识。我们共同开发了适用于工业需求并集成到mimic全栈部署系统中的新一代通用操控模型。FLUX-mimic是基于FLUX 3骨干的动作视频模型。","我们的理论是，FLUX 3必须学习世界的内部表示才能生成视频。FLUX-mimic遵循这一理论，并从FLUX骨干学习到的世界表示中解码动作。这种方法最早由mimic-video开创，在视频预测路径提取的中间特征基础上训练轻量级动作解码器。","架构概览：FLUX-mimic是如何构建在FLUX 3之上的","这种方法的成功取决于两个相关但不同的方面：FLUX学习到的世界模型的质量以及该世界模型特征表示的质量。世界模型的质量直接关系到生成质量：如果模型不理解世界的行为，就无法模拟它。然而，即使是最好的世界模型也无法帮助动作解码器，如果它不可访问：如果特征空间在模态之间的因果关系上保持非线性纠缠，那么从特征表示理解这些关系仍然和从原始输入理解它们一样困难——表示质量至关重要。","生成质量和表示质量长期以来一直是分开研究和处理的。生成方法能够产生高质量的世界模型，使模拟成为可能，并且它们在计算投资上表现出可预测的规模规律。然而，与更专门的表示学习方法相比，它们产生的表示较不易解耦，这限制了其在需要世界理解的任务中的用途。","由于生成模型本身依赖于它们自己的特征，它们在表示质量上的差异似乎有些不合常理。生成模型中表示的改进应当提升其世界模型的质量，并使其在下游任务中更可用。在我们的工作《Self-Flow：https://bfl.ai/research/self-flow》中，我们展示了如何在单一框架中统一生成与表示学习，并正是观察到了这种相互促进的改进：世界模型得到了提升——通过视频、图像和音频生成质量来衡量——其表示质量也提升了——通过机器人控制任务在仿真中的成功率来衡量。","Self-Flow 与 Flow Matching (FM) 对比。左图：每种模态的生成误差（Fréchet距离），均归一化为 FM=100（数值越低越好）。右图：在微调后四组任务中的操作任务成功率平均值（数值越高越好）。","Self-Flow 的扩展规律依旧成立，而 FLUX 3 是这一规律的应用：Self-Flow 的扩大版本。它在数千万小时的一般视频内容上进行训练，从第一天起就尽可能广泛地学习世界动态，并在数十万小时的以人类和机器人操作任务为重点的视频内容上训练，以准备作为视觉智能的核心。这种规模化将 Self-Flow 的实验室成功转化为现实。模拟实际工厂应用中 FLUX-mimic：覆盖生产和物流工作的日常场景——将零件放入结构化托盘，将电子控制单元插入紧密配合的夹具中，组装组件，以及处理传统自动化无法接触的柔软、灵活材料如密封件和电缆。","基准测试显示，即使 FLUX 骨干网络完全冻结，动作解码器也超过了以往的视觉-语言-动作模型——在这种设置下，以往模型无法成功。这凸显了规模化赋予我们的骨干网络丰富的世界知识及操作能力，以及 Self-Flow 如何使这些知识可以从骨干网络的特征表示中轻松解码。当骨干网络与动作解码器一起进行微调时，FLUX-mimic 达到最先进的成功率。","虚线表示每个模型在 20 次自主试验中的中位成功率。数值越高越好。","一个在可解码表示中暴露世界知识的骨干网络，会改变教机器人新任务所需的方式。如果物理知识已经包含在表示中并且易于获取，适应一个任务就不再是教模型世界如何运作——它只需学习这个特定任务如何映射到它已知的知识上。昂贵的部分在机器人开始移动之前就已经完成。","这一点直接体现在新任务所需的示范数据量上。在我们的 Self-Flow 实验中，与没有 Self-Flow 的视频模型相比，动作预测在训练步数上只需一半即可达到相同的成功率——更好的表示使世界知识更容易提取，因此达到相同能力所需的数据更少。mimic-video 论文报告视频动作模型比视觉-语言-动作模型的样本效率高达 10 倍；FLUX-mimic 结合了这两种效果。","同样的好处也体现在行为上。FLUX-mimic 天然能够从失败中恢复：机器人抓取失败后会自我纠正，再次抓取并完成任务。任何示范集都不可能覆盖任务可能出错的每种情况。未曾示范的恢复必须来源于别的地方——来自一个已经知道世界如何运作的模型。","骨干网络预测的未来（上图），以及机器人从解码动作产生的展开结果（下图）。","现实世界部署提出了一个硬性约束：模型必须像世界一样快速行动。FLUX-mimic 的主要计算成本在骨干网络中。它是模型中最大的组成部分，在 mimic 优化的部署堆栈中，其延迟实际上决定了整个系统的上限。","这也是我们方法论第二次体现价值的地方。更好的表示意味着每个参数具备更多能力：一个学习了结构良好的世界模型的模型，比未学习的模型达到同一性能水平所需的容量更少。对于部署来说，这直接转化为能够运行一个更小的骨干网络——更小的骨干网络意味着更快的骨干网络。","在100万次训练步骤的CLIP评分；分数越高越好。主干网络的深度是部署延迟的主要驱动因素，因此更少的层意味着模型运行更快。","因此，FLUX-mimic的主干网络可以优化到在单个NVIDIA RTX 5090 GPU上从输入到世界表示的时间少于80毫秒——这与人类视觉反应时间处于同一数量级。","实际部署需要对完整部署堆栈进行额外优化，以避免增加任何额外延迟：mimic 的优化涵盖从动作解码器，到传感器、模型与执行器之间的进程间延迟消减，再到实时分块，使预测和执行重叠，保持机器人平稳运行而无抖动。最终结果是一个反应时间为101毫秒的自包含机器人系统。","所有这些都回到了自动化最重要的地方：工厂。奥迪运营着汽车行业中最自动化的生产网络之一——这让它对传统自动化仍然止步的环节有清晰的认识。尽管机器人技术投资已达数十年，但涉及灵活零件和精细操作的任务仍然依赖人工，这在很大程度上出于经济原因：高端生产的品类多样性使得常规编程的机器人单元在每种情况下重新设计成本过高。基于学习的系统改变了这种计算方式。","“与mimic合作，奥迪一直在测试和部署FLUX-mimic。我们已经看到这些机器人能够完成传统机器人技术根本无法实现的复杂软体操作工作。这对于辅助员工、提高效率，以及在生产和物流操作中扩展灵活自动化来说都具有重大影响。对于我们来说，与mimic和Black Forest Labs等先锋公司合作，对于推动物理人工智能前沿并在实际生产环境中验证这些创新至关重要。”——奥迪生产实验室，Christoph Schneider","FLUX-mimic 是一个专门构建的机器人模型——也是未来发展的一个验证实例。它的样本效率和鲁棒性源自 FLUX 3 的骨干网络以及它所展现的高质量表示，而非特定任务的工程设计。这正是该方法能够跨任务、跨行业和跨硬件进行迁移的原因。通过 mimic 阅读更多内容：https://www.mimicrobotics.com/blog/introducing-flux-mimic。","一个模型，以视觉智能为核心，能够生成图像、视频和音频——并驱动生产线上的机器人。内容创作和物理 AI 是同一基础的两种应用。"],"translationStatus":"translated","bodyOrigin":"source-page","editorial":{"summary":"Black Forest Labs 公开 FLUX 3 早期版本，该模型从训练之初联合处理图像、视频与音频，并与 mimic 合作形成视频动作模型 FLUX-mimic，相关机器人已在奥迪测试和部署。","background":"FLUX 1 和 FLUX 2 主要生成图像，FLUX 3 扩展至联合生成视听内容。材料称，视频预测占总训练算力成本的 95% 以上，而带音频的 720p 视频中，音频所占令牌不足 0.5%。","viewpoint":"Aioga 判断，这次更新的重点并非单纯增加一种生成模态，而是尝试把视频、音频与机器人动作视为同一物理现实的不同表征。不过，“世界模型”式解释目前主要来自发布方论述。","implications":"加入动作预测后，文本到视频和图像到视频的人类评分最初最多下降 10%，训练 3500 步后恢复此前质量并具备动作预测能力。值得关注的是，该结果来自一次大规模训练运行。","nextStep":"后续值得关注 FLUX-mimic 在更多机器人任务中的测试与部署表现，以及视频质量恢复现象能否在不同训练设置下重复。材料尚未提供具体任务成功率、安全指标或与其他模型的对比结果。","evidenceRefs":["title","summary","articleBody","source"],"status":"published","aiGenerated":true,"autoApproved":true,"generatedBy":"aioga-editorial:gpt-5.6-sol","reviewedBy":"aioga-editorial-review:gpt-5.6-sol","generatedAt":"2026-07-26T06:11:27.210Z","sourceHash":"b19862d73839ff2b","review":{"approved":true,"groundedness":96,"clarity":92,"duplicationRisk":18,"blockingIssues":[],"notes":["“公开 FLUX 3 早期版本”可能被理解为已公开模型权重或开放使用；来源仅明确发布方披露了早期版本及其机器人应用，建议可选地改为“披露 FLUX 3 早期版本”。","“Aioga 判断”属于明确标注的编辑观点，不构成事实错误；若面向不了解该名称的读者，可补充其身份或改为“我们认为”。"]},"validation":{"passed":true,"mode":"ai-auto","revisions":0,"checks":["schema","length","source-attribution","low-source-overlap","no-html","independent-ai-review"]}},"tags":["模型更新","Hacker News 热门（buzzing.cc 中文翻译）"],"translations":{"zh-CN":{"title":"FLUX 3 x mimic：新一代视频动作模型","summary":"Black Forest Labs 发布多模态基础模型 FLUX 3，联合训练图像、视频和音频，其中视频预测占训练算力的 95% 以上。该模型与机器人公司 mimic 合作推出 FLUX-mimic，已在奥迪生产线上测试部署。加入动作预测后，视频生成质量最初下降最多 10%，但经 3500 步训练后恢复原有水平。","category":"模型更新","source":"bfl.ai","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic：新一代视频动作模型 - Aioga AI资讯","description":"Black Forest Labs 发布多模态基础模型 FLUX 3，联合训练图像、视频和音频，其中视频预测占训练算力的 95% 以上。该模型与机器人公司 mimic 合作推出 FLUX-mimic，已在奥迪生产线上测试部署。加入动作预测后，视频生成质量最初下降最多 10%，但经 3500 步训练后恢复原有水平。","url":"https://www.aioga.com/news/cmrywwsva05zgrolgrh7uc4ls/"},"en":{"title":"FLUX 3 x mimic: Next-generation video action model","summary":"Black Forest Labs released the multimodal foundation model FLUX 3, jointly training images, videos, and audio, with video prediction accounting for more than 95% of the training compute. The model collaborated with robotics company mimic to launch FLUX-mimic, which has been tested and deployed on Audi's production line. After adding action prediction, the video generation quality initially dropped by up to 10%, but returned to its original level after 3,500 training steps.","category":"Models","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic: Next-generation video action model - Aioga AI News","description":"Black Forest Labs released the multimodal foundation model FLUX 3, jointly training images, videos, and audio, with video prediction accounting for more than 95% of the training co...","url":"https://www.aioga.com/en/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:04:40.796Z"},"ja":{"title":"FLUX 3 x mimic：次世代のビデオアクションモデル","summary":"Black Forest Labsはマルチモーダル基礎モデルFLUX 3を発表し、画像、ビデオ、音声を共同で訓練しました。その中でビデオ予測は訓練計算力の95%以上を占めています。このモデルはロボット会社mimicと協力してFLUX-mimicを展開し、すでにアウディの生産ラインでテスト導入されています。動作予測を追加した後、ビデオ生成の品質は当初最大で10%低下しましたが、3500ステップの訓練後には元のレベルに回復しました。","category":"モデル更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic：次世代のビデオアクションモデル - Aioga AIニュース","description":"Black Forest Labsはマルチモーダル基礎モデルFLUX 3を発表し、画像、ビデオ、音声を共同で訓練しました。その中でビデオ予測は訓練計算力の95%以上を占めています。このモデルはロボット会社mimicと協力してFLUX-mimicを展開し、すでにアウディの生産ラインでテスト導入されています。動作予測を追加した後、ビデオ生成の品質は当初最大で10...","url":"https://www.aioga.com/ja/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:04:51.248Z"},"ko":{"title":"FLUX 3 x mimic: 차세대 비디오 액션 모델","summary":"Black Forest Labs는 멀티모달 기초 모델 FLUX 3을 발표했으며, 이미지, 비디오, 오디오를 공동으로 학습했는데, 그중 비디오 예측이 학습 연산 능력의 95% 이상을 차지합니다. 이 모델은 로봇 회사 mimic와 협력하여 FLUX-mimic을 출시했으며, 아우디 생산 라인에서 테스트 배포되었습니다. 동작 예측이 추가된 후 비디오 생성 품질은 처음에 최대 10% 하락했지만, 3500 스텝 학습 후 원래 수준으로 회복되었습니다.","category":"모델 업데이트","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic: 차세대 비디오 액션 모델 - Aioga AI 뉴스","description":"Black Forest Labs는 멀티모달 기초 모델 FLUX 3을 발표했으며, 이미지, 비디오, 오디오를 공동으로 학습했는데, 그중 비디오 예측이 학습 연산 능력의 95% 이상을 차지합니다. 이 모델은 로봇 회사 mimic와 협력하여 FLUX-mimic을 출시했으며, 아우디 생산 라인에서 테스트 배포되었습니다. 동작 예...","url":"https://www.aioga.com/ko/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:05:32.480Z"},"es":{"title":"FLUX 3 x mimic: modelo de acción de video de nueva generación","summary":"Black Forest Labs lanzó el modelo base multimodal FLUX 3, entrenando conjuntamente imágenes, videos y audio, donde la predicción de videos ocupa más del 95% del poder de cómputo del entrenamiento. Este modelo se lanzó en colaboración con la empresa de robótica mimic como FLUX-mimic, y ya se ha probado y desplegado en la línea de producción de Audi. Tras incorporar la predicción de acciones, la calidad de generación de video inicialmente cayó hasta un 10%, pero después de 3.500 pasos de entrenamiento volvió a su nivel original.","category":"Modelos","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic: modelo de acción de video de nueva generación - Aioga Noticias de IA","description":"Black Forest Labs lanzó el modelo base multimodal FLUX 3, entrenando conjuntamente imágenes, videos y audio, donde la predicción de videos ocupa más del 95% del poder de cómputo de...","url":"https://www.aioga.com/es/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:05:29.890Z"},"fr":{"title":"FLUX 3 x mimic : modèle d'action vidéo de nouvelle génération","summary":"Black Forest Labs a publié le modèle de base multimodal FLUX 3, entraînant conjointement des images, des vidéos et de l'audio, avec plus de 95 % de la puissance de calcul dédiée à la prédiction vidéo. Ce modèle a été lancé en collaboration avec la société de robotique mimic sous le nom de FLUX-mimic et a été testé sur les lignes de production d'Audi. Après l'ajout de la prédiction de mouvement, la qualité de génération vidéo a initialement chuté de 10 % au maximum, mais est revenue à son niveau initial après 3500 étapes d'entraînement.","category":"Modèles","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic : modèle d'action vidéo de nouvelle génération - Aioga Actualités IA","description":"Black Forest Labs a publié le modèle de base multimodal FLUX 3, entraînant conjointement des images, des vidéos et de l'audio, avec plus de 95 % de la puissance de calcul dédiée à...","url":"https://www.aioga.com/fr/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:06:10.094Z"},"de":{"title":"FLUX 3 x mimic: Neue Generation des Videobewegungsmodells","summary":"Black Forest Labs veröffentlicht das multimodale Basismodell FLUX 3, das Bilder, Videos und Audio gemeinsam trainiert, wobei die Videovorhersage mehr als 95 % der Trainingsrechenleistung ausmacht. Das Modell wurde in Zusammenarbeit mit dem Roboterunternehmen mimic als FLUX-mimic auf den Audi-Produktionslinien getestet und eingesetzt. Nach der Hinzufügung der Bewegungsprognose sank die Videoerstellungsqualität zunächst um maximal 10 %, erreichte jedoch nach 3500 Trainingsschritten wieder das ursprüngliche Niveau.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic: Neue Generation des Videobewegungsmodells - Aioga KI-News","description":"Black Forest Labs veröffentlicht das multimodale Basismodell FLUX 3, das Bilder, Videos und Audio gemeinsam trainiert, wobei die Videovorhersage mehr als 95 % der Trainingsrechenle...","url":"https://www.aioga.com/de/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:06:12.819Z"},"pt-BR":{"title":"FLUX 3 x mimic: Novo modelo de ação em vídeo de nova geração","summary":"A Black Forest Labs lançou o modelo fundamental multimodal FLUX 3, treinando conjuntamente imagens, vídeos e áudio, sendo que a previsão de vídeo consome mais de 95% da capacidade computacional de treinamento. Esse modelo, em parceria com a empresa de robótica mimic, deu origem ao FLUX-mimic, que já foi testado e implementado na linha de produção da Audi. Após a adição da previsão de ações, a qualidade da geração de vídeo inicialmente caiu até 10%, mas após 3500 passos de treinamento voltou ao nível original.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic: Novo modelo de ação em vídeo de nova geração - Aioga Notícias de IA","description":"A Black Forest Labs lançou o modelo fundamental multimodal FLUX 3, treinando conjuntamente imagens, vídeos e áudio, sendo que a previsão de vídeo consome mais de 95% da capacidade...","url":"https://www.aioga.com/pt-BR/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:06:48.259Z"},"ru":{"title":"FLUX 3 x mimic: новое поколение моделей для видео-анимации","summary":"Black Forest Labs выпустила мультимодальную базовую модель FLUX 3, которая обучается совместно на изображениях, видео и аудио, при этом предсказания видео составляют более 95% вычислительной мощности обучения. Эта модель была выпущена в сотрудничестве с робототехнической компанией mimic под названием FLUX-mimic и уже протестирована и внедрена на производственной линии Audi. После добавления предсказания действий качество генерации видео сначала снизилось максимум на 10%, но после 3500 шагов обучения вернулось к исходному уровню.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic: новое поколение моделей для видео-анимации - Aioga Новости ИИ","description":"Black Forest Labs выпустила мультимодальную базовую модель FLUX 3, которая обучается совместно на изображениях, видео и аудио, при этом предсказания видео составляют более 95% вычи...","url":"https://www.aioga.com/ru/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:06:54.137Z"},"ar":{"title":"FLUX 3 x محاكاة: الجيل الجديد من نماذج حركة الفيديو","summary":"أصدرت مختبرات الغابة السوداء نموذج FLUX 3 متعدد الوسائط، وهو نموذج أساسي يتم تدريبه بشكل مشترك على الصور والفيديو والصوت، حيث يشغل التنبؤ بالفيديو أكثر من 95٪ من قدرة الحوسبة التدريبية. تعاون هذا النموذج مع شركة الروبوتات mimic لإطلاق FLUX-mimic، وقد تم اختباره ونشره على خط إنتاج أودي. بعد إضافة التنبؤ بالحركة، انخفضت جودة إنتاج الفيديو في البداية بنسبة تصل إلى 10٪، لكنها عادت إلى المستوى الأصلي بعد 3500 خطوة تدريب.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x محاكاة: الجيل الجديد من نماذج حركة الفيديو - Aioga أخبار الذكاء الاصطناعي","description":"أصدرت مختبرات الغابة السوداء نموذج FLUX 3 متعدد الوسائط، وهو نموذج أساسي يتم تدريبه بشكل مشترك على الصور والفيديو والصوت، حيث يشغل التنبؤ بالفيديو أكثر من 95٪ من قدرة الحوسبة التدر...","url":"https://www.aioga.com/ar/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:07:33.619Z"},"hi":{"title":"FLUX 3 x mimic: नई पीढ़ी का वीडियो एक्शन मॉडल","summary":"Black Forest Labs ने मल्टीमॉडल बेस मॉडल FLUX 3 जारी किया, जो इमेज, वीडियो और ऑडियो का संयुक्त प्रशिक्षण करता है, जिसमें वीडियो पूर्वानुमान प्रशिक्षण कंप्यूटिंग पावर का 95% से अधिक हिस्सा लेता है। इस मॉडल को रोबोट कंपनी mimic के साथ मिलकर FLUX-mimic के रूप में लॉन्च किया गया, जिसे ऑडी की प्रोडक्शन लाइन पर परीक्षण और तैनाती की जा चुकी है। मूवमेंट प्रेडिक्शन जोड़ने के बाद वीडियो जनरेशन की क्वालिटी शुरू में अधिकतम 10% तक कम हुई, लेकिन 3500 स्टेप्स के प्रशिक्षण के बाद यह अपने मूल स्तर पर लौट आई।","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic: नई पीढ़ी का वीडियो एक्शन मॉडल - Aioga AI समाचार","description":"Black Forest Labs ने मल्टीमॉडल बेस मॉडल FLUX 3 जारी किया, जो इमेज, वीडियो और ऑडियो का संयुक्त प्रशिक्षण करता है, जिसमें वीडियो पूर्वानुमान प्रशिक्षण कंप्यूटिंग पावर का 95% से अधिक...","url":"https://www.aioga.com/hi/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:07:41.293Z"},"it":{"title":"FLUX 3 x mimic: nuovo modello di azione video di nuova generazione","summary":"Black Forest Labs ha rilasciato il modello di base multimodale FLUX 3, addestrando congiuntamente immagini, video e audio, con le previsioni video che rappresentano oltre il 95% della potenza di calcolo dell'addestramento. Questo modello è stato lanciato in collaborazione con l'azienda di robotica mimic come FLUX-mimic, e già testato e implementato sulla linea di produzione Audi. Dopo aver aggiunto la previsione dei movimenti, la qualità della generazione video inizialmente è diminuita fino al 10%, ma dopo 3500 passaggi di addestramento è tornata al livello originale.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic: nuovo modello di azione video di nuova generazione - Aioga Notizie IA","description":"Black Forest Labs ha rilasciato il modello di base multimodale FLUX 3, addestrando congiuntamente immagini, video e audio, con le previsioni video che rappresentano oltre il 95% de...","url":"https://www.aioga.com/it/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:08:20.740Z"},"nl":{"title":"FLUX 3 x mimic: nieuwe generatie videobewegingsmodel","summary":"Black Forest Labs heeft het multimodale basismodel FLUX 3 uitgebracht, dat gelijktijdig beelden, video en audio traint, waarbij videovoorspelling meer dan 95% van het rekenvermogen tijdens het trainen verbruikt. Het model is in samenwerking met het robotbedrijf mimic uitgebracht als FLUX-mimic en is getest en ingezet op de productielijn van Audi. Na toevoeging van bewegingsvoorspelling daalde de videoproductiekwaliteit aanvankelijk met maximaal 10%, maar na 3500 trainingsstappen herstelde deze zich tot het oorspronkelijke niveau.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic: nieuwe generatie videobewegingsmodel - Aioga AI-nieuws","description":"Black Forest Labs heeft het multimodale basismodel FLUX 3 uitgebracht, dat gelijktijdig beelden, video en audio traint, waarbij videovoorspelling meer dan 95% van het rekenvermogen...","url":"https://www.aioga.com/nl/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:08:19.083Z"},"tr":{"title":"FLUX 3 x mimic: Yeni nesil video hareket modeli","summary":"Black Forest Labs, görüntü, video ve sesi birlikte eğiten çok modlu temel modeli FLUX 3'ü yayınladı; bunun içinde video tahmini, eğitim hesap gücünün %95'inden fazlasını oluşturuyor. Bu model, robotik şirketi mimic ile iş birliği içinde FLUX-mimic'i piyasaya sürdü ve Audi üretim hattında test edilip uygulandı. Hareket tahmini eklendikten sonra, video üretim kalitesi başlangıçta en fazla %10 düşse de, 3500 adım eğitim sonrası eski seviyesine geri döndü.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic: Yeni nesil video hareket modeli - Aioga AI Haberleri","description":"Black Forest Labs, görüntü, video ve sesi birlikte eğiten çok modlu temel modeli FLUX 3'ü yayınladı; bunun içinde video tahmini, eğitim hesap gücünün %95'inden fazlasını oluşturuyo...","url":"https://www.aioga.com/tr/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:08:59.673Z"},"vi":{"title":"FLUX 3 x mimic: Mô hình hành động video thế hệ mới","summary":"Black Forest Labs ra mắt mô hình cơ bản đa phương thức FLUX 3, được huấn luyện kết hợp hình ảnh, video và âm thanh, trong đó dự đoán video chiếm hơn 95% công suất huấn luyện. Mô hình này hợp tác với công ty robot mimic ra mắt FLUX-mimic, đã được thử nghiệm triển khai trên dây chuyền sản xuất của Audi. Khi bổ sung dự đoán hành động, chất lượng tạo video ban đầu giảm tối đa 10%, nhưng sau 3500 bước huấn luyện đã phục hồi về mức ban đầu.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic: Mô hình hành động video thế hệ mới - Tin tức AI Aioga","description":"Black Forest Labs ra mắt mô hình cơ bản đa phương thức FLUX 3, được huấn luyện kết hợp hình ảnh, video và âm thanh, trong đó dự đoán video chiếm hơn 95% công suất huấn luyện. Mô hì...","url":"https://www.aioga.com/vi/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:08:58.004Z"},"id":{"title":"FLUX 3 x mimic: Model gerakan video generasi baru","summary":"Black Forest Labs merilis model dasar multimodal FLUX 3, melatih gambar, video, dan audio secara gabungan, dengan prediksi video menyumbang lebih dari 95% dari daya komputasi pelatihan. Model ini bekerja sama dengan perusahaan robotik mimic untuk meluncurkan FLUX-mimic, yang telah diuji dan diterapkan di lini produksi Audi. Setelah penambahan prediksi aksi, kualitas pembuatan video awalnya turun hingga 10%, tetapi setelah 3500 langkah pelatihan kembali ke tingkat semula.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic: Model gerakan video generasi baru - Berita AI Aioga","description":"Black Forest Labs merilis model dasar multimodal FLUX 3, melatih gambar, video, dan audio secara gabungan, dengan prediksi video menyumbang lebih dari 95% dari daya komputasi pelat...","url":"https://www.aioga.com/id/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:09:37.747Z"},"th":{"title":"FLUX 3 x mimic: โมเดลเคลื่อนไหววิดีโอรุ่นใหม่","summary":"Black Forest Labs เปิดตัวโมเดลพื้นฐานมัลติชัดเจน FLUX 3 ซึ่งฝึกอบรมร่วมกับภาพ วิดีโอ และเสียง โดยที่การทำนายวิดีโอคิดเป็นกำลังการฝึกมากกว่า 95% โมเดลนี้ร่วมมือกับบริษัทหุ่นยนต์ mimic เปิดตัว FLUX-mimic และได้ทดสอบปรับใช้ในสายการผลิตของ Audi หลังจากเพิ่มการทำนายการเคลื่อนไหว คุณภาพการสร้างวิดีโอในตอนแรกลดลงสูงสุด 10% แต่หลังจากการฝึก 3,500 ขั้นตอนกลับมาที่ระดับเดิม","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic: โมเดลเคลื่อนไหววิดีโอรุ่นใหม่ - ข่าว AI Aioga","description":"Black Forest Labs เปิดตัวโมเดลพื้นฐานมัลติชัดเจน FLUX 3 ซึ่งฝึกอบรมร่วมกับภาพ วิดีโอ และเสียง โดยที่การทำนายวิดีโอคิดเป็นกำลังการฝึกมากกว่า 95% โมเดลนี้ร่วมมือกับบริษัทหุ่นยนต์ mim...","url":"https://www.aioga.com/th/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:09:42.900Z"},"pl":{"title":"FLUX 3 x mimic: nowa generacja modelu akcji wideo","summary":"Black Forest Labs wydało multimodalny model bazowy FLUX 3, wspólnie trenując obrazy, wideo i audio, przy czym przewidywanie wideo stanowi ponad 95% mocy obliczeniowej użytej do treningu. Model we współpracy z firmą robotyczną mimic został wprowadzony jako FLUX-mimic i przetestowany na liniach produkcyjnych Audi. Po dodaniu przewidywania akcji jakość generowanego wideo początkowo spadła maksymalnie o 10%, ale po 3500 krokach treningu wróciła do pierwotnego poziomu.","category":"模型更新","source":"Hacker News 热门（buzzing.cc 中文翻译）","aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"FLUX 3 x mimic: nowa generacja modelu akcji wideo - Aioga Wiadomości AI","description":"Black Forest Labs wydało multimodalny model bazowy FLUX 3, wspólnie trenując obrazy, wideo i audio, przy czym przewidywanie wideo stanowi ponad 95% mocy obliczeniowej użytej do tre...","url":"https://www.aioga.com/pl/news/cmrywwsva05zgrolgrh7uc4ls/","contentTranslated":true,"sourceHash":"1b21191ead8ef6bf","translatedAt":"2026-07-26T02:10:23.012Z"}}}}