文 | 奇点研究社，作者｜徐婷，编辑｜奇博士
10月30日深夜，快手在港交所发布公告：旗下视频生成业务可灵完成增资交割，198.02 亿元增资款全部到账，约合 29.05 亿美元，可灵投后估值约 179 亿美元，折合人民币超过 1200 亿元，高于快手当日约 968 亿元的港股收盘市值。
“一个可灵，比整个快手还贵”，尤其考虑到，2026 年上半年，可灵收入占快手总收入的比例仅为 2.17%。资本市场买的显然不是可灵当下的收入，而是它在 AI 视频赛道里的未来。
资本故事讲得再大，最终仍要回到创作现场接受检验。最近，我们用可灵4.0制作一条探展视频，真人出镜环节全部交给 AI，采用展会实拍素材与 AI 人物结合的方式，拆分生成 7 组画面，再通过音画同出、视频修改和时间线等功能完成后续制作。测试前，我们准备了约 3000 字提示词，尽可能明确主体关系、镜头运动、构图和声音要求，减少指令歧义。
实测结果喜忧参半，个别镜头的人物融合、运镜和构图已经相当自然，几乎看不出明显违和；但另一些片段里，AI 会把台上的机器人换成台下的主播，把一镜到底变成几次硬切，甚至在修改视频后，将原本正常的声音变成一串难以辨认的乱码。
AI已经能生成不少惊艳镜头，为什么成片仍要反复试错？可灵4.0距离“一站成片”，还差哪些关键能力？在180亿美元估值与行业资源重组之下，稳定交付能否成为AI视频的商业分水岭？
可灵4.0再升级，从生成画面到执行创意
在国内互联网公司的大模型竞争中，快手选择了一条相对聚焦的路线：没有将主要筹码押在通用聊天机器人上，而是持续投入视频生成。这与快手原有的业务生态密切相关。广告、电商、短剧、影视和创作者内容，都有大量视频生产需求，模型能力也有机会通过会员订阅、企业服务和 API 调用转化为收入。
这是一门离钱更近，也更烧钱的生意。视频生成的付费场景相对直接，但训练、推理、算力和人才，每一项都价格不菲。可灵由此从快手内部的技术项目，逐渐走向公司的战略中心。
这种权重变化，清晰地写在组织与资本动作里。2025年4月，快手设立可灵 AI事业部，负责人直接向程一笑汇报，7月可灵推进独立分拆融资安排，融资规模近30亿美元，投后估值约180亿美元，腾讯、阿里云、百度以及多家地方产业投资基金和文娱产业方进入股东名单。9月30日，原社区科学线负责人于越转任可灵AI公司董事兼 CEO，负责经营管理。10月30日，这轮独立融资正式完成交割，实际到账增资款约 198.02 亿元，折合约 29.05 亿美元。
从一级事业部到独立融资，再到专设 CEO，可灵正在获得一家独立公司的资源配置，也开始面对一家独立公司的经营考题。
商业化数据是它眼下最有分量的筹码。根据快手披露的数据，2026 年第一季度，可灵 AI 收入超过 6.5 亿元，同比增长超过 300%；第二季度收入超过 8.5 亿元，同比增长超过 200%，上半年合计超过 15 亿元。
但收入增长，并不等于商业模式已经高枕无忧。随着业务独立性增强，它还需要回答几个更现实的问题，诸如收入增长能否延续，算力成本如何消化，企业客户是否愿意持续续费，以及产品能否建立足够稳固的竞争壁垒等等。
关于可灵筹备赴港 IPO 的消息已见诸报端，不过相关安排仍处于早期阶段，上市时间和规模均存在调整空间。资本给了可灵更大的空间，也抬高了它的及格线。接下来，市场追问的不只是模型还能生成什么，更是技术高光能否变成持续回报。
可灵 4.0正是在这样的背景下到来，也是可灵尝试向专业创作流程再迈进一步的产品节点。
根据快手官方信息，可灵4.0升级重点围绕真实、可控和专业展开：单次原生生成时长最长可达30秒；图片、视频、主体与音色等参考素材可以灵活组合，最多支持15项参考素材和 10 张关键帧；提示词长度上限也进一步提高。官方希望通过这些升级，更好地处理高动态场景、复杂运镜和连续叙事。
这些能力变化对应着AI视频行业评价标准的转变。早期模型首先要证明自己能否把文字描述转换成一段画面：人物动作是否自然，画面是否逼真，光影是否接近真实摄影。随着生成质量提升，创作者开始提出更复杂的要求：人物跨镜头要保持一致，镜头要按指定的景别、构图和节奏移动，多个片段之间还要维持基本的空间关系和叙事连贯性。
模型需要处理的除了“画面里有什么”，还包括素材各自承担什么作用、哪些元素可以改变、哪些信息必须保留，以及创作者希望镜头如何展开，可灵 4.0 对参考素材、运镜和生成时长的升级，正是在回应这些需求。
与模型能力同步演进的还有产品形态。可灵逐步将生成、编辑、画布和时间线等能力整合到同一工作空间，希望减少用户在不同工具之间切换的成本。字节跳动 Seed 团队此前推出的 Seedance 系列，也把长叙事、多模态参考和视频编辑作为产品迭代的重要方向。
各家的实现方式不尽相同，但竞争目标逐渐清晰：创作者期待的已不只是一个能生成片段的按钮，还包括一套能协助完成作品的工具。
如果说AI视频的上半场拼的是生成能力，下半场考验的则是对创作意图的执行能力。但更长的生成时长、更多的参考素材和更丰富的编辑能力，是否就意味着创作者能更准确地控制最终结果？
奇点用一条探展视频进行了一番实测，展会现场是实拍素材，AI 人物需要进入既有环境，完成不同的镜头运动；多个片段之间还要维持人物和场景关系，并与口播、声音及后期处理衔接起来。
我们想看看，一条没有真人出镜的探展视频，从素材整理、镜头生成到配音配乐，究竟能被 AI 包揽多少？
可灵4.0一手实测，好镜头与好成片之间，差几次“重来”
这次测试全程使用可灵 4.0，视频采用实拍素材与 AI 人物结合的方式，拆分为 7 组画面，并在画布中完成生成、修改和素材整理。为了尽可能减少指令歧义，我们借助 AI 助手和团队剪辑同事整理了约 3000 字提示词，明确主体关系、构图、景别、运镜和声音要求。
换句话说，这不是一次完全放任模型的随机测试，而是一场有专业辅助、有明确交付目标的实战。
可灵 4.0 的上限，确实足够惊喜。表现最好的几段素材，能够准确完成提示词要求的镜头推移，构图、景别和切换节奏也基本符合预期。更难得的是，AI人物与实拍环境的融合相当自然，人物没有明显“贴”在背景上的感觉，与环境互动时也不显得突兀。
这类画面已经接近可以直接交付的状态。它说明可灵 4.0 并不只是能生成几段适合展示的样片，在条件合适时，确实能够产出具有专业使用价值的内容。
声音也是本次测试中相对亮眼的部分，大部分环境声和口播可以直接使用，整体可用率高于画面，音画同出减少了逐段配音的步骤，对于需要快速完成短片的创作者，具有明确的效率价值。
但高光表现并不能稳定复现。原因一是模型对主体与场景关系的处理仍会出现偏差。在一次生成中，原本应该站在台下的主播，直接替换掉了台上的机器人；
另一段素材里，AI 又用角色人物换掉了背景视频中的主持人。即使补充“围观”等关系描述，结果也没有明显改正。还有片段出现人物与场景比例失衡，让正常探展突然有了“误闯小人国”的观感。
这些bug的核心不是画质，而是模型是否真正理解画面中“谁是谁、谁在哪里、谁与谁互动”。
其次是长镜头与连续叙事。我们专门设计了一组一镜到底的指令，但几次尝试后，得到的仍是不同场景之间的硬切，人物虽然一直在走，镜头却没有形成连续运动，周围环境也被大幅修改，提示词中对素材使用的限制没有被稳定执行。时长增加，并不自然等于叙事能力升级，模型可以生成更长的内容，却未必能持续维持空间、人物和镜头关系。
最后是修改与声音的一致性。部分片段经过视频修改后，不仅没有按指令修正，还擅自添加字幕；原本正常的声音，也可能在后续生成中变成乱码，提前为人物主体配置的声音，没有被稳定引用，不同片段中的同一个人物，声线也会出现差别。画布和时间线确实改善了素材管理，但 BGM、转场、音效与修改流程，仍未形成足够顺畅的成片闭环，任何一处出现问题，都要“推到重来”。
这也是本次实测最直接的感受，可灵4.0已经抬高了好镜头生成的上限，但稳定交付一条符合要求的视频，仍然需要创作者在关键环节兜底。
这期间产生的成本很容易被单次生成的价格掩盖，提示词调试、反复生成、筛选素材、失败修改、等待时间，以及人工检查和修补，共同构成了视频真正的生产成本。对个人创作者而言，它们消耗的是时间和精力；对广告、短剧和影视团队来说，它们还会影响工期、预算和交付验收。
可灵分拆，小云雀并入即梦，AI视频下半场换了打法
从实测效果来看，可灵 4.0 已经能在部分场景生成具有专业使用价值的画面，但稳定执行复杂要求、在修改中保留有效内容、顺畅完成整条生产流程，仍有提升空间。这些问题也关系到 AI 视频行业下一阶段的竞争焦点。
可灵、Seedance、Vidu、海螺和通义万相等产品，正在围绕画面质量、人物一致性、连续叙事、生成成本和垂直场景持续迭代。单次版本升级可以带来阶段性的能力变化，随着基础生成能力不断提升，单看某个样片是否惊艳，越来越难以完整衡量一款产品的实际价值。
可灵的商业化进展已有较明确的公开数据，收入增长也让它成为视频生成领域值得持续观察的玩家。不过，收入规模、用户增长、盈利能力和长期竞争力是不同维度，没有同口径的竞品数据，也很难据此断言它在商业化上已经全面领先。
竞争同样发生在应用生态上。字节拥有 Seedance、即梦、剪映、抖音和火山引擎等不同环节的产品与服务；快手也有内容平台和快影。可灵在获得更独立的融资与经营空间后，既需要继续与母公司协同，也需要面向外部开发者和企业客户建立自己的产品生态。
最近有报道称字节旗下 AI 短剧创作平台小云雀并入即梦，产品仍保留独立入口，相关短剧能力已接入即梦。这与可灵独立融资形成了有趣的对照，一边为单一业务争取独立经营空间，另一边将相近产品的能力与运营进一步整合，两种组织选择并不能直接证明哪条路更优，也不足以代表整个行业的统一趋势，但都指向同一个现实问题：企业需要重新判断，哪些能力应该单独发展，哪些产品应当整合，哪些环节能够创造持续的商业价值。
当越来越多第三方创作平台可以接入不同的视频模型，单纯拥有生成能力，就不一定能成为某款应用的独占优势。用户可以比较不同工具的效果、速度和价格，也可以按照任务更换模型。应用要留下用户，需要提供模型之外的价值：更适合某种创作任务的工作流、更精细的编辑与协作能力、更可靠的内容管理，或与分发和商业转化的直接连接。
因此，AI视频产品的价值，可以从三个相互关联的维度重新衡量。
首先是生成稳定性。同样的任务，能否较稳定地生成符合要求的结果？成功画面是偶然出现，还是能够在相似条件下反复复现？对专业用户而言，一次惊艳的结果固然有用，但稳定性才决定它能否成为生产流程里的常规工具。
其次是修改的可控性。当人物、动作或声音出现偏差，工具能否在保留其他有效内容的前提下完成修正？如果每次修改都有可能引入新的错误，模型看似提供了更多编辑能力，实际却可能增加返工。
最后是综合的交付成本。除了模型生成所需的时间和积分，还应计算提示词准备、筛选素材、失败重试、人工修复以及音画后期的投入。把这些成本放在一起，才能判断一条可用视频究竟需要多少资源，也才能比较不同工具在真实工作流中的效率。
这三个维度进一步连接着商业回报。稳定的生成和修改能力，可以减少重复劳动；更顺畅的工作流，可以让模型进入广告、短剧、影视等专业生产场景；可预期的单位交付成本，则影响客户是否愿意长期使用和续费。
对于模型公司来说，用户规模、ARR 和融资估值依然重要，但企业续费、API 使用、算力成本和最终利润，同样决定增长能否持续。
在这一过程中，混合使用多个模型和工具，很可能长期存在。创作者可以用一个模型生成画面，用另一个工具处理声音，再通过专业软件完成剪辑与后期，模型厂商要努力争取让自己的能力进入用户反复运行的生产流程。
180 亿美元估值为可灵的技术潜力和商业前景定价，也为其后续发展提供了空间，4.0 的能力升级展示了它在专业创作方向上的努力。接下来需要继续验证的，是这些技术能力能否转化为更稳定的结果、更可控的修改过程，以及客户能够接受的交付成本。
好镜头吸引人，稳定交付留住人。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Wen, |, 奇, 奇, |, 婷, 博士, 博士, 博士, 博士, 博士.
Late on the night of October 30th, the expressman at the Hong Kong Treasury publish announcement: The video-generation operation Kling completed the replenishment and the full replenishment of $19,802 million was received, amounting to approximately US$ 2,905 million. The Kling post-investment valuation was approximately US$ 17.9 billion, equivalent to more than US$ 120 billion, which is higher than the market value of the HK stock collection of approximately US$ 96.8 billion by the expresser.
“A Kling is more expensive than an entire hand”, particularly considering that in the first half of 2026, Kling's income represented only 2.17 per cent of the total income of the hand. The capital market apparently bought not the current income of Kling, but its future in the AI video track.
Capital stories go so far as to finally go back to the creation scene for testing. Recently, we used Kling4.0 to produce an exhibition video, all of which was given to AI, and by the way split was actually combined with AI, split produced seven sets of images, and then finished the next production with the sound drawings, video modifications and time lines. Before the test, we prepared approximately 3000 words prompt, identifying as much as possible the main relationship, shot motion, graphs and sound requirements to reduce the ambiguity of the instructions.
The results are mixed, with individual shot characters having a natural combination, mirrors and images, which hardly appear to be a clear contradiction; but in other clips, AI replaces the robots on the stage with the anchors, turning the mirrors into a few pieces, even after modifying the video, into an incognito code.
I've been able to create a lot of amazing shots. Why do you have to try it again and again? Kling4.0 is “one stop in one piece”, what are the key capabilities? Under the $18 billion valuation and restructuring of industry resources, could stable delivery become a commercial watershed for AI video?
Kling4.0 Upgrade from generation to implementation of creativity
In the competition for LLM of the Internet company in the country, the fast-tracker chose a relatively focused route: instead of placing the main chips on general chat robots, it continued to feed into video production. This is closely related to the pre-existing operational ecology. Advertising, electronics, short dramas, video and creator content are in large demand for video production, and modelling capacity has the opportunity to be translated into income through member subscriptions, business services and API calls.
It's a business that's closer to money and more expensive. The video-generated paid scene is relatively straightforward, but each of it is expensive to train, reason, calculator and talent. Kling thus moved from fast-track in-house technical projects to the strategic centre of the company.
This shift in weight is clearly written in organizational and capital actions. In April 2025, the Ministry of Kling AI was set up, with the head reporting directly to Cheng Yi, and in July Kling promoted independent split financing arrangements amounting to nearly $3 billion, with a post-investment valuation of about $18 billion. On 30 September, the former head of the community science line was transferred to the board of directors and CEO of Kling AI, in charge of management. On 30 October, the independent financing round was officially completed, resulting in an actual replenishment of approximately 198.2 million yuan, or $2.905 billion.
From the first level of business to the independent financing, to the dedicated CEO, Kling is acquiring the resources of an independent company and is also beginning to face the business test of an independent company.
Commercialization data are the most significant leverage it has at the moment. According to fast-disclosed data, in the first quarter of 2026, Kling AI earned more than $650 million, representing an increase of more than 300 per cent over the same period; in the second quarter, revenues exceeded $850 million, representing an increase of more than 200 per cent over the same period and a total of more than $1.5 billion during the first half of the year.
But income growth does not mean that business models are already high. As business independence increases, it also needs to answer a few more realistic questions, such as whether income growth can be sustained, how the cost of calculation can be absorbed, whether business clients are willing to sustain their fees, and whether products can create sufficiently strong barriers to competition.
Information on Kling ' s preparations for a visit to Hong Kong IPO is available in the press, although the arrangements are still in their early stages and there is room for adjustment both in listing time and size. Capital gave Kling greater space and raised its pass line. Next, the market asks not only what models can generate, but also whether technological light can be translated into sustainable returns.
It is against this background that Kling 4.0 is also the product node where Kling is trying to move further towards a professional creation process.
According to fast-track official information, Kling4.0 upgrades are focused on real, manageable and professional: single birth generation can last up to 30 seconds; assets, for example, for pictures, videos, subjects and voice, can be assembled flexibly, supporting up to 15 reference assets and 10 key frames; and prompt ceilings are further increased. Through these upgrades, the Government hopes to better deal with high-dynamic scenarios, complex mirrors and continuous narratives.
These changes in competencies correspond to changes in the evaluation criteria for the AI video industry. The early model begins with proof of whether it can translate the narrative into a picture: Whether the person's movement is natural, whether the picture is authentic or whether the light is close to real photography. As the quality of generation increases, the creators begin to make more complex demands: people cross shot to be consistent, shot to move according to a given view, image and rhythm, and to maintain basic spatial relations and narrative coherence between multiple segments.
Models need to be addressed in addition to “what's in the picture”, including the role of asset, what elements can be changed, what information must be kept, and how the creators want shot to proceed. Kling 4.0 The upgrading of reference asset, mirrors and generation time frame responds to these needs.
There are also product patterns that evolve in tandem with model capabilities. Kling gradually integrates capabilities such as generation, editing, canvasing and timeline into the same workspace in the hope of reducing the cost of switching between different tools. The byte Jump Seedance series, previously launched by the byte team, also uses long narratives, multi-mode references and video editing as important directions for product iterative.
The way in which they are achieved varies, but the objective of competition is increasingly clear: what creators expect is more than just a button that can generate a clip, including a set of tools that can assist in the completion of their work.
If the first half of the AI video is a capacity to generate, the second half is an ability to execute creative intent. But is a longer generation, more reference to asset and more editorial capacity meant that creators could control the end result more accurately?
The site of the event was a live video of the discovery of asset, and the AI person needed to enter the established environment to complete the different shot movement; several segments also had to maintain the relationship between the person and the scene and be linked to the broadcast, the sound and the processing of post-production.
We'd like to see how much I'd like to get from asset to shot to voiceover.
Kling4.0's hand-to-hand, between a good shot and a good piece, there's a few "replays" missing.
The entire test was conducted using Kling 4.0, the video was taken in the form of a real combination of asset and AI characters, and split was a set of 7 images, which were produced, modified and collated in a canvas. In order to minimize the ambiguity of the instructions, we have been able to organize around 3000 words prompt with the help of AI ' s assistant and team editing colleagues, specifying the main relationships, the structure, the scene, the mirror and the sound requirements.
In other words, this is not a random test of a completely permissive model, but a battle with professional support and clear delivery objectives.
The ceiling of Kling 4.0 is indeed enough surprise. The best-performing segments of asset were able to accurately complete the shot transfer required by prompt, and the maps, scenes and tempo were generally in line with expectations. Even more rare is the natural mix of AI figures with the live environment, where people do not feel clearly “posted” in the context and interact with the environment.
Such images are close to the state of direct delivery. It states that Kling 4.0 is not just capable of producing several pieces of suitable samples that can actually produce content of professional use when the conditions are right.
Sound is also a relatively bright part of this test, with most environmental sound and oral broadcasts being directly used, overall availability being higher than the picture, and sound drawings being combined with steps to reduce the number of out-of-section voiceovers, with clear efficiency value for creators who need to quickly complete short films.
But high-light performance is not stable. One reason is that the model still differs in the treatment of the relation between the subject and the scene. In a single generation, the robots on the stage were replaced directly by the anchor that should have been under the stage;
In another asset, AI replaced the character with the host in the background video. Even if the description of relationships, such as “spanning”, was added, the result was not clearly corrected. There is also an imbalance between the number of people and the number of people on the scene, which has suddenly given the view of a normal tour of the country “by mistake”.
The core of these bugs is not paint, but whether the model really understands “who, where and who interacts with” in the picture.
This was followed by long shot and successive narratives. We have designed a set of one-stop instructions, but after several attempts, we still get a hard cut between the different scenes, while the characters are walking, shot has not developed a continuous movement, the surrounding environment has been significantly altered, and the restrictions on asset in prompt have not been applied steadily. The increase in time does not automatically amount to an upgrade in narrative capability, and models can generate longer content without necessarily continuing the relationship between space, character and shot.
Finally, there is consistency between changes and voices. Some of the clips, modified by video, not only were they not amended by order, but subtitle were added without permission; the original sound, which was normal, may have become random in subsequent generation, with the voices assigned to the subject of the person in advance, were not consistently quoted, and the same person in the different segments, the tone of the sound, was different. While it is true that the canvass and time lines have improved asset management, the BGM, the switch, the sound and the modification processes have not yet formed a sufficiently smooth closed circle, and any problem has to be “reduced back”.
This is also the most straightforward experience of this exercise, where Kling4.0 has raised the ceiling of shot, but the steady delivery of a timely video still requires the creator to go to the bottom of the critical link.
The costs incurred during this period are easily masked by the single-generated price, and prompt debugging, recurring generation, screening asset, failure modification, waiting time, and manual inspection and repair together constitute the real production cost of the video. For individual creators, they consume time and energy; for advertising, short play and video teams, they also affect the duration, budget and delivery of acceptance check.
Kling split, Little Skylark merged into Jimeng.
In terms of results, Kling 4.0 has been able to generate a picture of professional utility in part of the scene, but there is still room for improvement to stabilize complex requirements, retain valid content in modifications and complete the entire production process smoothly. These issues also relate to the next phase of competition in the AI video industry.
The equivalent of Kling, Seedance, Vidu, Hailuo and Qwen are continuing to overlap around the quality of images, consistency of characters, continuous narratives, generation of costs and vertical scenarios. A single-version upgrade can bring about a phased change in capacity, and it is becoming increasingly difficult to measure the actual value of a product in its entirety as the base generation capacity increases and a single sample is amazing.
The progress of Kling commercialization has been more clearly documented, and income growth has made it a player in the area of video generation that deserves sustained observation. However, the dimensions of income size, user growth, profitability and long-term competitiveness are different, and it is difficult to assert that it has taken a full lead in commercialization, without data on the same calibre of competition.
Competition also occurs in applied ecology. byte has products and services in different links, such as Seedance, Jimeng, CapCut, Douyin and Volcengine; fast hands also have content platforms and quick shadows. Kling, after obtaining more independent financing and business space, needs to build its own product ecology, both in partnership with the parent company and for outside developers and business clients.
There have been recent reports of the integration of the small sparrow of the AI short play creation platform under the banner of byte into Jimeng, the product still retaining an independent portal, and the associated short play capability has been accessed into Jimeng. This is an interesting contrast to the independent financing of Kling, which, on the one hand, seeks independent operating space for a single business, and on the other hand, further integrates the capacity and operation of nearby products. The two organizational options do not directly prove which path is better or more representative of a uniform trend across the industry, but point to the same reality: Businesses need to redefine which capabilities should be developed individually, which products should be integrated and which links can create sustainable commercial value.
When more and more third-party creative platforms can access different video models and simply have the capacity to generate, they may not necessarily be the exclusive advantage of a given application. Users can compare the effects, speed and prices of different tools, and can also change the model according to mission. To leave users behind, applications need to provide value other than models: workflow, which is more suitable for a creative task, more sophisticated editing and collaborative capabilities, more reliable content management, or direct links to distribution and commercial transformation.
Thus, the value of AI video products can be re-measured from three interrelated dimensions.
The first is to generate stability. Can the same task produce more stable results that meet the requirements? Is the picture of success accidental or can it be repeated under similar conditions? For professional users, while an amazing outcome is useful, stability determines whether it can become a regular tool in the production process.
The second is controllability of changes. Can tools be modified while retaining other valid content when people, actions or voices deviate? If each change is likely to introduce new errors, the model appears to provide more editorial capacity, while in practice it may increase back-to-work.
Finally, there are combined delivery costs. In addition to the time and points required for model generation, input from prompt preparation, screening asset, failed retry, manual restoration and sound post-production should be calculated. Bringing these costs together will make it possible to determine how much resources a usable video requires and to compare the efficiency of different tools in real workflow.
These three dimensions are further linked to commercial returns. A stable capacity for generation and modification reduces duplication of effort; a smoother workflow allows models to enter professional production sites such as advertising, short dramas and video; and predictable unit delivery costs affect the client ' s willingness to use and continue to pay.
For model companies, user size, ARR and financing valuation remain important, but the sustainability of growth is equally determined by the continuation costs, API use, computing costs and final profits.
In this process, multiple models and tools are used in combination and are likely to exist for a long time. The creator can generate images from one model, use another tool to process sound, then complete the clips with post-production through professional software, and the model manufacturer will try to get its capabilities into the user-run production process.
The $18 billion valuation, which priced the technical potential and business prospects of Kling, also provided space for its subsequent development, and the upgrading of 4.0 demonstrated its efforts in the direction of professional creativity. This is followed by continued validation of whether these technological capabilities can be translated into more stable outcomes, more manageable revision processes, and delivery costs that are acceptable to clients.
Good shot attraction, steady delivery to retain.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.