Founded Year

2021

Stage

Series D | Alive

Total Raised

$447M

Valuation

$0000 

Last Raised

$250M | 16 days ago

Revenue

$0000 

Mosaic Score
The Mosaic Score is an algorithm that measures the overall financial health and market potential of private companies.

+177 points in the past 30 days

About Fal

Fal offers tools for building and optimizing generative media applications, including an inference engine for diffusion models and support for training artificial intelligence (AI) models. Fal's services cater to developers looking to integrate AI capabilities into their applications. It was founded in 2021 and is based in San Francisco, California.

Headquarters Location

2261 Market Street Suite 10467

San Francisco, California, 94114,

United States

800-952-5210

Loading...

Loading...

Expert Collections containing Fal

Expert Collections are analyst-curated lists that highlight the companies you need to know in the most important technology spaces.

Fal is included in 3 Expert Collections, including Artificial Intelligence (AI).

A

Artificial Intelligence (AI)

37,207 items

Companies developing artificial intelligence solutions, including cross-industry applications, industry-specific products, and AI infrastructure solutions.

G

Generative AI

2,951 items

Companies working on generative AI applications and infrastructure.

U

Unicorns- Billion Dollar Startups

1,297 items

Latest Fal News

A16Z最新洞察:视频模型从狂飙到分化,产品化是下一个机会

Oct 28, 2025

A16Z最新洞察:视频模型从狂飙到分化,产品化是下一个机会 但今年,节奏变了。如果你密切关注基准测试,可能会感觉“进步”放缓了:大多数主流模型都能生成 10–15 秒带同步音轨的视频,效果已经相当惊人,但也不再令人惊讶。 这并不是坏事。在A16Z合伙人贾斯汀·摩尔看来,我们正在进入一个新的阶段:视频模型的“产品时代”。 简单来说,视频模型的进步,不再体现在模型参数或基准分数上,而是体现在多样性和专业化上。比如,我们开始看到不同模型在特定能力上各自突破:物理模拟、卡通风格、多镜头剪辑……没有哪一个模型能“通吃全场”,但每一个都在变得更擅长某一个维度。 与此同时,更大的机会开始从模型本身,转向“围绕模型”的产品构建:那些能简化创作流程、抽象出复杂操作的工具,正变得比模型本体更有价值。 今天,就让我们跟着贾斯汀·摩尔来看看视频模型在过去一年的变化。 过去几年,各大扩散模型实验室不断发布性能更强的新版本,在各种测试榜单上不断刷新纪录。大家逐渐形成一个共识:总有一天会出现一个“神级模型”,在所有视频生成任务中表现最出色,成为行业默认标准。 但这个假设最近被打破了。上个月发布的 Sora 2,在 LMarena 等测试中甚至不如 Veo 3,表现不升反降。很多人开始怀疑,扩散模型的技术进步是不是开始变慢了。所谓“最强模型”的概念,在视频领域可能根本不存在。 其实,大语言模型也走过类似的路径。2023到2025年,主流模型性能持续上升,之后在各种评测中逐渐趋于稳定。到了这个阶段,各家研究机构开始把重心放在具体场景和垂直领域上,而不是单纯追求更高分。同时,基于这些模型的AI产品也开始快速落地。 回头看,视频模型在公开测试上进展放缓其实也可以理解。过去几年,它们在“真实感”上突飞猛进,现在很多生成视频已经非常逼真。到了这个阶段,再想做得“更真实”就很难了,因为它已经几乎和现实看不出差别。 这就像17、18世纪的油画大师们,已经能画出接近照片的肖像和风景。那时大家不再纠结谁更写实,而是开始关注作品的风格和审美取向。 如果“更真实”不再是模型的优势来源,接下来会发生什么?我们可能会看到更多风格化、专业化的模型出现。每个模型不再追求通用,而是各有特长。资源丰富,选择也变多了。视频生成,正在进入一个“百花齐放”的新阶段。 02 那时我说:距离 AI 拍出像皮克斯那样的短片,还有很长的路要走。而现在,一切都变了。 谷歌推出了 Veo 模型,直接登上多个排行榜榜首;OpenAI 正在用 30 人团队、3000 万美元预算,制作一部完整的 AI 动画长片(虽然不是完全由模型生成,但依然是一次飞跃)。如今,视频长度更长,物理细节更真实。篮球从篮板反弹再落地?已经是标配。 不过,虽然整体水平在变好,我们也看到了另一个趋势:模型正在变得专一,各有特长。 为什么会这样?很简单,没有一款模型能满足所有用户的需求,比如有的团队专注提速和成本控制;有的专攻后处理阶段,让模型在某些场景表现特别好。 比如: Sora 2:可以根据一句话生成有趣的多镜头视频,像在帮你“拍短剧” Wan:开源模型,支持很多风格化插件(LoRA),适合定制风格 Grok:速度快、成本低,特别适合动画内容 Sora 适合创作趣味内容,比如用一句话生成一段“霍金打篮球”的视频,或把你和朋友放进某部电影里。它更像是一位“故事导演”,适合普通用户和 meme 创作者。但它在物理表现、音视频同步方面还不太稳定,经常出现嘴型不对、声音延迟等问题。 相比之下,Veo 就更“专业”。它缺乏幽默感,需要你提供更清晰的指导,但它的动作、镜头、音画同步更精确,更适合内容创作者、影视工作者这类对质量要求高的用户。 这种“专业化”趋势也带动了整条生态链的发展。像 Fal、Replicate 这样的 AI 视频云平台,已经托管了几十种模型,供用户按需选择。Krea 这类编辑工具,则提供了一个中心平台,让用户可以和多个模型打交道,并建立自己的工作流程。 当然,一些大公司仍在努力追求“万能型模型”,那种什么都能做、表现都顶尖的“上帝视角”模型。我们当然也希望它能早点出现。但在这之前,不同模型在不同场景里“各显神通”,已经是一个非常现实、非常值得期待的阶段了。 03 熟悉我的朋友都知道,我平时会用各种视频和图像生成模型,去尝试做一些非常定制化的内容。这个过程往往涉及好几个工具来配合使用。 举个例子:如果我要做一个“定制家具展示视频”,我通常会用到 Ideogram、nano-banana 和 Veo3;如果是想在已有视频中添加“产品赠品”的动画片段,那就要靠 nano-banana、Hedra,再加上一些编辑工具,比如 Krea 和 Kapwing。 这些组合工作流程其实挺复杂的,不是每个人都有时间、精力去折腾这么多工具。我们确实需要更好的一体化产品来简化整个创作过程。现在模型的能力已经很强,但对应的产品进度,依然有很多“追赶空间”。 很多创作者正在手动拼接多个模型的功能,来完成模型本可以自动做到的事情。 这些本可以由模型自动处理的工作,如今却依然依赖创作者手动拼接,正是产品体验和创作效率之间的巨大断层。好消息是,有些团队已经开始尝试解决这些问题。 Runway 就发布了一套工具,可以让用户修改镜头角度、生成下一个镜头、切换风格、改变天气,甚至在画面里加东西或删东西。 OpenAI 的 Sora Storyboard 也支持更细致地控制视频中每一帧的动作;而谷歌刚发布的 Veo 3.1,更像是一次“产品更新”而非“模型升级”,它围绕音频控制和视觉控制做了很多功能增强。 其实,这就像我们过去看到的大语言模型(LLM)一样:即便模型性能不再突飞猛进,围绕它构建实用产品的空间依然非常大。视频模型现在也处在这个阶段,能力不缺,缺的是好用的产品。 未来,我相信我们会看到越来越多“小而美”的模型,专门为某个行业或某种场景优化,比如室内设计、营销、动画制作等等。 同时,我们也需要更强大的“创意工具包”来打通各种模态,让视频、配音、音乐这些元素的生成与编辑更顺畅,最终形成一整套真正闭环的 AI 视频工作流。 本文来自微信公众号 “乌鸦智能说” ,作者:智能乌鸦,36氪经授权发布。 该文观点仅代表作者本人,36氪平台仅提供信息存储空间服务。

Fal Frequently Asked Questions (FAQ)

  • When was Fal founded?

    Fal was founded in 2021.

  • Where is Fal's headquarters?

    Fal's headquarters is located at 2261 Market Street, San Francisco.

  • What is Fal's latest funding round?

    Fal's latest funding round is Series D.

  • How much did Fal raise?

    Fal raised a total of $447M.

  • Who are the investors of Fal?

    Investors of Fal include Kleiner Perkins, Sequoia Capital, First Round Capital, Village Global, Andreessen Horowitz and 24 more.

  • Who are Fal's competitors?

    Competitors of Fal include Crusoe and 8 more.

Loading...

Compare Fal to Competitors

R
Radium

Radium provides enterprise artificial intelligence (AI) cloud solutions, focusing on the lifecycle of artificial intelligence development and deployment. Its offerings include a platform for training, fine-tuning, and deploying AI models, with features that support AI computing. The company's services are aimed at sectors that require AI capabilities, such as research institutions and enterprises interested in generative AI technologies. It was founded in 2019 and is based in Toronto, Canada.

Fireworks AI Logo
Fireworks AI

Fireworks AI is a company that provides a generative AI platform as a service within the technology sector. The company focuses on product iteration and cost reduction for its clients. It was founded in 2022 and is based in Redwood City, California.

CHAI AI Logo
CHAI AI

CHAI AI is a conversational AI platform in the technology sector, focused on the development and deployment of large language models. The company provides a platform where users can create and interact with AI chatbots, utilizing in-house trained LLMs and reinforcement learning techniques to improve user interaction. CHAI AI serves the social AI engagement sector, offering tools and technologies that allow developers to contribute to and utilize the platform's ecosystem. It was founded in 2021 and is based in Palo Alto, California.

Lambda Logo
Lambda

Lambda provides graphics processing unit (GPU) compute for artificial intelligence (AI) within the cloud computing industry. The company offers GPU instances and clusters for AI training and inference, as well as GPU cloud infrastructure. Lambda also offers Kubernetes services, inference endpoints, an application programming interface, and NVIDIA GPU workstations and desktops for deep learning applications. Lambda was formerly known as Lambda Labs. It was founded in 2012 and is based in San Jose, California.

Aethir Logo
Aethir

Aethir is a decentralized cloud computing infrastructure provider that offers access to GPUs for AI model training, fine-tuning, inference, and cloud gaming. The company primarily serves the artificial intelligence and gaming industries with its distributed cloud solutions. It was founded in 2021 and is based in Singapore, Singapore.

io.net Logo
io.net

io.net focuses on providing artificial intelligence (AI) solutions. The company offers a platform to harness global graphics processing unit (GPU) resources to provide computing power for AI startups, allowing users to create and deploy clusters for machine learning applications. It primarily serves the AI development and cloud computing industries. io.net was formerly known as Antbit. It was founded in 2022 and is based in New York, New York.

Loading...

CBI websites generally use certain cookies to enable better interactions with our sites and services. Use of these cookies, which may be stored on your device, permits us to improve and customize your experience. You can read more about your cookie choices at our privacy policy here. By continuing to use this site you are consenting to these choices.