Agnes 怕不是疯了!图片和视频 API 都敢免费Agnes Lost Her Mind! Offering Free Image and Video APIs


AI free token/AI免费积分

图片有点过于抽象…全球前十的 AI 大模型 Agnes 宣布无限期免费开放全模态 API。图片说实话,这种决策其实挺罕见的。以前免费开放的 AI 模型,要么是刚起步的初创公司想要冲一冲量,要么是模型本身还差那么点意思。在 Claw-Eval 测试榜单中,Agnes-2.0-Flash 在部分任务上超越了不少大厂的知名模型。图片且 Agnes 的图像和视频模型也都进入了 Artificial Analysis 榜单前列,获得多个国际评测体系的认可。所以,这样一个已经初具实力的公司,为什么要免费呢?图片 模型越强,也就越贵 AI 兴起后,我都是一个前沿体验者。用到现在,我最直观的感受是:模型越来越强,成本也越来越高。尤其是在 Agent 中使用时,模型需要反复地调用工具、读取上下文、进行多轮推理,一个项目下来,账单直接爆炸。图片对于专业的开发者和团队来说,也许早就习惯了这种账单,但对于普通玩家来说,这是一道让很多人不愿迈过的门槛。我觉得这是现在 AI 行业非常明显的一个矛盾。AI 厂商不断提升模型能力,希望吸引更多用户,但模型越强,使用成本越高,反而让更多人望而却步。所以我觉得 Agnes 免费开放,本质上就是在回答这个问题:当模型已经足够可用时,如何让更多的普通人参与进来?答案也显而易见,就是两个字:免费。图片6月1日,Agnes 正式宣布免费,到6月7日,官方公开了开放后的首周数据。文字模型调用量超过1万亿Token,图片模型生成量超过200万张,视频模型生成量超过200万秒。

这些数字说明一件事,当价格合适时,大家都愿意去更多地尝试使用 AI。

图片

我自己的使用习惯也有了变化,以前在使用时,心里总是下意识的计算成本,去尝试如何以更少的成本完成项目,结果出来的效果不理想,还要重新进行一轮,成本反而更高。

而现在想到就去用,用完再看效果,这种心态的变化,直接影响了我使用频率和探索的深度。

 免费不缩水,还在偷偷升级 

原本我以为,Agnes 这次免费只是为了扩大声量,之后不会继续迭代这些免费模型。

没想到就在前几天,也就是11号,Agnes 再次升级了免费的这几款模型。

升级点如下:

1、Agnes-2.0-Flash 上下文扩展至1M

2、Agnes-image-2.1-Flash 新增4K画质

目前顶级模型的上下文长度大多在 1M 左右,现在 Agnes 已经与顶尖旗舰模型看齐了。

图片

以后进行长文档总结时会更丝滑,而图像模型新增的 4K 分辨率也代表着在商业场景上更近了一步。

 我把 Agnes 接入了 Codex ① agnes-2.0-flash

agnes-2.0-flash 已经是我现在使用的主力模型了,用了这么长时间,我对这个模型的感受就是:用它既能省钱,也能省力。

我把 Agnes 接入了 Codex。用过 CC 的应该知道,Codex 官方是不支持第三方模型的,但是民间有邪修方法可以接入第三方模型,就是需要折腾一下。

Agnes 官方文档中也罗列了接入各个 Agent 工具的方法,其中就包括了 Codex。详细步骤官方文档中都写清楚了,我就不赘述了。

图片

Agnes 接入文档链接:

https://agnes-ai.com/doc/overview

接入成功后,就可以在 Codex 中正常使用 Agnes 的模型了。

图片1.小说总结现在 Agnes 的文字模型也支持1M上下文了,在进行中长篇小说解读时,就不再需要分割解读,而是进行整体解读。正好我最近对《海底两万里》这本书有兴趣,但是又没有时间去亲自读,就直接上传给了Agnes,让它帮我总结一下这本中篇小说。图片这篇小说的字数在30万左右,而1M Token保守估计可以算作70万个中文字符,但因为是上下文,还要计算AI发送给我们的字符,所以数字砍半,这个模型在阅读30到40万字的文章时,是不需要切片理解的。图片最后总结的内容基本准确,与原书内容一致。我把它总结的结果发给其它AI模型进行验证,确实和原书内容一样。图片2.制作可视化图表我这里有一份上个月的消费清单,我想看在哪一部分的开销最多,分别开销都是多少。我就让 Codex 帮我把这份清单表格,转化为可视化图表。图片它是直接用HTML的方式创建了图表,节省了写 Python 脚本,我们看看效果如何:图片我没有指定图表类型,它便自动汇总了所有可分析的数据,包括品类、支付方式、甚至订单状态都总结了出来。这样一眼看过去,就知道上个月电子产品买得最多了。3.企业问答知识库Web工具这次我们再让 agnes-2.0-flash 制作一个Web工具,看看它的 Coding 能力如何。我准备让它根据企业资料做一个问答知识库工具。通过这个工具直接检索公司的资料。图片执行过程中断了一次,不知道是模型的问题还是 Codex 的问题,毕竟是第三方接入,难免有些适配问题。恢复后,Web 工具顺利创建完成,耗时约两分钟,速度还是很快的。图片实测成功运行,所有问题都能成功回答,且答案与我提供的公司资料一致。② agnes-image-2.1-flash

这个模型支持图改图、多图融合、局部修改、背景替换、风格转换、文字编辑、图像修复等多种任务。

宣布免费之后,网友也是很热心的制作了多个生图生视频的 Agnes Skill 发布在 Github 上。我用的是这个:图片直接复制 Github 的链接,让 Agent 安装这个技能就可以,之后发送你的 API 给 Agent,就可以开始调用 Agnes 生图和视频了。图片1.多图融合这次模型升级了4K,在商业上有更好的使用空间了,正好最近也是618,这个模型在电商方面有很多用途。比如多图融合,发一张模特和衣服的照片,就能让Agnes生成这个模特穿着这个衣服的照片。图片效果如下:图片对比一下原图,人物样貌和裙子细节都很还原:图片2.图片编辑如果对模特的某些部位不满意,还能让Agnes进行修改,例如把模特的头发换成金色的。图片只是变了个发色,整个人的气质都不一样了。图片3.背景替换还可以改变融合后的图片背景,让穿着新衣服的模特置身于一个更合适的环境中。图片图片4.创意效果最后再来给大家生成一些创意图片,看看这个模型生成的图片效果如何。图片这张穿着羽绒服的小鸟真实感很强,连翅膀也做了袖子,虽然现实中不可能发生这样的场景,但还是很好地融合了出来。图片卡通风格也可以驾驭。图片图片图片图片图片它出图的速度确实很快,15 秒左右就能完成一张图片。上面这些素材都是由 agnes-image-2.1-flash 生成的,不花钱的感觉就是爽 图片图片③ agnes-video-2.0

提示词:

生成一个视频:高强度动作惊悚场景,手持晃动镜头,电影感动态模糊,女人冲进驾驶舱并僵住,因为她意识到飞行员座位是空的,警报闪烁红光,颠簸震动着机舱,她向前冲刺并准备把自己绑在飞行员座位上。

 
 
 

 

 

提示词:

甜美的中国女生,甜美谈恋爱氛围,生成男友视角手机拍摄的vlog,在水城威尼斯;接头看着路过的白马发呆;在水城的河道里乘着船;民宿旅店床上嘟着嘴捏脸蛋的大特写,以及其它适合威尼斯旅拍的地方,手机拍摄。

 
 
 

 

 

 写在最后 

写这篇文章的时候我一直在思考,大部分公司都是受利益而驱使,但 Agnes 这样把自家旗舰模型免费,于他们而言究竟有什么好处。

想了很久都没有想明白,但我为什么要以恶意去揣测别人的初衷。倒不是说赚钱恶意,赚钱很正常,但相比较赚钱来说,能够选择免费,君子论迹不论心。

或许他们官网的标语就是答案:Agnes 想让世界级的AI属于每一个人。

图片

虽然没有明确表明这种免费能持续多久,但在当前这个时间点,Agnes 确实让大家都能免费用上大模型。

官网:https://agnes-ai.com

本文发表于公众号【莫理】

关注我们,阅读更多精彩内容

▽▽▽

 

A bit too abstract... Agnes, one of the top ten AI large models globally, has announced the indefinite free opening of its full-modality API. Honestly, this kind of decision is quite rare. Previously, free AI models were either early-stage startups trying to build momentum, or models that were still lacking in capability. In the Claw-Eval benchmark, Agnes-2.0-Flash surpassed many well-known models from major tech companies on certain tasks. Moreover, Agnes's image and video models have also entered the top ranks of the Artificial Analysis leaderboard, earning recognition from multiple international evaluation systems. So, why would a company that already has considerable strength go free? The stronger the model, the more expensive it is. Since the rise of AI, I've been a frontier experimenter. My most intuitive feeling after using it is: models are getting stronger, but so are the costs. Especially when used in Agents, the model needs to repeatedly call tools, read context, and perform multi-round reasoning. One project later and the bill explodes. For professional developers and teams, maybe they're accustomed to such bills, but for ordinary users, this is a threshold many are reluctant to cross. I think this is a very obvious contradiction in today's AI industry. AI companies continuously improve model capabilities, hoping to attract more users, but the stronger the model, the higher the cost, which drives more people away. So I believe Agnes's decision to go free is essentially answering this question: when the model is already good enough, how do we let more ordinary people participate? The answer is obvious—two words: free. On June 1st, Agnes officially announced free access, and by June 7th, the official team released the first week's data after going open. Text model calls exceeded 1 trillion tokens, image model generation exceeded 2 million images, and video model generation exceeded 2 million seconds.图片图片图片图片图片图片

These numbers show one thing: when the price is right, everyone is willing to try using AI more.

图片

My own usage habits have also changed. Previously, when using AI, I would subconsciously calculate the cost and try to complete projects with less expense, but the results were unsatisfactory, requiring another round, which actually cost more.

Now I just use it when I think of something, then check the results afterward. This change in mindset has directly affected my usage frequency and the depth of exploration.

Free without compromise, secretly upgrading

I originally thought Agnes's free offer was just to build buzz, and they wouldn't continue iterating these free models afterward.

But a few days ago, on the 11th, Agnes upgraded these free models again.

The upgrade points are as follows:

1. Agnes-2.0-Flash context expanded to 1M

2. Agnes-image-2.1-Flash now supports 4K quality

Currently, the context length of top-tier models is mostly around 1M, and now Agnes is on par with leading flagship models.

图片

Long document summarization will be smoother in the future, and the new 4K resolution for the image model represents a step closer to commercial scenarios.

I integrated Agnes into Codex ① agnes-2.0-flash

agnes-2.0-flash has already become my primary model. After using it for so long, my feeling is: it saves both money and effort.

I integrated Agnes into Codex. Those who've used CC should know that Codex officially doesn't support third-party models, but there are unofficial methods to integrate them—it just takes some tinkering.

Agnes's official documentation also lists methods to integrate with various Agent tools, including Codex. The detailed steps are all written in the official docs, so I won't repeat them.

图片

Agnes integration documentation link:

https://agnes-ai.com/doc/overview

After successful integration, you can use Agnes's models in Codex as usual.

1. Novel summarization Agnes's text model now supports 1M context, so when interpreting medium-to-long novels, there's no need to split them anymore—you can process the whole thing at once. I recently got interested in "20,000 Leagues Under the Sea" but didn't have time to read it personally, so I uploaded it directly to Agnes and asked it to summarize the novella. This novel is about 300,000 Chinese characters, and 1M tokens can conservatively be estimated as 700,000 Chinese characters. But since context also includes characters sent by AI back to us, the effective number is halved, so this model can read 300,000 to 400,000 characters without slicing. The final summary was basically accurate and consistent with the original book. I sent its summary to other AI models for verification, and indeed it matched the original content. 2. Creating visual charts I have last month's spending list and want to see which category had the highest expenses and how much each accounted for. I asked Codex to convert the list into a visual chart. It created the chart directly using HTML, saving the need to write a Python script. Let's see the result: I didn't specify the chart type, yet it automatically summarized all analyzable data, including categories, payment methods, and even order status. A quick glance shows I spent the most on electronics last month. 3. Enterprise Q&A knowledge base Web tool This time let's have agnes-2.0-flash build a Web tool to test its coding ability. I plan to have it create an enterprise Q&A knowledge base tool based on company materials, allowing direct retrieval of company information. During execution it paused once—unsure whether it was a model issue or a Codex issue, since third-party integration inevitably has some compatibility problems. After resuming, the Web tool was created successfully in about two minutes, which is quite fast. The test ran successfully—all questions were answered correctly, and the answers matched the company materials I provided. ② agnes-image-2.1-flash图片图片图片图片图片图片图片图片

This model supports various tasks including image-to-image, multi-image fusion, localized editing, background replacement, style transfer, text editing, and image restoration.

After the free announcement, netizens enthusiastically created multiple Agnes Skills for image and video generation and published them on Github. I'm using this one: Simply copy the Github link and let the Agent install the skill, then send your API key to the Agent, and you can start calling Agnes for image and video generation. 1. Multi-image fusion This time the model upgraded to 4K, opening up better commercial possibilities. With 618 approaching, this model has many applications in e-commerce. For example, with multi-image fusion, you upload a photo of a model and a piece of clothing, and Agnes can generate an image of the model wearing that outfit. Here's the result: Comparing with the original photo, the person's features and the dress details are both well preserved. 2. Image editing If you're not satisfied with certain parts of the model, you can also ask Agnes to make modifications, such as changing the model's hair color to blonde. Just changing the hair color completely transforms the whole vibe. 3. Background replacement You can also change the background of the fused image, placing the model in a more suitable environment. 4. Creative effects Finally, let me generate some creative images to see how this model performs. This image of a bird wearing a down jacket is very realistic—the wings even serve as sleeves. Though such a scene is impossible in reality, the integration turned out beautifully. Cartoon styles are also handled well. The image generation speed is indeed fast, taking about 15 seconds per image. All the above materials were generated by agnes-image-2.1-flash. The feeling of not spending money is just satisfying. ③ agnes-video-2.0图片图片图片图片图片图片图片图片图片图片图片图片图片图片图片图片图片图片

Prompt:

Generate a video: high-intensity action thriller scene, handheld shaky camera, cinematic motion blur, a woman rushes into the cockpit and freezes as she realizes the pilot's seat is empty, alarm lights flash red, the cabin shakes violently, she sprints forward and prepares to strap herself into the pilot's seat.

Followed

Follow

Replay Share Like

Close

Watch more

More

Exit fullscreen

Mo Li Followed

Share video

, duration 00:05

0/0

00:00/00:05

Switch to landscape mode

Continue playing

Progress bar, 0%

Play

00:00

00:05

00:05

Speed

Fullscreen

Playing at speed

0.5x 0.75x 1.0x 1.5x 2.0x

Ultra HD Smooth

Continue watching

Is Agnes crazy? Even daring to make image and video APIs free

Watch more

Repost

,

Is Agnes crazy? Even daring to make image and video APIs free

Mo Li Followed

Share Like Watch

Synced to Look-look Write your comment

Video details

Prompt:

A sweet Chinese girl, sweet dating atmosphere, generate a boyfriend-perspective phone-shot vlog in the water city of Venice; standing at the intersection staring blankly at a passing white horse; riding a boat through the canals; an extreme close-up of pouting and pinching her cheeks on a homestay bed; and other places suitable for Venice travel photography, shot on a phone.

Followed

Follow

Replay Share Like

Close

Watch more

More

Exit fullscreen

Mo Li Followed

Share video

, duration 00:13

0/0

00:00/00:13

Switch to landscape mode

Continue playing

Progress bar, 0%

Play

00:00

00:13

00:13

Speed

Fullscreen

Playing at speed

0.5x 0.75x 1.0x 1.5x 2.0x

Ultra HD Smooth

Continue watching

Is Agnes crazy? Even daring to make image and video APIs free

Watch more

Repost

,

Is Agnes crazy? Even daring to make image and video APIs free

Mo Li Followed

Share Like Watch

Synced to Look-look Write your comment

Video details

Final thoughts

While writing this article I kept wondering: most companies are driven by profit, so what benefit does Agnes get from making its flagship models free?

I thought about it for a long time and still couldn't figure it out, but why should I assume ill intent toward others' motivations? I'm not saying making money is bad—making money is normal—but compared to choosing to make money, choosing to go free is what matters. Judge by deeds, not by heart.

Perhaps the answer lies in their official slogan: Agnes wants world-class AI to belong to everyone.

图片

Although it hasn't explicitly stated how long this free access will last, at this current moment, Agnes has indeed allowed everyone to use large models for free.

This article was published on the WeChat public account [Mo Li]

Follow us to read more great content

△△△

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry