Agnes免费生图批图API+一键生图软件!Agnes Free Image Generation/Processing API + One-Click Image Creation Software!


AI free token/AI免费积分

昨天发了一篇免费调用大模型的文章,里面提到了 Agnes 这个平台除了提供免费的文字模型之外,还提供了图片和视频模型。视频我试了一直排队中,图片已经搞定了。

我已经确认,可以正常“生图”,正常“P 图”。然后,我又忍不住搓了一个不知道有啥用的“AI 生图软件”!

今天主要分享一下这个生图模型和生图软件。

有了 AI 之后,基本上软件自由了,很多场景不需要使用第三方软件了,全部使用自己开发的软件!

我们先来看看文档,了解一下 Agnes Image 这个图片模型!

图片

网址:https://agnes-ai.com

 

当前最新版本为 Agnes Image 2.1 Flash,2.5预览版在路上。这是 Sapiens AI 推出的升级图像生成模型,支持 文生图 和 图生图 两种工作流。

图片

与之前版本相比,Agnes Image 2.1 Flash 在 高信息密度图像 生成方面性能有所提升,更适合需要复杂视觉细节、更丰富构图以及更清晰语义对齐的场景。

我是第一次接触这个平台,还挺新鲜的。我也没用过之前版本,所以也没啥概念。

那就直接试一下吧。

我喜欢用一些极其简单的提示词,来体验生图模型的效果。如果简单提示词都可以很好的话,下限就有保障了!

1. 看一看效果

提示词:

1apple

图片

这个真实度还可以啊!

提示词:

1girl

图片

整个真实感还是挺好的,不像有些模型,全是浓妆艳抹的网红美女。可惜,这个美女的颜值不在我的审美上!

提示词:

一个木质的电脑桌上放着一台苹果电脑,清晨的阳光打在桌上

图片

整体感觉还可以,但是屏幕细节一般,阳光不错,但感觉不像清晨。

然后试一下修改图片:

修改电脑的壁纸,改成 macOS 26 的经典壁纸,电脑画面属于登录状态。

图片

它成功定位到了屏幕这个区域,并进行了修改,所以它的理解能力还可以。

但是我说的是 macOS 26、登录状态。其实这两个是有点难理解的,因为一个是比较新,还有一个是登录状态,至于到底是登录的界面还是登录后的状态,可能也有歧义。所以导致它这个界面很杂糅,它对新内容的理解还不是很好。从生成的画面可以看出来,它对这一块的训练是不充分的。

继续修改图片:

把衣服改成黑色,然后胸口有一个很有设计感的Jarivs。

图片

这张图片,我本来默认生成的是一件米色的 T 恤,然后我就尝试让它改了一下衣服的颜色,以及在衣服上加了一些内容。

这个执行得还是非常不错的!

改完,发现我把 Jarvis 拼错了,这是我的锅啊,它也没有自作主张给我纠正,这是对的。

图片生成和修改的效果大概就是这个样子。

它主要就是生成图片的能力和修改图片的能力。我也就是粗略地测一下,保证它基本上是能用的!

确认是可以用的之后,我们来看一下怎么用。

先来看一下它的 API 文档,看具体写了哪些东西,然后我会分享一下我开发完成的软件以及使用方法。

2. API 文档

因为这个免费提供的是“API”,所以需要调用 API 才可以。简单看一下它的文档情况。

Base URL:

https://apihub.agnes-ai.com

Endpoint:

POST https://apihub.agnes-ai.com/v1/images/generations

模型名称:

agnes-image-2.1-flash

上面 3 条是调用 API 最关键的信息啊,可以记录一下。

请求参数:

参数 类型 是否必填 描述
model string 是 模型名称,请使用 agnes-image-2.1-flash
prompt string 是 图像生成或图像编辑的文本指令
size string 是 输出图像尺寸,例如 1024x768
image string[] 图生图必填 输入图像数组,支持公共图像 URL 或 Data URI Base64
return_base64 boolean 否 文生图输出需要以 Base64 返回时使用
extra_body object 否 高级工作流的附加参数
extra_body.response_format string 否 输出格式,常见值:url, b64_json

这是它主要的请求参数。

 

cURL 示例:

curl https://apihub.agnes-ai.com/v1/images/generations \  -H "Authorization: Bearer YOUR_API_KEY" \  -H "Content-Type: application/json" \  -d '{    "model": "agnes-image-2.1-flash",    "prompt": "A luminous floating city above a misty canyon at sunrise, cinematic realism",    "size": "1024x768",    "extra_body": {      "response_format": "url"    }  }'

这个接口可以文生图、图生图,可以传 base64 和 URL 形式的图片,也可以返回 URL 和 base64 的图片。

提示词结构和示例:

[主体] + [场景 / 环境] + [风格] + [光照] + [构图] + [质量要求]日出时分薄雾峡谷上方的发光浮空城市,电影级写实风格,广角构图,丰富的建筑细节,柔和的金色光线,高视觉密度将场景转换为霓虹倒影的雨浸赛博朋克夜晚,同时保留原始构图和主体布局。充满飞行器、全息招牌、密集人群、霓虹灯光的未来城市市场,电影级写实风格,超精细,高信息密度构图

这是官方提示词结构和一些示例。

 

定价:

最后来看一下价格是什么样的。

生成图像 $0.003 / 张 $0 / 张
  收费 免费

前面说过,它的文字模型、图片模型和视频模型都是免费的。所以我们都处于一张零元的状态,它也标注了一个收费的价格,收费的价格是“2 分”一张,这便宜到离谱!

好了,参数模型啊,价格啊,全都看完了,接下来就可以真的来使用了。

3. 生图软件

有了文档之后,生成图片就非常简单了,可以直接把文档交给 Agent,也可以让 Agent 写一个可视化的软件。

我觉得有些流程类的非常适合用 skill 来完成,但是这些图形类的,我更喜欢用界面来进行管理和操作。固定成界面和软件还有一个好处,就是省 token 以及排除不同模型带来的不稳定性。

我之前刚好有一个图片软件的架子,是测试 Opus 的时候生成的。当时的测试题目是,让它参考“美图秀秀”的布局生成一个桌面软件。

当时大概做了三个功能,主要完成图片的压缩,格式转换,裁切,以及拼图这些操作。

图片

这个是裁切和格式转换功能。可以很方便地拖拽、裁切,然后支持常见媒体平台的比例以及常见的文件格式。

我主要是用作公众号封面用的。另外 AI 生成的图片有时候太大了,我会转换一下格式和压缩一下。

图片

这个是用来拼图的。

文章里面有时候图片太多了,拼成一张,文章会更加紧凑一些。

另外一个重点是,模仿 QQ 截图,开发一个截图功能。

我一直都用 QQ 截图的,因为它这个功能确实做得很好,很多细节都考虑到了。但是它现在有个巨大的问题,就是截完图之后不能选择格式,而且很多图片一截就是 6M+。我会把我的图片上传到自己的对象存储里面,它就会占用我的空间和流量。所以我非常需要一个截完图能自动压缩的功能。

我就是为了这一个功能,才要手搓这个截图功能。目前除了定位有一点问题,其他都 OK 了。

上面是本来就有的功能,因为这个 AI 生图模型又加了一个功能:

图片

生图功能界面非常简单:

图片

左边是输入区域,右边是预览区域。

只要选一下生成类型,然后输入内容,选择比例,点击生成就可以了!

有了软件之后,API 使用就非常简单了。只要获取 API Key,配置到软件里面就可以了,其他都不用管。API Key 的获取方式请参考上一篇。

配置的方式也极其简单。点击右上角的设置按钮,找到 AI 生图:

图片

把 Agnes 的 API 放到里面就可以了!

配置完成之后,立马就可以愉快地玩耍了。这个生图是没有任何门槛的,注册获取API以及生成图片是不需要任何复杂的授权和认证的,只要你有一个邮箱,注册了就能使用。

但是天下没有免费的午餐,要什么“代价”,你们自己想清楚就可以了。另外,白嫖这种东西都是先到先得,这是亘古不变的道理!

还有,有人说我上一篇是广告图,我就稍微强调一下,这两篇文章都不是“软广”,我文章中所有的内容和操作都是真实的,你们只要看对你们有没有用就行了,有用就拿去用,没用扔一边就好了。

如果硬要说是“软广”的话,就是我夹带一下我自己的软件~~哈哈!

发完之后,立马就有人找我了啊,看来我的带货潜力也是很强哦!

其实我也一直在思考一个问题啊:一个AI博主,如何在不卖课、不带货、不接商单、不骗粉丝的情况下用 AI 赚到钱,存在这样的模式么?

大家可以讨论一下!

 

Yesterday, I published an article about free access to large language models. In it, I mentioned that the Agnes platform not only offers a free text model but also image and video models. I tried the video, which seems to be in the queue, but the image one is ready to go.

I've confirmed that I can successfully "generate images" and "edit pictures." Then, almost reflexively, I created an "AI image generation software" that I'm not quite sure what people will use it for yet!

Today, I want to share this image generation model and the software I made.

With AI, software development becomes almost effortless. We're at the point where we don't need to rely on third-party software; we can develop everything ourselves!

Let's start by looking at the documentation and understanding the Agnes Image model.

Website: https://agnes-ai.com

The current version is Agnes Image 2.1 Flash, with version 2.5 in preview. This is an updated image generation model by Sapiens AI, supporting both text-to-image and image-to-image workflows.

Compared to previous versions, Agnes Image 2.1 Flash performs better on high-information-density images, making it more suitable for scenes requiring complex visual details, richer compositions, and tighter semantic alignment.

This is my first time using this platform, so it's all new to me. I haven't used the previous versions, so I don't have any points of comparison.

So, let's just give it a try.

I like to use extremely simple prompts to test the performance of image generation models. If it handles simple prompts well, it's a good sign of quality!

1. Let's see the results.

Prompt:

1apple

The realism is pretty good!

Prompt:

1girl

The overall realism is very good; it's not like some models that only produce heavily made-up, influencer-like women. Unfortunately, this girl's looks aren't to my taste!

Prompt:

A wooden computer desk with an Apple computer on it, with morning sunlight shining on the table.

The overall feel is decent, but the screen details are poor, and while the sunlight is present, it doesn't really look like early morning.

Now, let's try editing a picture:

Change the computer's wallpaper to the classic macOS 26 wallpaper, and show the computer in a logged-in state.

It successfully identified the screen region and made changes, so its comprehension is solid.

However, I specified macOS 26 and a logged-in state. These are hard concepts to capture visually — “26” is too new, and a logged-in state could imply the login screen or the desktop. The resulting image is a bit of a mess, indicating that the model isn't well-trained on understanding specific interface states yet.

Keep modifying the picture:

Change the shirt to black and add a stylish Jarvis logo to the chest.

This worked very well!

After the edit, I realized I spelled Jarvis wrong — that's on me. The model didn't try to correct it, which is good practice for AI assistants.

That's about the level of image generation and editing capability we're looking at.

I'm just doing a rough test to make sure it's functional.

Confirming it works, let's look at how to use it.

First, check the API documentation to see what's involved. Then I'll share the software I developed and how to use it.

2. API Documentation

Since this is provided as an "API," we need to call it via code. Let's look at the docs quickly.

Base URL:

Endpoint:

Model Name:

These three points are critical for making the API call. You should take note of them.

Request Parameters:

These are the main parameters for the request.

cURL Example:

The API supports text-to-image and image-to-image operations. You can send images in base64 or as a URL, and receive the output as a URL or base64.

Prompt Structure and Examples:

Here are the official prompt structures and some examples.

Pricing:

Finally, let's check the price.

As mentioned earlier, the text, image, and video models are free. So, we're currently operating at zero cost per image. The listed price for paid use is "2 cents," which is ridiculously cheap!

Alright, we've covered the parameters, model, and pricing. Next up is the actual usage.

3. Image Generation Software

With the documentation in hand, generating images is straightforward. You can feed the docs to an agent or, like I did, get the agent to write a desktop application.

I find that for repetitive tasks like this, a simple UI is much better than relying on command-line or API calls every time. It saves tokens and standardizes the workflow.

I already had a framework for a photo editing software, which I built while testing Opus. The initial task was to create a tool resembling "Meitu Xiu Xiu" with common image operations.

Currently, it handles basic image compression, format conversion, cropping, and stitching.

This is the cropping and format conversion section. It allows for easy dragging and resizing, supports standard social media ratios, and various file formats.

I mostly use this for converting my WeChat article covers. Additionally, AI-generated images often come in large sizes, so this helps in downscaling.

This is the stitching feature.

Sometimes, articles have too many images scattered around, so stitching them into one image makes the layout cleaner.

Another key feature is the screenshot function, inspired by QQ's screenshot tool.

I've always used QQ for screenshots because the feature is excellent and covers many edge cases. However, the post-processing is weak — you can't choose the format (it's always PNG) and the files are huge (often 6MB+). Since I upload my images to my own storage, this wastes space and bandwidth. I desperately needed a tool that could automatically compress screenshots after capture.

This is why I built the screenshot feature myself. Besides some minor positioning issues, it works well.

The existing features are covered above. Since we are discussing a new AI image model, I added a new section:

The AI image generation interface is very simple:

The left side is the input area, and the right is the preview area.

Just select the generation type, enter your prompt, choose the aspect ratio, and hit "Generate"!

After adding this feature, using the API is trivial. You just need to get your API Key from the Agnes dashboard and plug it into the software settings.

The setup is super simple. Click the settings icon in the top right, go to "AI Image Generation":

Paste your Agnes API key into the field.

Once configured, you are ready to go. Generating images is free and requires no complex authentication beyond registering an account.

However, remember that free services often have hidden costs (like data collection or rate limits), and availability isn't guaranteed — it's a "first-come, first-served" deal.

Also, some people have accused me of shilling (做广告). I want to be clear: this is not sponsored content. I am sharing because I found the service useful during my experiments. If it helps you, great. If not, move on.

If anything, the "shilling" is for my own software, which I'm sharing openly for free anyway!

Actually, this leads me to a broader question: How can AI content creators make money without selling courses, endorsing products, accepting ads, or deceiving their audience? Is there a viable business model here?

Feel free to share your thoughts!

图片

图片

图片Welcome to the NVIDIA Developer Zone

图片Get early access to the latest NVIDIA technologies.

图片Join the conversation on our online forums.

图片Stay up-to-date with developer newsletters.

图片Learn and engage with NVIDIA experts at GTC.

图片Connect with other developers in local communities.

图片Find the right software for your next project.

图片Access the latest NVIDIA software, drivers, and documentation.

图片Get step-by-step guidance for setting up and running NVIDIA hardware.

图片Explore a library of sample code and tutorials.

图片

图片

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