AI Transformations (beta)

ImageKit provides easy-to-use URL-based transformation parameters to perform various AI transformations on images. These transformations are powered by AI models and can be used to perform tasks like background removal, drop shadow, retouching, upscaling, object-aware cropping, generating an image via a text prompt, generating variations of an image, etc.

For example, to remove the background from an image, you can use the tr=e-removedotbg transformation parameter in the URL. Here is an example URL:

https://ik.imagekit.io/ikmedia/blue-bmw.jpg?tr=e-removedotbg

1. Except fo-face, fo-auto and object-aware cropping, all AI transformations are currently in beta.
2. AI transformations typically take longer to process compared to basic transformations. The response time can range from a few seconds to a minute, depending on the complexity of the transformation. If the transformation takes longer, ImageKit may return an intermediate HTML response with the message The asset is currently being prepared and a 200 status code. In this case, a special response header is-intermediate-response:true is included. You can use this header to identify intermediate responses and handle them appropriately.
3. Only one such async operation will be allowed in a single transformation in a chain of transformations. If you want to apply multiple AI transformations, you can chain them together using a colon (:) separator.
   tr:e-removedotbg,h-100:e-dropshadow - ✅ (e-dropshadow is separated by a colon)
   tr:e-removedotbg,h-200,e-dropshadow - ❌ (e-dropshadow is separated by a comma)
4. If a request fails due to any reason, the error response on the corresponding URL is cached for a short duration. You should wait for a few minutes before ImageKit accepts and initiates a new request on the same URL. Learn about error troubleshooting.
5. Most of these transformations cannot be used inside layers, with the exception of e-removedotbg, and e-dropshadow.
6. The outputs of AI transformations are cached forever to prevent redundant utilization of your extension credits. Purging the cache for a URL will still delete all cache entries for that URL. Also, changing any of the image settings via your dashboard (default quality, auto-format, device-based resize, auto-rotation, chroma subsampling, preserve metadata, PNG compression) may invalidate existing cache entries.
7. The Restrict Unsigned URLs setting can be toggled on or off specifically for AI transformations. It will be turned off by default so you can easily try these transformations out. However, we strongly recommend turning it on for production use cases to prevent unauthorized access to your AI transformations.

Using AI transformations consumes extension units. Different AI transformations consume different amounts of extension units as per your pricing plan.

The standard pricing plan for AI transformations is as follows. We are constantly working on improving the AI models and reducing the cost of these transformations. The pricing plan is subject to change.

Remove the background from an image using the e-removedotbg transformation parameter. You can use e-removedotbg inside image overlay layers in image requests.

Syntax: e-removedotbg

Here is an example URL:

bg-genfill transformation extends an image beyond its original boundaries, allowing you to add new visual elements in an image while preserving the original image content.

Syntax: bg-genfill[:-prompt-${text}] or bg-genfill[:-prompte-${urlencoded_base64_encoded_text}]

It must be used with height, width, and either the pad_resize or pad_extract crop strategy.

Here is an example URL:

You can also provide a text prompt to control the generative fill.

https://ik.imagekit.io/ikmedia/footwear.jpg?tr=bg-genfill-prompt-flowers,w-1000,h-960,cm-pad_resize

Add a drop shadow around an object in an image using the e-dropshadow transformation parameter. You can control the direction, elevation, and saturation of the light source.

Syntax: e-dropshadow[:-az-${0-360}][:_el-${0-90}][:_st-${0-100}]

• az (Azimuth): This represents the angle of the light source. A value of 0 or 360 indicates the light is directly behind the object. The default value is 215, which creates a natural shadow effect.
• el (Elevation): This refers to the elevation angle of the light source above the ground. The default value is 45. Lowering this value results in a larger, more diffused shadow.
• st (Saturation): This controls the intensity of the shadow. Higher values produce a more intense or darker shadow. The default value for the st parameter is 60.

Examples:

Improve the quality of an image using the e-retouch transformation parameter.

Syntax: e-retouch

Here is an example URL:

Increase the resolution of an image using the e-upscale transformation parameter.

Syntax: e-upscale

Here is an example URL:

You can generate an image based on a text prompt. The text prompt can simply describe the image you want to generate.

Syntax: {URL_ENDPOINT}/ik-genimg[:-prompt-${text}]/...filepath.../filename.extension

Here filepath and filename.extension is up to you to decide. Once the image is generated, it will be accessible at the provided path. Changing the text prompt on the same path will not generate a new image. To generate another image, either purge the cache for this path, change the filepath or filename, or append a random query parameter to the URL.

Examples:

You can use any other image transformation parameter along with ik-genimg to modify the generated image further.

For example, remove the background from the generated image and resize it down to 500px width:

https://ik.imagekit.io/ikmedia/ik-genimg-prompt-A man eating a burger/gen-burger-image.jpg?tr=e-removedotbg,w-500

Let's add a text overlay layer on the generated image:

https://ik.imagekit.io/ikmedia/ik-genimg-prompt-A%20woman%20wearing%20red%20top/gen-woman-red-top.jpg?tr=w-500,l-text,i-30%%20off,fs-45,lfo-top_left,pa-10,bg-white,l-end

If you have an image and want to generate variations of it, you can use the e-genvar transformation parameter. This will generate a new image with slight variations from the original image. The variations include changes in color, texture, and other visual elements. However, the model will try to preserve the structure and essence of the original image.

Syntax: e-genvar

Here is an example URL:

For more details, refer to the object-aware cropping documentation.

We can also crop out a particular object of interest out of multiple different objects present in an image.

Crop an image based on the detected face using the fo-face transformation parameter. Optionally, zoom in or out using the z parameter.

https://ik.imagekit.io/ikmedia/tr:w-150,h-150,fo-face/docs_images/examples/example_portrait_1.jpg

Automatically focus on the most important part of an image using the fo-auto transformation parameter.

https://ik.imagekit.io/ikmedia/tr:w-200,h-300,fo-auto/docs_images/examples/Smart_crop_example.jpg

This is what the final result looks like with smart crop auto mode. The main subject is right in the center of the final thumbnail.