Generative Background Fill has quickly become one of our most popular generative AI features with just under a million unique transformations so far in 2024. Our AI-driven image editing models continue to evolve and today, we’re excited to introduce the newest version of Generative Fill, now powered by advanced image editing models for more accurate results.

Generative Fill uses artificial intelligence to add, remove, expand, or edit images  by adding content around the edges to extend or blend multiple images. 

This process, known as outpainting,  enhances our padded crops feature by intelligently filling new areas when adjusting aspect ratios, rather than just adding solid backgrounds like traditional padding.

Generative Fill takes this process further by analyzing the image and using generative AI to fill new areas with matching background imagery so that the end result matches and fits into the original style and setting. 

Check out our original blog post or explore the documentation for more details on our improved Gen Fill feature or try it out for yourself on our Generative Fill demo.

A pad crop with a solid color fill:

Loading code examples

A pad crop using Generative Fill:

Loading code examples

• Cleaner expansions. Based on user feedback, we’ve reduced distracting objects in the extended areas, making expansions appear cleaner and more natural.
• Improved image quality. Our updated models deliver more precise, detailed images with a higher resolution in every expansion.
• Latest technology. The tool uses an advanced model designed for more realistic, seamless image generation.
• Smart prompts for consistency. To ensure consistent results, Generative Fill now generates optimized prompts automatically without specific user instructions.

While there are many improvements, a few things remain the same:

• Syntax. You can continue using the same syntax for Generative Fill as before, making this update easy to adopt. For syntax details, check out the documentation.
• Transformation counts. Despite the improvements, Generative Fill continues to use the same transformation count, so there’s no impact on your current usage.

The following examples show how the features’ improvements look in the real world. Though the initial results are still quite good, the improvements in the output of the new version of Generative Fill are pretty remarkable.

In this example, we can see that the old result isn’t bad. However, the new version has much more detail. It intelligently detects and handles the model’s shadow and can even add fine-tuned details to the floor, such as the grooves between tiles. The upper part of the generated image is much cleaner and more natural.

Old Generative Fill (left) and New Generative Fill (right)

In this example, the new Generative Fill result maintains the details of the wall behind the subject, creating a seamless scene.

Old Generative Fill (left) and New Generative Fill (right)

When the image is expanded, we notice some artifacts in the background that may not be ideal. It also seems that the flowers may make the AI think the woman is perhaps on a table. The new result, however, recognizes that it is the floor and keeps the output clean and straightforward.

Old Generative Fill (left) and New Generative Fill (right)

The latest updates to Cloudinary’s Generative Fill bring greater accuracy and quality to image outpainting fills, making for faster content creation without specialized editing skills, adaptable visual storytelling through automatic backgrounds, and rapid content scaling across platforms. Lastly, the time savings is big, especially for those launching complex campaigns that demand varied visual assets.

Try out the new Generative Fill to experience the difference, and sign up for a free Cloudinary account today.