Recipe eight: How to create a cloning effect with exactly.ai

1. One of our clients requested a series of images that used a cloning effect.

They sent us reference images where the backgrounds were cloned to form a pattern within the subject in the foreground. Rather than simply replicating these, our task was to create a new series featuring different subjects, all while achieving the same visual effect.

2. At first, we tried solving this with traditional approaches.

We loaded the reference images into various tools like Midjourney, Leonardo, and Stable Diffusion. Despite numerous attempts and combinations of prompts, none of these tools produced the desired effect. That’s when we turned to exactly.ai to create a custom model.

3. As soon as we loaded the references into exactly.ai, it was clear that the standard prompt descriptions wouldn’t cut it.

We needed the model to understand that the object should seamlessly replicate the background. Breaking away from the usual rules, we wrote highly detailed descriptions for each image in the dataset, explaining the specific requirements for cloning.

4. To our surprise, exactly.ai immediately produced the cloning effect we’d been chasing.

After countless failed attempts with other ai tools, this was a breakthrough. The model handled both flat and spatial backgrounds with ease, delivering impressive results across different mediums. Here are some examples of how the model interpreted cloning in the styles of different artists:

5. The key to success was in how we structured the prompts.

We divided each prompt into two sections: the first describes the scene and its elements, while the second – the "cloning directive" – instructs the AI to ensure the object visually replicates the background. Here’s an example:

Here are two images generated from that exact prompt:

6. Then I experimented with different prompts, to generate a wide variety of cloned visuals.

We even experimented with swapping out the cloning prompt for something from our earlier experiment with piles of clothes, like this: Silhouette in the form of a collage of clothes on a yellow background. It’s hard to believe that from just six initial reference images sent by the client, exactly.ai was able to generate such a vast range of cloned variations.

7. Naturally, we couldn’t stop there.

Building on the success of the first cloning model, we developed another, pushing the creative possibilities even further. The new model generated captivating results with more intricate patterns. We also tested how the cloning model performed with objects not included in the original dataset, like cars, and the results were equally impressive.

8. So, what’s next? Exactly 3.0 has just been released, and the updates have introduced some significant improvements.

I’ve already started experimenting with a third cloning model. Stay tuned – we’ll be sharing the results soon as we continue exploring this exciting new frontier!

Ready to explore the creative possibilities of exactly.ai? Sign up now.