Exploring Insmind's IA: creating and retouching images with mixed results

Continuing with my tools tests based on artificial intelligence applied to the image, this time I decided to test Insmind, a platform that integrates image generation and editing in a pretty intuitive way. The interesting thing about this tool is that it is not limited only to creation from scratch: it also allows you to expand images, delete elements, replace them or even improve your resolution.

In this article I share my experience testing several key functions, using as a practical case the image of a playful rabbit interacting with a ball.

Starting point, image generation

The first test was to request the IA to create an image with this description:

In the plane there is a rabbit of fallen ears, white with clear brown spots, on a beige surface. The rabbit pushes and bites a yellow and red plastic ball, showing itself playful and curious. His movement transmits tenderness and energy, as if he was entertaining with his favorite toy. The camera remains fixed, focusing on the rabbit as it interacts with the ball.

The result was very correctboth in composition and in detail. The image generated faithfully represented the scene: the rabbit was well defined, the expression was consistent with the text and the colors of the environment were harmonious. Insmind responded quickly and with remarkable quality at this stage.

"Expand" function: image extension without loss

Then I tried the tool Expand, which allows to extend the canvas and generate consistent content on the edges. Here the performance was excellent. The image expanded naturally, maintaining the continuity of the background and texture of the soil without the change being noticed. Very useful to adapt proportions without affecting general aesthetics.

"Erase" function in Insmind

The next step was to try remove the ball with the tool Erase, requesting that the area be rebuilt with the normal legs of the rabbit. This is where the tool showed its most obvious limitations.

Although he technically removed the ball, the result was unrealistic: the legs were distorted, with a kind of diffuse or even deformed effect that made it look like the animal was "burning" below. A clear example of how far the current model reaches in complex tasks of rebuilding key elements.

Replace function

To solve the previous problem, I tested the option Replace, which allows to replace selected elements with new ones generated by IA. In this case, I changed the yellow and red ball for a Orange basketball.

The result was positivebut with a small side effect: the IA unexpectedly modified some elements of the environment, such as the rabbit nose, which changed slightly in shape. Nothing dramatic, but something to consider if you are looking for a precise and localized edition.

Conclusions: good basis, room for improvement in complex edition

Insmind is a tool with a very friendly environment and powerful functions for create, expand and replace visual content quite effectively. The generation model responds well, especially with detailed descriptions, and the overall performance is fast and fluid.

However, in more complex tasks such as precise reconstruction, the tool still has limitations. Although it may be very useful in basic or prototyping retouching work, in projects where total visual fidelity is required, it is necessary to review and possibly retouch the results manually.

As a point in favour, all these tests were carried out in the free version of the tool, which makes it an attractive option to explore IA solutions without having to invest from the start.