InvokeAI

If you've been on the internet lately, then it's likely that you've been a witness to the constant flood of AI-generated works. For better or worse, this marks a new age in the creative fields, where a computer is able to generate poetry, artwork, and many other things solely based off of a few prompts.

That does sound like some futuristic scenario, but it is indeed possible to have an AI model generate content based on your prompts, on your very own computer. If AI-generated artwork is of interest to you, then you may find InvokeAI's offering worthy your consideration. It's a creative engine for several AI models, granting the end user a lot of freedom when it comes to parametrizing art generation.

This is quite the exhaustive tool in regard to generative art. With how much you're allowed to configure your settings to influence the end result, it's likely that a good chunk of the AI art you've seen was made through InvokeAI.

Before hopping in, it's worth noting that you're in for a rather lengthy installation process, and you'll need some competent hardware to boot. Last-gen midrange graphics cards should work fine, but if you happen to have a lesser configuration, you can still run the software. Just keep in mind that generating art can take a good while on lower-tier machines.

Regardless, once you're into the browser interface, you'll be able to start dictating how you want your generative art to be handled. Several models are available for use, with Stable Diffusion 1.5 being the default choice. If you're after other types of art, it's worth consulting the other options, as each has their respective strengths and weaknesses.

You can then go right ahead and decide on the prompts to use. You can get as creative as you'd like here, and as per usual, the more specific, the better. The same applies to the number of generative steps used: the more steps, the more accurate the resulting artwork.

After subsequent generations, you'll be able to train your AI model to better respond to your demands. This is done via the Seed and Variations menu, where you'll be able to set thresholds and variation amounts. If your model is having trouble generating faces, you can have a go at Face Restoration, which uses an algorithm to spot any irregularities and correct them accordingly. You're free to modify the strength of the chosen algorithms, of which there are two: gfpgan and codeformer.

Your art can also be upscaled to enhance picture clarity. Choose a scale and algorithm strength that's right for you, and the ESRGAN upscaling architecture will improve the presentation of your generated image.

If you wish to train your AI to generate art that is akin to something specific, you can try using the Image to Image feature. Through it, users are able to upload a series of images to help dictate subsequent generations. This marks a change from simply using seeds and prompts, as having other visual elements as a source can greatly improve the accuracy of your end result.

Users also have the option of utilizing Unified Canvas. By way of this function, you can take advantage of Inpainting and Outpainting, which work to refine certain parts of an existing image. The former simply refines your image within the bounds of your marked area, while the latter extends beyond it.

There's a lot to InvokeAI: it's an exhaustive toolkit for generated art that fully harnesses the power of AI, while giving rein to the end user to influence the resulting image in more ways than one. For those reasons, it's a piece of software worth trying out, even if you're not all that intrigued by AI art, because you can never quite know what your model might generate in the end.