Tech

Review: Recraft AI Image Generator Puts Humans Back in the Loop

For artists, creating AI images is often a depressing experience.

As a professional photographer, my job is all about creative control — I pick my subject, frame my shot, adjust my aperture/speed/ISO, and choose the perfect moment to fire the shutter.

The result are photos that capture my specific creative style and vision.

Creating an AI image with today’s leading AI image generators is totally different. It’s more like visual Russian roulette — you plug in a prompt, press a button, and hope that the end result looks something like your vision.

If you push the button again, the system will create images that look entirely different from your first run. There’s no creative control.

Now, a new model that’s surging onto the scene is quickly changing that. Recraft is the most powerful image generator on the market today. And it’s succeeding specifically by putting humans back in the loop.

The Enigma of the Panda

Recraft launched its powerful new V3 image generation model with a clever tease.

The company began participating in the Artificial Intelligence Image Arena, an ongoing public…

Recraft’s V3 model quickly flew to the top of the leaderboard. But the company chose to release it under a code name, Red Panda.

That prompted immediate and rampant speculation about who was behind this new model. Was it the latest version of DALL-E, released early and anonymously by OpenAI? Some kind of stealth Chinese competitor secretly beating the world?

Eventually, Recraft revealed the identity of their new “Red Panda” model. V3 is powerful in its own right. But it’s doubly so because it puts the process of image creation back into the hands of humans.

Best of all, you can use it for free. As soon as Recraft revealed Red Panda’s true identity, I created an account and began to test the model out.

Back in Control

The experience of using Recraft V3 starts the same as pretty much every other AI image generator on the market.

You enter a prompt, select some parameters, and let the system generate a few images. Recraft is fast, and as its leaderboard position indicates, powerful and realistic.

Here’s an image of “an incredibly beautiful and realistic Autumn scene”

Screenshot of the Recraft interface

With many AI image generators, this would be the end of the process. But with Recraft, creating an initial image is just the beginning.

Once your image is in place, you can begin to use Recraft’s suite of tools to edit and add to it.

The system is structured a bit like an online version of Photoshop. Your image exists on a canvas, where you can make adjustments, add additional elements, create layers, and more.

Some of these are basic, but powerful. Recraft has an Eraser function that lets you quickly remove things from your generated image, replacing them with a new, seemless AI-generated background.

If I want to get rid of the leaves on my AI-generated road, for example, it’s a single click.

Screenshot of using the Eraser tool
Results of the tool

That might seem simple. But the ability to quickly remove erroneous elements from an AI image is extremely helpful.

Like Large language models, AI image generators often hallucinate strange objects or design elements that shouldn’t be in your scene. The ability to quickly remove them without needing to load the image in Photoshop is a big time saver.

Recraft also lets you merge real images with AI ones, and edit them using the tool’s AI functions.

I can upload a photo of my own, all-too-real minivan into my AI image, use Recraft to automatically remove the background, and resize, flip, and reposition the image.

Integrating a real photo into the scene

I can even change the color palette of the real image. Purple Odyssey, anyone?

Changing the color scheme with AI

You’ll notice that the image isn’t perfect. Because of the angle of my original photo, the driver-side rear wheel well isn’t visible.

Recraft lets me fix that, too. I can circle the missing wheel well with Recraft’s Lasso tool and type in a prompt, like “Add the missing wheel well.”

Recraft will create a new one with AI and add it in.

The resulting image

The ability to add objects goes much deeper than making simple fixes, though.

I love cows. This image needs cows.

Adding one is a simple matter of selecting a relevant area and asking for more cows to be present in it.

Adding cows

Much better!

More cows!

Recraft can add far more fanciful things, though. Let’s put a flyer saucer in the sky. Again, I just Lassoed the blank area and asked for “A flying saucer UFO.”

My final, bizarre image

Recraft is excellent at handling text, too. You can generate text directly in the system’s AI image generator, which I’ve found is much better at creating accurate text than other competitors.

But Recraft also provides an even simpler solution. With the system’s Text tool, you can type out the exact text you want.

Using the Text and Frame tools

You can then use the Frame function to seamlessly blend it into the AI image; Recraft will consider the full scene and integrate the text using a visual style and color scheme that fits the vibe of the scene.

The text fits the image’s vibe

The end result is an image that’s just as bizarre as many AI-generated images you’ll find on the Internet today.

But crucially, it’s my bizarre image.

I’ve customized every aspect of it, adding text, real-world imagery, and additional design elements that match my concept of what the image should be. It’s created by AI, but it’s also a product of my own human process.

This is clearly a silly example. But Recraft excels at making genuinely useful images, too.

Imagine that you need an AI illustration for your food-themed Facebook page.

You can easily ask Recraft to generate realistic, seasonally appropriate images with integrated text. And you then then edit them to make them exactly how you want.

Generating a food-themed image with overlaid text

The system excels at creating logos, product mockups, and much else, too.

A Recraft-generated logo for a fictional accounting firm

It’s a unique combination — the power of a world-class AI image generator, merged with a suite of tools that lets you inject your own human creative vision deeply into the resulting images.

Back in the Driver’s Seat

One of my biggest takeaways from testing Recraft is that using the software is hard.

That’s a really good thing.

I don’t mean that the Recraft interface is clunky or confusing, or that the software is overly complex. On the contrary, Recraft has done a great job building a straightforward and feature-rich product.

What I mean is that — like using Photoshop or learning the settings on your DSLR camera — working with Recraft to make images takes work.

You need to experiment with different prompts to create the elements you want in your image, come up with the perfect text to complement the visuals, add your own human-created photos or illustrations, remove backgrounds, delete visual elements, pick and consistently implement a color palette, and more.

That’s very different from using existing AI image generators like Midjourney or DALL-E.

Again, those programs are typically a one-shot operation; you type in a prompt, and the machine does the rest.

Yes, DALL-E offers simple editing tools. But they provide nowhere near the level of control that you get in Recraft. You’re still fundamentally relying on random chance to get a usable image.

With Recraft, the need for intense human intervention in the process is a powerful feature, not a bug. Recraft isn’t just spinning up images at random, from the murky depths of its neural network brain — your human effort, ideas, and creativity are a key part of the process.

And when you get a final image out of Recraft, it’s shot through with the fruits of that human effort, and your own human ideas — even if a computer helped you make it.

For a creative, that level of control and hands-on interaction is extremely heartening. Working with Recraft feels like using a totally new tool for realizing your creative vision, not like outsourcing your creativity to a fundamentally dumb, soulless machine.

Protectable Images

Importantly, the level of effort and knowledge required to use tools like Recraft could offer another tangible, valuable benefit for artists: copyright protection.

Despite the best efforts of certain artists, the US Copyright Office has said consistently that the output of pure AI image generators cannot be copyrighted.

The office, though, has left open the door for the possibility that images combining AI elements with true human work and creativity could eventually be protectable under America’s existing copyright system.

I recently attended an industry conference where five of the world’s top legal minds in the copyright space sat around a table and discussed the future of copyrights and AI.

They all agreed that the copyright office is likely to eventually allow protection for images that integrate a hefty dose of human creative effort, even if those images also use some AI.

The key question, they all agreed, was how much human effort is enough to make an image protectable. The Copyright Office has made clear that prompt-based, one-shot image generators don’t qualify, even if the prompt that an artist uses is long and complex.

But the images created by a human-in-the-loop tool like Recraft just might. An image like my “Typical California Day” creation does include AI elements, and much of the process of creating it relies on the power of AI.

But it also includes my own human decisions and ideas — as well as original imagery produced in the real world, like the shot of my minivan.

As tools like Recraft better integrate AI with human creativity, it’s possible that the images made with them will eventually be copyrightable. This would open up all kinds of legal protections for artists, and make AI images easier for brands and buyers to use.

In short, adding the human back into AI image creation doesn’t just enrich the creative experience for the artist — it may ultimately unlock all kinds of benefits and protections that are currently only available for fully human works.

Personally, I’m not worried about copyrighting my Recraft images. I’m just delighted to have an AI image generator that lets me control the creative process — and that can handle practical tasks like overlying text.

Recraft is powerful in its own right. But combine it with a human artist, and the possibilities grow exponentially.

Want to test it out for yourself? Recraft is currently free. Head to Recraft.ai to check it out.

I’ve tested thousands of ChatGPT prompts over the last year. As a full-time creator, there are a handful I come back to every day. I compiled them into a free guide, 7 Enormously Useful ChatGPT Prompts For Creators. Grab a copy today!

Thomas Smith

Thomas Smith is a food and travel photographer and writer based in the San Francisco Bay Area. His photographic work routinely appears in publications including Food and Wine, Conde Nast Traveler, and the New York Times and his writing appears in IEEE Spectrum, SFGate, the Bold Italic and more. Smith holds a degree in Cognitive Science (Neuroscience) and Anthropology from the Johns Hopkins University.

Leave a Reply

Back to top button

Discover more from Bay Area Telegraph

Subscribe now to keep reading and get access to the full archive.

Continue reading