From the makers of Stable Diffusion, Stable Doodle is the company’s latest free text-to-image and image-to-image combo. It takes your poor attempts at mouse-drawn images and converts them into a more cohesive AI illustration. The service is free, and users on the Stability AI-owned Clipdrop platform can use their mouse (or a drawing tablet) to create a simple black and white line drawing that can transform into a more interesting image in a variety of styles.
There are plenty of AI systems that let users add an image alongside a prompt when generating an AI output. According to Stability AI, their system uses the Stable Diffusion XL model alongside other adapter systems to add additional inputs into the text-to-image generator. The company said the T2I-Adapter licensed from Tencent ARC lets their regular system understand the outlines of sketches and generate watermark-laden images based on that plus an additional text prompt.
The company used several examples, from a simple chair to a more sketched out drawing of a living room to show the AI is capable of interpreting both simplified and more complex drawings into more-realized art. The company says this tool could be useful for designers or illustrators who want to “free up valuable time and maximize efficiency.”
As if they’re not already worried that companies will be incentivized to cut their jobs if a CEO thinks their lazy doodle could pass for harmless, if not completely ineffective, interior design proof of concept.
The Stable Doodle system still doesn’t work without the text prompt, and you can’t import your own sketches doodled up on paper. Gizmodo tried to put the system through its paces with both simple and more complicated prompts.
The system doesn’t do too well with complicated images. It can take a scribbled duck on a lake and turn it into an AI artwork just fine. Once you tell Doodle that the duck is smoking on a cigarette, and the AI will put lumps of steaming coal into its beak.
It’s also unclear just how much stock AI actually gives to interpret user-made images versus the prompts. I drew a picture of a tree surrounded by stumps, but included a small hollow on the main subject with a creature living inside. My prompt “A tree in winter surrounded by tree stumps” produced several well-crafted images, but no matter the style only about half of them included the hollow, and none tried to stick a little creature in there.
Stability AI did mention their system has limitations. The company wrote that “Stable Doodle uses algorithms to analyze the outline of an image to generate a visually pleasing and coherent result” adding that the final image’s accuracy “may vary depending on the complexity of the scene.”
Stability AI has been trying to stand out from the ever-growing pack of AI developers by proclaiming itself as the heralds of open source AI models. That doesn’t mean there aren’t other AI models meant to turn line drawings into “art.” Services like Fotor and Playform also advertise similar sketch-to-art services, though varying degrees of costliness and capability.
Closing on a year since DALL-E and Stable Diffusion 1.0 took the web by storm, you’ve probably seen your fair share of AI art generators. You might be thinking they’re old hat at this point, the whole lot of them nothing better than highly-capable rip-off machines. A few newer models are still trying to innovate on what came before, and alongside text-to-video, we’re starting to see more companies try and innovate with different use cases. Doubtless illustrators and designers are thrilled with AI that can jazz up awful art into less-awful (though still extremely derivative) art.
Want to know more about AI, chatbots, and the future of machine learning? Check out our full coverage of artificial intelligence, or browse our guides to The Best Free AI Art Generators, The Best ChatGPT Alternatives, and Everything We Know About OpenAI’s ChatGPT.