Join OpenSea’s flagship series……
Developed by Elman Mansimov, alignDRAW is the first text-to-image AI model (born in 2015). It generates blurry images with a resolution of 32 x 32 pixels—among the very first images ever created from text prompts.
Details as follows👇
Elman Mansimov was born in Azerbaijan in 1996. At the age of 15, he began studying computer science at the University of Toronto. He was mentored by Ruslan Salakhutdinov, and by early 2015, neural networks were already able to recognize photos and write captions for them.
At 19, Elman proposed an obvious yet inverse question: can a neural network read a sentence and draw it? This led to the creation of the alignDRAW model. The paper describing the model (“Generating Images from Captions with Attention”) (which generates images from captions using an attention mechanism) was submitted in November 2015. Co-authors include Emilio Parisotto, Jimmy Lei Ba, and Ruslan Salakhutdinov. The paper was later accepted by the ICLR 2016 conference for an oral presentation.
The model generates 2,709 images from 36 text prompts. Each image has a resolution of 32 x 32 pixels and is difficult to discern. Of the prompts, 21 appear in published papers; the remaining prompts—i.e., the “process prompts”—were released on a webpage at the University of Toronto, and they can still be viewed via http://archive.org (archived on November 2015).
One of the works presented in this article is: “a huge commercial airplane parked on a surfboard floating on water”—the 92nd image out of the 121 the model generated for that sentence. The prompt is the title of the work, and the blurred image is the art itself.
Over the next nearly six years, work in the field remained largely quiet. Elman earned his PhD from Kyunghyun Cho at New York University (NYU), and he is now a senior applied scientist in the New York division of Amazon.
alignDRAW existed only in academic citations until DALL-E, Midjourney, and Stable Diffusion made “text-to-image generation” a household phrase, prompting people to trace the source of this creativity.
In April 2023, the artist Alejandro Cartagena reached out to Elman to discuss the matter of putting the original images on-chain. Fellowship ( @fellowshiptrust ) published these images as NFTs. Each NFT stores the original images generated in 2015 in native byte format and achieved fully on-chain storage in May 2024.
After that, the art world quickly followed suit: alignDRAW appeared in succession at the Paris Photo fair in November 2023, at the HERE space on Cork Street 9 in London, and later that same year in December in Christie’s online auction.
In 2024, the Worcester Art Museum acquired three text-prompt-generated “alignDRAW” series prints and included them in the exhibition titled “New Terrain: 21st-Century Landscape Photography.”
In the same year, in June, the work made an appearance during Art Basel in an exhibition titled “Collaborations with the Artificial Self,” hanging between Harold Cohen’s “AARON” system and the works “Infinite Images” by Holly Herndon and Mat Dryhurst.
The international head of Photography at Christie’s, Darius Himes, remarked: “This is AI’s Nicéphore Niépce.”
Lev Manovich also holds the same view: “the images generated by alignDRAW are on par with the first permanent photographs captured by Niepce between 1826 and 1827.” To mark the project’s tenth anniversary, Elman reconstructed the original model in 2025 using modern hardware, while the Fellowship organization published a monograph that includes 2,709 images and essays exploring the origins of this experiment.
Incorporate alignDRAW 2438 into our flagship collecting series, as a tribute to this pioneering attempt to turn written language into images: a work created in 2015 by a 19-year-old—an image of a “plane on a surfboard” made up of 32 x 32 pixels. Even a decade later, it is still continuously reprinted, rerun, and generating buzz.
Welcome home, 2438 🌊
