Ideogram launched in August 2023 with the objective of fixing the infamous textual content downside that has plagued AI picture fashions. Even as these instruments have turn out to be adept at producing amazingly sensible scenes and characters, any textual content included in pictures – from protest indicators to t-shirt slogans – usually seems warped past recognition.

Last fall, when Ideogram debuted its software program, opponents like Midjourney, DALL-E 2, and Stable Diffusion struggled mightily to deal with textual content. But the sphere has seen fast advances since then. Stable Diffusion 3.0 focuses closely on textual enhancements, whereas DALL-E 3 can now produce some legible phrases and phrases in pictures.

Even so, Norouzi believes Ideogram’s distinctive strategy outperforms rival fashions. The newest model boasts increased textual content accuracy charges general, he says, and reveals particular ability at dealing with prolonged, advanced sentences. Just have a look at this instance from final 12 months produced with the immediate “a photograph of an adorable kitten wearing a t-shirt with the words ‘ask me about my AI startup” on a number of picture fashions.

Clockwise from prime left: Ideogram, OpenAI’s DALL-E 2, Stability AI’s Stable Diffusion, and Midjourney. (Image: Bloomberg/Ideogram)

There’s a transparent winner right here.

The new software program additionally features a new characteristic referred to as “magic prompt” that routinely expands on the written prompts customers submit. For instance, it would construct on a easy phrase like “a cute pika with bumblebee antennae” by producing extra descriptive sentences in regards to the pika’s stance, expression, and different particulars.