Toronto-based artificial intelligence (AI) startup Ideogram has raised $80 million in Series A funding, as the startup releases its “1.0” update to its text-to-image generation platform and new subscription model.
“They’re innovating on an intuitive product that acts as a printer of imagination to millions of users.”
-Andreessen Horowitz general partner Martin Casado
American venture capital firm Andreessen Horowitz led the round, with participation from returning investor Index Ventures and new investors Redpoint Ventures, Pear VC, and SV Angel. Andreessen Horowitz co-led Ideogram’s $22.3-million CAD ($16.5 million USD) seed round following the startup’s launch just six months ago.
Martin Casado, a general partner at Andreessen Horowitz, has joined Ideogram’s board of directors as part of the financing.
“We have been investors in Ideogram since their seed round, and are very excited to lead their Series A as they continue to push the boundaries of generative model research,” Cosado said in a recent blog post. “They’re innovating on an intuitive product that acts as a printer of imagination to millions of users.”
Much like DALL-E, Midjourney, and other AI text-to-image generators, Ideogram allows users to simply type a prompt and click “generate.” Within 30 seconds, Ideogram will provide four different image interpretations of the provided prompt that users can download and freely use, or generate new image batches until they are satisfied.
RELATED: Ideogram launches with $22.3 million CAD for generative AI text-to-image platform like DALL-E
Ideogram said in a blog post that its Series A financing was raised to accelerate its growth and build more capable generative media models. The company added it is hiring for roles in its engineering, research, design, and operations departments.
In addition to announcing its Series A round, Ideogram this week unveiled what it’s calling its new text-to-image model, called “Ideogram 1.0.”
The company says the new model is “trained from scratch” and is its most advanced text-to-image model to date. Alongside improved photorealistic displays, Ideogram claims the update substantially reduces error rates in rendered text, with output images displaying legible and coherent text in a variety of fonts and styles. Text generation has been a known weakness of AI-generated images, often generating requested text into an unreadable script.
By way of comparison, the feature image of this article was generated with the prompt “Racoon wearing a Toronto Raptors jersey in Toronto with cinematic 3D rendering.” BetaKit used the same prompt six months ago and received this image.
The new model also improves prompt adherence, Ideogram says, which means it can understand longer and more complex prompts. The model also features a “magic prompt,” giving users the ability to input a simple prompt and allow the model to extrapolate a longer prompt. Ideogram provides an example of three emojis being interpreted into a four-sentence prompt that generates a Halloween-themed image.
The final component of Ideogram’s 1.0 update is a new subscription model, which will still allow free use of 100 generated images, or 25 prompts, per day. The basic subscription tier costs $7 USD per month and allows 400 uncompressed images per day while the plus subscription tier costs $16 USD per month and allows unlimited image generation.
Ideogram’s CEO Mohammad Norouzi was previously a senior staff research scientist at Google and worked on Google’s text-to-image system Imagen with fellow Ideogram co-founders Chitwan Saharia, William Chan, and Jonathan Ho.
“Mohammad, Jonathan, William, and Chitwan are among the pioneers of diffusion models — and they’ve created a consumer-grade user experience combined with state-of-the-art model research,” Cosado wrote.
Feature image generated by Ideogram with Alex Riehl prompt: “Racoon wearing a Toronto Raptors jersey in Toronto with cinematic 3D rendering.”
The post Midjourney competitor Ideogram closes $80-million Series A as it launches latest text-to-image model first appeared on BetaKit.
Originally published on BetaKit : Original article