Taking a break from complaining about Google Gemini’s racially confused AI imagery, Marc Andreessen and his compatriots at the mega VC firm Andreseen Horowitz (a16z) have upped their investment in a rival image generation startup, Ideogram, leading the firm’s $80 million Series A financing, the two companies announced today.

In addition, Martin Casado, General Partner at a16z, is joining Ideogram’s board, and the company today announced a new version of its “trained from scratch” image generating model, Ideogram 1.0, that “offers state-of-the-art text rendering, unprecedented photorealism and prompt adherence, and a new feature called Magic Prompt,” according to a blog post by the company.

Ideogram 1.0 is currently available for users to try for free on the company’s site, though it requires a Google or Apple account to log in. Users can also generate images from within the company’s Discord server.

The company offers a free tier for users capped at 100 image generations per day, alongside monthlty subscriptions allowing 400 daily images ($7 monthly, billed annually as Basic tier), and unlimited daily images ($16 per month, billed annually as Plus tier).

VB Event

The AI Impact Tour – NYC

We’ll be in New York on February 29 in partnership with Microsoft to discuss how to balance risks and rewards of AI applications. Request an invite to the exclusive event below.

 


Request an invite

Also joining in Ideogram’s $80 million Series A are prior investor Index Ventures, alongside newcomers Redpoint Ventures, Pear VC, and SV Angel. The company’s seed round was previously led by a16z and joined by AIX Ventures, Golden Ventures, Two Small Fish Ventures, and industry experts Ryan Dahl, Anjney Midha, Raquel Urtasun, Jeff Dean, Sarah Guo, Pieter Abbeel, Mahyar Salek, Soleio, Tom Preston-Werner, and Andrej Karpathy.

Text and typography generation in AI images is no longer a uniquely differentiating feature

Founded by ex Google Brain AI researchers, Ideogram made waves when it first debuted in August 2023 by offering text and typography baked directly into AI generated images, something that rivals such as Midjourney did not offer at the time.

However, the game has since changed considerably since then, with not only Midjourney introducing text generation within images as part of its V6 release, but also OpenAI’s DALL-E 3 introducing the feature for its users as well (accessible via ChatGPT). For example, it is possible to have AI images generated with characters holding up signs that have messages printed on them, or storefronts with legible signage.

Nowadays, letter formation is more accessible through a host of AI image generators, which means Ideogram’s differentiator is lessened. Still, the results are not always consistent or aligned with what the user prompted, and Ideogram pointed out in its blog post announcing 1.0 and the Series A today that its research shows human evaluators prefer Ideogram over Midjourney V6 and DALL-E 3. See the following graphs to support these fundings:

Ideogram graph showing user preferences compared to AI generations by DALL-E 3 and Midjourney V6.
Ideogram graph showing user preferences compared to AI generations by DALL-E 3 and Midjourney V6. Credit: Ideogram

Yet Ideogram also stood out when it first launched by offering users the ability to select from a range of pre-curated image styles such as “3D rendering, cinematic, painting, fashion, product, illustration, conceptual art, ukiyo-e.” Now the website has further evolved to include options for different aspect ratios, image weights, public/private visibility of generations to other Ideogram users, and a toggle to turn on the new Magic Prompt feature (described further down in this piece).

But Ideogram still seeks to stand out with new features

Ideogram isn’t resting on its laurels, either. The company’s newest feature, Magic Prompt, automatically expands on user-inputted text prompts to make them more descriptive and detailed, producing ideally more high quality imagery.

However, this feature too is similar to OpenAI’s integration of DALL-E 3 image generating AI with ChatGPT, which also takes a user’s text prompt and modifies it automatically with new, more vividly descriptive language, all in the background and invisible to the user.

These features make sense in that they can help a user better communicate with the underlying AI model, effectively translating what the user wrote into a more machine-friendly format, without the user having to do the work of trial and error (though in our experience, some of this is also still required).

While undoubtedly exciting for the burgeoning AI art community and potentially enterprise users and marketers, the advent of Ideogram 1.0 and its continued funding to build out the product will also likely lead to an increase in spammy AI image generations — already a growing problem on the web — as recent examples show.

VentureBeat uses Ideogram, Midjourney, DALL-E 3 and other AI tools to create article imagery.

VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.


Source link