Signals Inbox·August 26, 2026·AI Creative Tools

Why is Gamma buying Lica?

Gamma is buying Lica to own more of the design intelligence behind its AI creation tools: structured layouts, typography, editability, animation and brand control. Lica gives Gamma a research team already working on the problems that become important once generating a decent-looking first draft is no longer enough.

Q1What exactly did Gamma buy when it acquired Lica?

Gamma mainly bought a small design-research team that can help it build much better AI design technology.

The deal is unusual if we look at it as normal software M&A. Lica was founded in 2023, had raised just $4 million from Accel, South Park Commons, Village Global and others, and never disclosed a large customer base or meaningful revenue. Gamma, meanwhile, already had more than 100 million users by the time the acquisition was announced. Buying Lica was never going to move Gamma's revenue or distribution much.

The way Gamma structured the acquisition gives us a better clue. Lica co-founders Priyaa Kalyanaraman and Purvanshi Mehta are joining Gamma to lead a new design research lab. Gamma has not announced plans to keep Lica running as an independent product, and the purchase price was not disclosed.

TechCrunch's reporting on the deal also included a revealing comment from Gamma CEO Grant Lee: Gamma still sees presentations as a core use case, but the company now wants to explore what presentations could become when they are more visual, interactive and fluid.

Gamma is paying for Lica's researchers, technology and accumulated knowledge about how AI should understand design.

Q2Was Lica really another AI presentation startup?

Lica started close to the AI presentation market, but by the time Gamma bought it, the company had moved much deeper into AI design research.

Lica's first product turned screenshots and screen recordings into presentations and product videos. That explains why Gamma and Lica were already operating around the same basic problem: helping people communicate visually without spending hours manually designing everything.

But Lica kept moving. The company later worked with e-commerce businesses to generate marketing videos that followed brand guidelines. More recently, its public work shifted heavily toward research on how AI understands layouts, typography, editable layers, vectors, animation and brand identity.

Its website these days describes Lica as building the underlying technology needed to train and evaluate AI systems that can behave like design partners. The language is much closer to an AI research lab than a lightweight presentation app.

Gamma did not need another prompt-to-presentation product. Lica had already started working on harder problems underneath the product Gamma wants to build next.

Q3What did Lica actually build that Gamma needs?

Lica built one of the largest structured graphic-design datasets we have seen, along with benchmarks and tooling for teaching AI how real designs are constructed.

The centerpiece is LICA, short for Layered Image Composition Annotations. The research team assembled about 1.55 million graphic-design layouts across 20 categories and roughly 972,000 unique templates. Those designs contain more than 24 million individual components.

The important part is the structure. A normal image dataset gives an AI model the finished pixels. LICA keeps the text, images, vectors and groups separate, along with information about where they sit, how large they are, which fonts they use, their opacity and how the elements relate to one another.

Nearly 449,000 layouts are presentations, representing about 29% of the whole dataset. Around 600,000 are Instagram posts. The rest includes education material, flyers, business documents, videos, posters, infographics and other visual formats.

That mix lines up remarkably well with where Gamma is going. Gamma began with presentations, but its product now covers social posts, documents, websites, images and standalone visual assets too.

Q4Is AI still bad at the kind of design Gamma needs?

Yes, current AI models are surprisingly weak once Gamma asks them to make precise design decisions instead of simply generating something attractive.

Lica built GraphicDesignBench specifically to measure this gap. The benchmark tests frontier models on 49 tasks across layout, typography, infographics, templates and animation.

The result was rough. Lica classified only two of those 49 tasks as mostly solved. Twenty-five were partially solved and 22 remained unsolved.

Some individual failures show how far there is still to go. Component detection scored only 6.4% mean average precision. Font recognition topped out at 23.7% across 167 font families. Lica also found that models often performed reasonably on simple single-element tasks and then deteriorated sharply when several elements had to interact correctly.

That is very different from the impression people get from looking at the best AI-generated images online. GPT, Gemini and other models can produce beautiful graphics. Professional design asks for something more annoying and much less glamorous: keep this exact layout, use this exact font, move this object, preserve everything else and do it again across 30 variants.

Today's models still struggle badly with that.

Q5Why can't Gamma just use GPT, Claude or Gemini?

Gamma can keep using frontier AI models, but those models currently do not give it enough control over the design itself.

Gamma already works this way. The company integrates models from outside providers rather than trying to train one giant general-purpose foundation model from scratch. That makes sense: spending billions to recreate GPT or Gemini would be absurd for Gamma.

The difficult part sits one level above the model.

Imagine asking an AI system to replace one photograph in a presentation while keeping every object exactly where it was, preserve a company's typography across 40 slides, turn a landscape design into a vertical social post, animate three elements in the correct sequence and then produce another version for a different audience.

The underlying model can help with all of those jobs. But Gamma still needs software that understands what each object is, what can change, what must stay fixed and how all the pieces fit together.

Lica spent much of its research effort on exactly that layer. Its engineering team was even developing a custom design file format built for AI agents, where designs remain structured and editable rather than becoming finished pictures that have to be regenerated whenever something changes.

Gamma can therefore keep benefiting from every improvement OpenAI, Anthropic and Google make while owning more of the design intelligence that turns those models into a useful product.

Q6Why is editability becoming such a big deal for Gamma?

Editability is becoming critical for Gamma because its most serious users do far more than generate a presentation once and leave it untouched.

Gamma's recent Creation Index gives us an unusually good look at this behavior. The company analyzed more than 625 million lifetime creations made by over 100 million people, using aggregated product data through earlier this year.

The most interesting group is Gamma's top 1% of creators. Gamma calls them "supercreators," and they produce 11% of everything made on the platform. More importantly for the Lica acquisition, they make nearly twice as many revisions to each Gamma as the average user.

Gamma also found a broader shift away from blank-prompt generation. In the first quarter of 2025, 59% of creations started from scratch. One year later, that had fallen to 31%. Meanwhile, creation from users' own notes, documents, decks and transcripts had risen sharply.

Gamma is increasingly being used to transform and refine existing work rather than simply generate something from nothing.

That pushes the product toward exactly the problems Lica has been studying. The more users revise, reuse, adapt and personalize existing material, the more frustrating uncontrolled AI generation becomes. Moving one thing while accidentally changing three others stops being a small technical flaw. It becomes a core product problem.

Q7Is brand consistency one of the real reasons Gamma wants Lica?

Yes, better brand control is probably one of the most commercially useful things Lica can bring to Gamma.

Gamma has been pushing further into work done by teams and companies, where "make it look good" is rarely enough. Businesses want the right fonts, colors, logos, layouts and visual style every time.

Gamma has already been moving this way. Users can create custom themes, bring visual references into image generation and use Gamma Imagine to produce posters, diagrams, infographics, social graphics and other assets. Imagine can also apply a saved Gamma theme so generated graphics follow its typography and colors.

Lica approached the same problem from the research side. Before the acquisition, its commercial product generated creative assets for brands, and its research explicitly focused on personalization and identity. The goal was to teach AI systems how to understand a brand from its colors, fonts, visuals and previous designs, then produce new work that stays consistent.

Current models remain shaky here. Lica's own experiments found that leading multimodal models were often able to describe the general feel of a design while missing exact fonts, spatial positions and other details that a real brand team cares about.

For Gamma, fixing that opens a much more valuable use case than prettier one-off decks. A company could potentially generate dozens of presentations, sales materials, social assets and localized variations while keeping the same visual identity automatically.

Q8Is Gamma still mainly a presentation company?

Gamma is still overwhelmingly used for presentations today, but its recent product launches show that it wants a much broader role in visual communication.

Gamma's own Creation Index says decks represented 89.7% of creations in the first quarter of this year. Documents accounted for another 6.9%, and document volume had grown 36.1% year over year. So presentations clearly remain the center of the business.

The product around those presentations is expanding quickly, though. Gamma now has a standalone AI design canvas called Imagine for creating infographics, posters, logos, invitations, diagrams and social graphics. It has image editing, websites, social posts and an AI creation agent that can work from multiple sources, research a topic, develop an outline and generate a finished piece of content.

Gamma's changelog also shows the product becoming more design-heavy. Recent updates added six-column layouts, better font handling in PowerPoint exports, more consistent image styles and an agent that can see and modify an entire Imagine canvas through conversation.

Lica adds another dimension: motion. Its dataset contains roughly 27,000 animated designs where individual elements have their own keyframes, timing and transitions. About 6,600 of those are animated presentations.

Gamma's management is now openly talking about presentations becoming more interactive, fluid and multimodal. We should take that fairly literally. Gamma appears to see the slide deck as the starting format for a much larger AI communication product.

Q9Is Gamma trying to become the next Canva or Adobe?

Gamma is clearly moving into territory occupied by Canva and Adobe, although Gamma is targeting a somewhat different user.

Grant Lee has described the opportunity as the large gap between professional creative software such as Adobe and Figma and traditional workplace tools such as PowerPoint. Millions of people need good visual communication for work but do not want to become professional designers.

That describes Gamma's customer remarkably well.

Gamma Imagine makes the overlap much clearer than it was a year ago. Someone can now use Gamma to create a presentation, a website, a social graphic, an infographic, a poster or other standalone visual content. These are jobs people already do in Canva and Adobe products.

But Gamma's approach is much more automation-heavy. Instead of giving users increasingly sophisticated design tools and asking them to learn those tools, Gamma tries to make more of the design decisions itself.

Lica fits that strategy because better automation requires more understanding underneath it. If Gamma wants a non-designer to type "make this look like our brand, adapt it for LinkedIn and animate the important parts," Gamma needs much more than an image generator.

Gamma seems to be moving toward visual creation software where the AI does most of the design work and the human keeps enough control to fix what matters.

Q10Why is Gamma buying Lica now instead of building all this itself?

Gamma can afford to make a longer-term research bet now because the company has already reached a level of scale where another incremental presentation feature is unlikely to be enough.

Gamma crossed $100 million in annual recurring revenue while profitable and raised $68 million at a $2.1 billion valuation. When the company announced those numbers, it said it had reached that scale with a team of about 50 people and had been profitable for two years.

Its usage has kept climbing since then. Gamma's newer data covers more than 625 million creations across 210 countries and territories, while the company now says more than 100 million people have used the product.

That gives Gamma something Lica never had: a huge environment in which research can quickly become a real product and get tested through actual user behavior.

Building the same research group internally would mean recruiting people with expertise across layout generation, typography, vector graphics, animation, computer vision and design tooling, then giving that new team enough time to develop datasets and benchmarks before it could start attacking the harder product problems.

Lica arrives with much of that work already underway.

For a profitable company that has remained unusually lean, buying a concentrated team that has spent years on exactly the technical problems it now wants to solve is a pretty sensible shortcut.

Q11Does Lica give Gamma a real AI design moat?

Lica gives Gamma some unusual design infrastructure, but we would not call it a moat yet.

The 1.55 million-layout LICA dataset is genuinely large relative to earlier structured design datasets. For comparison, the Lica researchers put Crello, one of the richer previous datasets, at roughly 24,000 examples and CGL-Dataset at around 60,000. LICA also preserves far more information about the individual elements inside each design and includes animation, which earlier datasets generally did not.

Scale alone will not protect Gamma. Research gets published, competitors improve, foundation models absorb new capabilities and well-funded companies such as Canva, Adobe, Microsoft, Google and OpenAI can spend far more on AI than Gamma can.

The more interesting advantage could come from combining Lica's structured approach with Gamma's own usage. Gamma now sees hundreds of millions of real creations and has data showing how its most active users import material, revise designs, apply themes and export their work.

That creates a potentially powerful feedback loop. Lica can help Gamma understand what a design is made of, while Gamma can observe how people actually change those designs once AI generates them.

The acquisition gives Gamma better raw material for a moat rather than handing it one overnight.

Q12Did OpenAI buying NextSlide push Gamma to buy Lica?

OpenAI's NextSlide acquisition raises the pressure on Gamma, but the evidence does not show that Gamma bought Lica as a rushed response to OpenAI.

NextSlide is a much more direct presentation story. The startup built AI that could turn prompts, notes, documents and research into editable presentations, and OpenAI acquired the company this year. The NextSlide team is now working inside OpenAI.

The timing is impossible to ignore. AI presentation startups spent the first few years competing mainly with one another. Now the capability is being absorbed into much larger AI platforms.

That makes simple prompt-to-deck generation a dangerous place for Gamma to stop. If ChatGPT, Gemini, Microsoft Copilot and other general-purpose AI products can all generate reasonable presentations, Gamma needs users to feel a much larger difference when they use Gamma.

But Lica and Gamma had been talking before the deal, according to TechCrunch, and their founders already knew each other through shared investors including Accel and South Park Commons. Lica had also spent months publishing research that lines up closely with Gamma's current direction.

OpenAI looks more like evidence of the competitive pressure Gamma faces than the specific reason this acquisition happened.

Q13What does Lica get from joining Gamma?

Lica gets immediate access to a user base that would have been extraordinarily difficult for the startup to build on its own.

Priyaa Kalyanaraman described the fit quite clearly when discussing the acquisition with TechCrunch. Lica had concentrated heavily on frontier research, while Gamma had become very good at distribution and had taken its product to more than 100 million users.

That changes what Lica's research can become.

A small research-heavy startup can publish a strong benchmark, build an interesting model and test workflows with individual brands. Gamma can potentially expose the same ideas to people creating more than a million pieces of content on busy days and learn very quickly where the technology works, where users fight it and which kinds of control people actually want.

Lica also gets a much wider canvas. Its research already covered presentations, social graphics, branded creative, vectors and animation. Those formats increasingly exist inside the same Gamma product ecosystem.

For Lica's founders, joining Gamma is therefore less of a retreat from their original idea than a much faster route to getting the idea into a mainstream product.

Q14What could make Gamma's Lica acquisition disappoint?

Gamma's Lica acquisition will disappoint if all this research produces clever papers but users barely notice a difference in the product.

That risk is real. Gamma has not announced a specific Lica-powered product, a release schedule or measurable performance targets for the new research lab. We know what the team wants to investigate, but we do not yet know which ideas will survive contact with millions of users.

Lica's own benchmark also shows how difficult the work is. Professional design is really a collection of different problems involving typography, spatial reasoning, vectors, animation, composition and brand identity. Progress in one area does not automatically fix the others.

And Gamma has a product-design constraint that Lica did not face at the same scale: Gamma became popular because it removed work. If better design control turns into dozens of new controls, settings and professional-design concepts, Gamma could reproduce the complexity it originally helped people escape.

The useful test is much simpler. Over the next generation of Gamma products, users should need fewer retries, make cleaner local edits, preserve brand rules more reliably and adapt one piece of work into several formats without rebuilding it.

If Lica's research produces those improvements, the acquisition will look smart. If users mainly get another set of AI-generation buttons, the strategic story will have been much more impressive than the result.

Q15Why is Gamma buying Lica?

Gamma is buying Lica because the next big problem for Gamma is making AI understand and edit design reliably, and Lica has spent years working on exactly that problem.

The pieces fit unusually well.

Gamma already has the distribution, with more than 100 million users and hundreds of millions of creations. It has also moved well beyond basic presentation generation through Agent, Imagine, image editing, websites, documents, social content and more advanced design workflows.

Lica brings a very different asset. Its team has built a 1.55 million-layout structured design dataset, studied where leading AI models fail across 49 real design tasks and developed technology around layers, typography, vectors, animation, personalization and brand identity.

As we saw above, Gamma is putting Lica's founders in charge of a new design research lab rather than treating Lica as another standalone app. That decision tells us more about the acquisition than the undisclosed purchase price.

The bet is that future AI creation tools will need to do much more than generate a beautiful first draft. They will need to understand why a design looks the way it does, change one part without breaking the rest, remember a company's visual identity, adapt the same idea for different people and formats, and eventually make presentations far more interactive than today's slides.

Gamma already has the audience for that product. Lica gives it a team that has been trying to build the missing design intelligence underneath it.

That is the clearest reason Gamma is buying Lica today.