Adobe Research

Adobe Research is the research division of Adobe Inc., founded in 2006 with headquarters in San Jose, California, and additional labs in Seattle, San Francisco, Cambridge (Massachusetts), London, Tel Aviv, Bangalore, and Bucharest. It develops the artificial intelligence and machine-learning technology that ships across Adobe's product lines: Photoshop, Premiere Pro, Lightroom, Acrobat, Illustrator, After Effects, Adobe Express, and the Adobe Firefly generative model family. As of April 2026, Adobe Research is one of the largest industrial AI research organizations focused on creative-tooling applications, with roughly 200 researchers across its labs and a publication record at major venues including SIGGRAPH, NeurIPS, ICCV, and CVPR.

At a glance

• Founded: 2006 (formalized as a corporate research division; earlier research history dates to Adobe's founding in 1982).
• Status: Research division of Adobe Inc., a publicly traded company (NASDAQ: ADBE).
• Funding: Operates within Adobe's annual research-and-development budget. Adobe spent approximately $3.6 billion on R&D in fiscal 2024, with Adobe Research representing a portion of that spend on advanced AI work.
• Senior leadership: Eric Larsen, Vice President and Head of Adobe Research. Gavin Miller, Senior Fellow and former head of Adobe Research, who continues to direct senior research strategy.
• Other notable leadership: Hanna Wallach (Senior Principal Researcher), Aaron Hertzmann (Principal Scientist), and other Adobe Fellows including Kirk Olynyk and Anders Forsberg.
• Open weights: Limited. Adobe Research publishes research output, but most production model weights remain closed. Selected research model code is released on GitHub.
• Flagship outputs: Adobe Firefly (image, video, vector, design, and audio generation), Generative Fill in Photoshop, Generative Expand, Generative Remove, Project Sky Replace, Project Music GenAI Control, Acrobat AI Assistant, Adobe Sensei (the unified AI/ML platform across Adobe products), and a research-paper output of more than 100 publications per year at major venues.

Origins

Adobe's research history reaches back to the company's founding in 1982 by John Warnock and Charles Geschke, with the early development of PostScript at the original Adobe Systems Inc. Through the 1990s and 2000s, advanced research at Adobe was distributed across product groups and individual researchers including David Salesin and Gavin Miller, who anchored the company's deep technology investments in image processing, color science, and computational graphics.

In 2006, Adobe formalized this distributed research effort into a centralized Adobe Research division, with Gavin Miller as one of its founding leaders. The division's structure, with research labs co-located near key product engineering centers, was designed to keep research output tightly coupled to product integration. Adobe Research has remained that way through the 2010s and 2020s.

The 2017 launch of Adobe Sensei marked the company's first public branding of its AI and machine-learning platform, with Sensei-powered features rolling into Photoshop, Premiere Pro, Lightroom, and other Creative Cloud products through 2018, 2019, and 2020. The 2022 to 2024 period saw the most consequential acceleration. Adobe Research delivered the underlying generative-image technology that became Firefly, launched as Firefly 1 in March 2023, with product integration into Photoshop, Illustrator, and Adobe Express. The Firefly Image Model 3 (October 2024), Firefly Video Model (early 2025), Firefly Vector Model, and Firefly Design Model continued the cadence through 2025 and into 2026.

The 2024 to 2026 period has seen Adobe Research expand into multimodal generative AI substantially, with Project Music GenAI Control, audio-to-video research, document-understanding work for Acrobat AI Assistant, and continued image-and-video work through the Firefly line.

Mission and strategy

Adobe Research's stated mission is to invent the future of digital experiences for creative professionals and knowledge workers. The strategic premise is that Adobe's product distribution to creative professionals (Photoshop, Premiere Pro, Lightroom, Illustrator, the broader Creative Cloud), enterprise document workflows (Acrobat, Document Cloud), and marketing operations (Experience Cloud) provides a durable distribution moat that justifies long-horizon research investment specifically tuned to Adobe's customer base.

The strategy combines four threads. First, generative AI for creative professionals through the Firefly model family and its integration into Adobe's product surface. Second, content-authenticity research through the Content Authenticity Initiative and the C2PA cryptographic provenance standard, directly tied to Adobe's role as an industry safety-and-trust leader. Third, document AI for the Acrobat product line and the broader knowledge-worker audience, including Acrobat AI Assistant. Fourth, fundamental research in graphics, vision, language, and audio that publishes openly and feeds into product engineering.

The competitive premise is that AI features built specifically for Photoshop, Premiere Pro, Illustrator, Lightroom, and other creative-professional tools are higher-value than horizontal foundation-model output, because they integrate cleanly into existing professional workflows. Adobe's commercial-safety positioning of Firefly (trained on licensed and public-domain content, with explicit indemnification for enterprise customers) is a direct strategic response to the legal exposure of generative AI in commercial creative work.

Distribution is Adobe's structural advantage. Firefly is integrated into Photoshop, Illustrator, Lightroom, Premiere Pro, Adobe Express, and the broader Creative Cloud, reaching tens of millions of paying creative professionals globally. Acrobat AI Assistant reaches the substantially larger Acrobat user base, including hundreds of millions of free and paid Acrobat users.

Models and products

• Firefly model family. Firefly Image Model 3, Firefly Video Model, Firefly Vector Model, Firefly Design Model, Firefly Audio Model. Closed weights. Trained on licensed and public-domain content for commercial-safe use. Powers generative features across Adobe's product line.
• Generative features in Photoshop, Illustrator, Lightroom, Premiere Pro, Adobe Express. Generative Fill, Generative Expand, Generative Remove, Project Sky Replace, Vector-to-Vector Generation, Text-to-Image, Image-to-Image, Image-to-Video, Generative Soundtrack.
• Acrobat AI Assistant. Document-understanding AI integrated into Acrobat. Powered by underlying foundation models with retrieval over the document context.
• Adobe Sensei. The unified AI/ML platform across Adobe products; the predecessor brand to the Firefly era for many Adobe AI features.
• Project Music GenAI Control. Music-generation research with conditional-generation control surfaces; surfaced as a research preview rather than a productized model.
• Research outputs. More than 100 papers per year at SIGGRAPH, NeurIPS, ICCV, CVPR, EMNLP, and other venues, with selected code releases on the Adobe Research GitHub.

Distribution channels include direct integration into Adobe's Creative Cloud and Document Cloud products, the standalone Firefly web app, Firefly APIs for enterprise customers, and continued open research output through publications and open-source code releases.

Benchmarks and standing

Adobe Research is positioned differently from frontier AI labs that publish benchmark results on standardized leaderboards (LMArena, GPQA Diamond, SWE-bench Verified). Adobe Research's evaluation framework focuses on creative-quality benchmarks, professional-user studies, and integration metrics tied to Adobe products rather than horizontal foundation-model leaderboards.

Firefly's image-generation quality has been consistently ranked among leading commercial image generators on Arena.ai's image-generation leaderboard, with Firefly Image Model 3 placing in the top tier alongside Ideogram, Black Forest Labs FLUX, and Stability AI outputs. Firefly's commercial-safety positioning, with explicit indemnification for enterprise customers using outputs in commercial contexts, is a structural differentiator from competitor models trained on broader scraped corpora.

Adobe Research's publication output across SIGGRAPH, NeurIPS, ICCV, and CVPR is consistently among the largest industrial-research outputs at those venues, comparable in scale to research output from Google DeepMind, Microsoft Research, and Meta AI / FAIR on creative-tooling-adjacent topics.

Leadership

As of April 2026, Adobe Research's senior leadership includes:

• Eric Larsen, Vice President and Head of Adobe Research. Adobe leader with research-management background.
• Gavin Miller, Senior Fellow and former head of Adobe Research. One of Adobe Research's founding leaders in 2006; continues to direct senior research strategy. Computer-graphics researcher with a publication record at SIGGRAPH and other venues.
• Hanna Wallach, Senior Principal Researcher. Machine-learning researcher with a publication record. Joined Adobe Research after a long tenure at Microsoft Research.
• Aaron Hertzmann, Principal Scientist. Computer-graphics and machine-learning researcher; ACM SIGGRAPH Significant New Researcher Award recipient.
• Senior research-and-engineering leadership across Adobe Research's principal lab locations and program areas.

Departures and arrivals are continuous. The 2024 to 2026 period has seen recruitment of senior generative-AI researchers and continued senior leadership stability across the principal program areas.

Funding and backers

Adobe Research operates within Adobe Inc.'s overall research-and-development budget. Adobe reported approximately $3.6 billion in research-and-development spending in fiscal 2024 (Adobe's fiscal year ends in early December), with Adobe Research representing a portion of that spend on advanced AI work. Adobe is publicly traded on NASDAQ as ADBE, with a market capitalization above $200 billion as of April 2026, and reports research-and-development spending in its 10-K filings without separately disclosing Adobe Research's specific allocation.

Open questions on Adobe Research's near-term resourcing are minimal compared to standalone AI labs, given Adobe's stable revenue base and consistent profitability. The principal allocation question is the balance between in-house Firefly model development and continued partnerships with external foundation-model providers, with Adobe's March 2024 partnership with OpenAI and other strategic relationships providing optionality.

Industry position

Adobe Research occupies a structurally distinctive position as the largest industrial AI research organization focused specifically on creative tooling, with product distribution to creative professionals, enterprise document workflows, and marketing operations. The Firefly model family and its commercial-safety positioning have been characterized in industry coverage as a structural differentiator in the enterprise generative-AI market, with the indemnification framework explicitly designed for enterprise customers concerned about legal exposure from generative AI outputs in commercial contexts.

Adobe Research's research-publication output is consistently among the largest at SIGGRAPH and other venues. Industry coverage has noted Adobe Research's recruitment of senior generative-AI researchers through 2024 to 2026, alongside continued senior leadership stability under Eric Larsen and Gavin Miller.

Competitive landscape

• Adobe Firefly. Adobe's productized generative-AI line, developed by Adobe Research and integrated across Adobe's product surface.
• Microsoft AI and Microsoft Research. Industrial-research peer with overlap on generative-AI for productivity, document understanding, and enterprise distribution.
• Google DeepMind and Google Research. Industrial-research peer with overlap on generative-image, generative-video, and other creative-AI research; Google's Imagen and Veo families are direct creative-tooling competitors.
• Meta AI / FAIR. Industrial-research peer with open-research output across image, video, and audio generation. The Movie Gen line is a direct creative-tooling competitor.
• Apple Intelligence and Apple Research. Industrial-research peer with creative-tooling overlap through Final Cut Pro, Logic Pro, and the broader Apple media-creation ecosystem.
• NVIDIA Research. Industrial-research peer with overlap on generative-AI infrastructure, image-and-video generation, and the underlying compute platform.
• OpenAI. Partnership relationship and frontier-AI provider; Adobe's March 2024 partnership with OpenAI provides API access to OpenAI's image and video models for selected use cases.
• Stability AI, Black Forest Labs, Ideogram, Midjourney. Direct creative-AI competitors on image generation, with substantially different commercial-safety positioning.

Outlook

• The continued cadence of Firefly Image, Video, Vector, Design, and Audio model releases through 2026 and 2027.
• Adobe Research's publication output and senior research-talent retention through the 2026 to 2027 generative-AI commercial expansion.
• The continued integration of generative AI features into Photoshop, Premiere Pro, Illustrator, Lightroom, and other Creative Cloud products.
• Adobe's commercial-safety positioning of Firefly as enterprise-AI markets continue to consolidate around indemnification and provenance-standard requirements.
• The competitive dynamic between Firefly and Google's Imagen / Veo, OpenAI's Sora, Stability's image and video lines, and other commercial generative-AI offerings.
• Continued senior research-talent recruitment from frontier AI labs into Adobe's industrial-research positioning.

Sources

• Adobe Research official site. Research-division reference.
• Adobe Firefly product page. Firefly model family and product reference.
• Adobe Research GitHub. Selected open-source code releases.
• Content Authenticity Initiative. Adobe-led industry initiative on content provenance and authenticity.
• Adobe Inc. 10-K filings. Adobe's research-and-development spending and financial reporting.