Adobe's Firefly Bet Isn't Working
Adobe staked its generative AI future on ethically trained models and stock-library licensing deals with Getty Images and Shutterstock. Eighteen months in, enterprise adoption is lukewarm, professional creatives still prefer Midjourney and Stable Diffusion, and the stock-photo partners are getting restless over revenue splits. What happens when you optimize for legal safety over product quality — and your competitors don't.
By Zoe Nakamura, Mobile Growth · Mar 12, 2026
Adobe Firefly's ethically trained AI image generator is losing to Midjourney and open-source alternatives on quality. Enterprise adoption is below targets, stock-photo licensing deals are underperforming, and creatives are voting with their wallets. A data-driven look at why Adobe's legal-safety-first AI strategy is backfiring.
Frequently Asked Questions
Is Adobe Firefly good for professional creative work?
Adobe Firefly has improved significantly since its March 2023 launch, but independent benchmarks and user surveys consistently rank it behind Midjourney, DALL-E 3, and Stable Diffusion XL for photorealism, prompt adherence, and artistic flexibility. In a January 2026 Blind survey of 2,400 professional designers, only 18% rated Firefly as their primary AI image generation tool, compared to 41% for Midjourney and 22% for Stable Diffusion variants. Firefly's main advantage is legal indemnification — Adobe offers IP indemnity for commercial use of Firefly-generated images, which matters for enterprise marketing teams but is less important to freelance creatives and agencies who prioritize output quality.
How does Adobe Firefly compare to Midjourney?
Midjourney consistently outperforms Adobe Firefly on image quality, artistic style range, and photorealism in independent benchmarks. In the February 2026 Artificial Analysis Image Arena rankings, Midjourney v6.1 scored an ELO of 1145 versus Firefly Image 3's 1038 — a significant gap. Midjourney also leads in prompt adherence and compositional complexity. However, Adobe Firefly has advantages in enterprise integration (it is embedded natively in Photoshop, Illustrator, and Express), legal safety (trained exclusively on licensed Adobe Stock, public domain, and openly licensed content), and IP indemnification for commercial outputs. For professional creatives who prioritize raw output quality, Midjourney remains the preferred tool. For enterprise marketing teams that need legal cover and workflow integration, Firefly is the safer choice — though 'safer' increasingly means 'slower to adopt.'
Is Adobe losing to AI competitors?
Adobe is not losing its core creative software business — Photoshop, Illustrator, Premiere Pro, and InDesign remain industry standards with strong retention. However, Adobe is losing the generative AI image creation market to Midjourney, OpenAI's DALL-E, and open-source models like Stable Diffusion and Flux. Adobe's Digital Media segment grew approximately 11% in fiscal 2025, but Firefly-specific revenue contribution remains undisclosed and is estimated at $400-600 million annually — well below the $1 billion run-rate target Adobe set for fiscal 2025. The risk is not that Adobe loses Photoshop customers today, but that a generation of creators builds workflows around non-Adobe AI tools, eroding the company's long-term relevance as generative AI becomes the primary mode of visual content creation.
What is Adobe's AI strategy?
Adobe's AI strategy centers on three pillars: Firefly (its family of generative AI models trained on licensed content), Sensei (its legacy machine learning platform for analytics and automation), and deep integration of AI features into existing Creative Cloud applications. CEO Shantanu Narayen and Chief Product Officer David Wadhwani have positioned Firefly as the 'commercially safe' alternative to competitors trained on scraped web data. Adobe has signed licensing deals with Getty Images, Shutterstock, and thousands of individual contributors to source training data. The company charges for Firefly usage through generative credits bundled with Creative Cloud subscriptions and standalone Firefly plans starting at $9.99/month. Critics argue this strategy prioritizes legal defensibility over model quality, resulting in outputs that lag competitors by 6-12 months.
Does Adobe Firefly use copyrighted images for training?
Adobe has stated that Firefly models are trained exclusively on Adobe Stock images (for which Adobe holds licenses), openly licensed content, and public domain works. This is a deliberate contrast to competitors like Midjourney, Stable Diffusion, and DALL-E, which were trained on large-scale internet scrapes that included copyrighted material. Adobe offers IP indemnification for Firefly outputs, meaning Adobe will cover legal costs if a customer is sued over a Firefly-generated image. However, this constrained training dataset is also Firefly's primary limitation — with approximately 400 million licensed images versus the billions of images in competitors' training sets, Firefly has less diversity, fewer stylistic references, and weaker performance on niche or culturally specific prompts.
How much revenue does Adobe Firefly generate?
Adobe does not break out Firefly revenue separately in its financial reports. Based on disclosed generative credit consumption, Creative Cloud attach rates, and standalone Firefly subscription data, analysts at Morgan Stanley and Bank of America estimate Firefly generated between $400-600 million in annualized revenue by Q4 FY2025 — a combination of incremental subscription upgrades, standalone Firefly plans, and API licensing to enterprise customers. This is significantly below the $1 billion annual run-rate that Adobe guided toward in its 2023 analyst day. Adobe CFO Dan Durn has said Firefly is 'accretive to Creative Cloud ARPU' but has declined to quantify the precise contribution, which analysts interpret as an acknowledgment that the numbers are below expectations.
What are the alternatives to Adobe Firefly for AI image generation?
The main alternatives to Adobe Firefly include Midjourney (best overall image quality, subscription-based at $10-60/month), OpenAI's DALL-E 3 and GPT-4o image generation (integrated into ChatGPT, strong at text rendering and instruction-following), Stable Diffusion and Flux (open-source models that run locally or via cloud services, maximum customization), Google's Imagen 3 (available through Gemini, strong photorealism), and Ideogram (excels at typography and text-in-image generation). For professionals embedded in Adobe's ecosystem, Firefly's integration with Photoshop's Generative Fill and Generative Expand remains a strong workflow advantage despite the model's quality gap. Canva's Magic Design suite is also a strong option for non-designers who need fast, template-driven AI generation.
Will Adobe Firefly get better?
Adobe has released three major Firefly model versions since March 2023, with each version showing measurable improvements in photorealism, prompt adherence, and resolution. Firefly Image 3, released in late 2025, narrowed the gap with Midjourney v6 meaningfully but did not close it. Adobe has indicated that Firefly Image 4, expected in mid-2026, will incorporate new training techniques and an expanded dataset through recently signed licensing agreements with additional stock libraries and individual photographers. However, the structural constraint remains: Adobe's commitment to licensed-only training data limits its dataset size to roughly 400-500 million images, versus the multi-billion-image datasets used by competitors. Whether architectural improvements can compensate for this data gap is the central technical question for Firefly's future.
Related Articles
Topics: Adobe, AI, Creative Tools, Generative AI
Browse all articles | About Signal