There is an instructive competition playing out in the creative software market right now. On one side: Adobe Firefly, a generative AI image tool backed by one of the most established software companies on the planet, trained on hundreds of millions of licensed Adobe Stock images, and priced from $9.99 a month for individuals to enterprise agreements negotiated on a case-by-case basis. On the other: an open-source plugin for Krita, a free digital painting application, maintained by an independent developer on GitHub, with over 10,000 stars and a community-run documentation site.

The Krita AI Diffusion plugin currently supports Stable Diffusion 1.5, SDXL, Flux 2, and the newer Z-Image model. It runs locally on consumer hardware – NVIDIA, AMD, and Apple Silicon GPUs are all supported – or connects to cloud backends for those without the compute. It supports inpainting, outpainting, generative fill, upscaling to 8K, real-time live painting, region-based prompting, and ControlNet-based pose and depth guidance. It costs nothing.

Adobe Firefly, in its Creative Cloud Pro incarnation, now costs $69.99 a month for the renamed all-apps plan as of June 2025, up from $59.99. Adobe positioned that increase as a reflection of Firefly’s expanding capabilities and the growing infrastructure cost of cloud-side AI generation.

Enterprise buyers are facing a more difficult reality: defining what better even means at a time when image quality differences are narrowing faster than their own procurement timelines.

What the Krita Plugin Actually Does

The Krita AI Diffusion plugin, developed under the project name Interstice, uses ComfyUI as its backend inference engine – a node-based interface for running diffusion models that has become a foundational layer for the open-source image generation ecosystem.

Within Krita’s canvas, the plugin exposes a clean interface that abstracts away most of the technical complexity. Artists can use Krita’s native selection tools to mark an area and trigger generative fill, exactly as they would in Photoshop’s equivalent feature. They can extend a canvas edge with outpainting, run live generation as they paint to see real-time AI interpretations of brushstrokes, apply ControlNet layers that anchor generation to a pose or depth map, and run upscaling passes at resolutions that consumer GPUs would normally struggle with.

Version 1.49.1, released April 11, 2026, supports Flux 2 Klein and Z-Image – two models that were not publicly available a year ago. The model roster updates when the underlying open-source ecosystem releases new checkpoints, not on a vendor’s product roadmap.

That model roster matters. Flux 2, released by Black Forest Labs in November 2025, represents a meaningful generational leap in image generation quality. The flagship Flux 2 Dev model runs on 32 billion parameters, supports up to 10 reference images simultaneously for style and character consistency, handles typography reliably enough for production use, and generates images at resolutions up to 4 megapixels. The Krita plugin absorbed these models within weeks of their release. Enterprise software with annual review cycles and quarterly roadmaps cannot match that cadence.

Adobe Firefly’s Actual Value Proposition

It would be inaccurate to dismiss Firefly as a product with no durable advantages. Adobe has made a specific and coherent bet: that the feature distinguishing it from free alternatives is not generation quality but legal safety.

Firefly is trained exclusively on licensed Adobe Stock content, public domain material, and openly licensed content. Adobe has offered IP indemnification to enterprise customers since Firefly’s initial launch – meaning that if a client faces a copyright infringement claim arising from Firefly-generated content, Adobe will defend that claim. For qualifying Creative Cloud plans, that indemnification is contractual.

Adobe has also built Firefly into its existing creative workflow applications. Generative Fill in Photoshop, Generative Extend in Premiere Pro, Text to Vector and Generative Recolor in Illustrator are all available without additional setup. Firefly Services, Adobe’s API layer for enterprise automation, offers over 30 generative AI and creative APIs for bulk asset production, background removal, resizing, and image variation at scale. That infrastructure is not replicated by a Krita plugin.

The Speed at Which the Gap Is Closing

None of that changes the underlying trajectory. Adobe’s pricing has moved in one direction: the Creative Cloud All Apps plan became Creative Cloud Pro in June 2025, rising to $69.99 from $59.99. Enterprise Creative Cloud Edition 4, available since June 4, 2024, saw increases of approximately 7 to 8 percent depending on region. Adobe framed these changes as the cost of access to a growing suite of generative AI capabilities.

That framing contains a telling detail. Adobe is now routing users to external models – including Flux, OpenAI, and Google – through its own interface. Models that are accessible independently at lower or no cost. The Flux models now available inside Firefly are the same models running inside the Krita plugin, for free, on a user’s local machine.

The open-source stack has moved from experimental to capable to approaching production-ready within approximately three years. By January 2026, Flux 2 Klein was running in under a second on consumer GPUs and was integrated into the Krita plugin on an ongoing basis. Enterprise software with annual review cycles cannot match that cadence – structurally, by design.

A Structural Problem That Applies Beyond Creative Tools

The Krita-Firefly dynamic reflects a structural problem affecting any product category where underlying model capabilities advance on open-source timelines. The traditional enterprise software value chain assumed that the vendor controlled access to the capability, and that competitive moats were defensible for years. Open-source AI disrupts that model in two ways simultaneously.

First, it makes the underlying capability freely available to anyone with sufficient compute. Second, it accelerates the capability improvement timeline because the global developer community contributes to model development, fine-tuning, and tooling in parallel – rather than a single vendor’s R&D team working in series.

Adobe recognized this early enough to pivot Firefly’s positioning away from “best model” and toward “legally safe model with frictionless integration.” That is a more defensible position – but not an indefinitely defensible one. As open-source models improve their safety processes – Black Forest Labs conducted independent third-party safety evaluations of Flux 2 Dev and Klein before release, and both include C2PA content provenance metadata – the safety gap will narrow too.

The Krita plugin will not replace Adobe Firefly in a large financial services firm’s content production workflow tomorrow. But the fact that a community-maintained, zero-cost tool now runs models that rival commercial offerings – and updates its model support on a timeline that commercial vendors cannot match – marks a structural shift in what enterprise buyers can reasonably expect from a software subscription. That shift is not going away.