Firefly Image 5 Preview is Adobe's generative AI model designed for high-fidelity image synthesis and professional creative workflows. Introduced at Adobe MAX 2025, it features native 4-megapixel generation, allowing the model to produce images at resolutions of approximately 2240x1792 pixels directly without initial upscaling. This upgrade significantly improves detail, sharpness, and color accuracy for commercial-scale applications.
The model's primary innovations include Layered Image Editing, which enables the deconstruction of images into independent, editable layers. This allows creators to manipulate specific elements—such as moving subjects or adjusting backgrounds—while the AI automatically re-projects shadows and re-harmonizes the composition. Furthermore, the model supports Custom Models, which allow artists to train the generator on their own artistic portfolios to maintain stylistic consistency across projects.
Technical refinements in Firefly Image 5 focus on photorealistic human rendering, specifically addressing common AI artifacts such as hand deformities and anatomical inaccuracies. The model also introduces a conversational Prompt to Edit interface, allowing users to describe modifications in everyday language. Like previous iterations, it is trained on Adobe Stock and public domain content to ensure the output is commercially safe and includes Content Credentials for provenance tracking.