Adobe Firefly Image Model 4

Firefly Image Model 4 is Adobe's fourth-generation text-to-image model, launched April 24, 2025 at Adobe MAX London. Trained exclusively on licensed content — Adobe Stock, public domain, and openly licensed work — it's the 'commercially safe' answer to Midjourney and DALL-E for enterprises that can't risk copyright lawsuits. Model 4 targets fast iteration with images up to 2K resolution, improved prompt fidelity, camera-angle and zoom controls, and better text rendering inside images. A companion Model 4 Ultra handles complex photorealistic scenes with fine detail. Both ship inside Photoshop, Express, and the Firefly web app.

The honest picture: 'commercially safe' comes with a creative ceiling — Firefly's training-data conservatism means it trails Midjourney and FLUX on artistic range and stylistic surprise. It's subscription-gated (Creative Cloud or Firefly plans from ~$9.99/mo) with a credit system, not open weights, so there's no self-hosting or fine-tuning. And since Adobe now also offers partner models (OpenAI, Google, Ideogram, Luma) inside Firefly, the platform is quietly admitting its own models aren't always the best choice.