Adobe Firefly: what Firefly Image 5 can do and how to use it

What is Adobe Firefly

Adobe Firefly is generative AI from Adobe, the company behind Photoshop, Illustrator and Premiere. The name covers two things: Adobe’s own family of Firefly models and the Firefly app, where you can run them alongside models from other developers.

Firefly Image 5 is the current in-house image model. Adobe announced it as a public beta at its MAX conference on October 28, 2025, and it became generally available in March 2026. Its main upgrades over earlier versions are native 4MP output without upscaling and editing photos with a plain sentence (Prompt to Edit).

The previous generation — Firefly Image 4 and Image 4 Ultra — came out in April 2025. Before that there were Image 3 (2024) and Image 2 (2023).

Adobe’s main pitch is commercial safety. Firefly is trained on licensed content such as Adobe Stock and on public domain works; Adobe says it doesn’t scrape the web for training data and has never trained Firefly on user content.

What Adobe Firefly can do

• Photorealistic 4MP images. Image 5 generates at high resolution straight away, without upscaling. Adobe highlights anatomically accurate portraits, lighting, textures and complex multi-layered scenes.
• Editing from a sentence (Prompt to Edit). Upload a photo and describe the change: “replace the grey sofa with a green velvet one and add a painting above it.”
• Generative Fill and Expand in Photoshop and the Firefly app: extend a background, remove or add an object, widen a square shot into a landscape one.
• Vector graphics — illustrations and icons you can edit in Illustrator.
• Custom Models. A model trained on your images that generates in your style or with your character. In public beta since March 2026, private by default.
• Video and audio. The Firefly app also does text- and image-to-video, music for your footage (Generate Soundtrack), voiceovers (Generate Speech) and a video editor.
• Partner models. The Firefly app gives access to more than 30 models: Nano Banana 2, Veo 3.1, GPT Image, FLUX, Kling, Runway and more.

Here’s what editing from a sentence looks like — we recreated a Prompt to Edit task with Nano Banana Pro:

And this is Generative Expand — a square shot turned into a wide one, with a door and a window box added on the sides:

What Adobe Firefly is best for

• Ads and commercial projects where legal safety matters: the model is trained on licensed content.
• Designers in the Adobe ecosystem: Generative Fill and edits right in Photoshop, vectors in Illustrator, video in Premiere.
• Product photography: clean light, textures, 4MP for print. Similar tasks are covered in our guide to AI product images for ecommerce.
• Editing and extending existing photos — swapping backgrounds and objects, changing the aspect ratio.
• On-brand series with Custom Models.

Adobe Firefly vs alternatives

	Adobe Firefly Image 5	Nano Banana Pro	GPT Image	Midjourney
Developer	Adobe	Google	OpenAI	Midjourney
Strength	commercial safety, Photoshop	complex scenes, faces and products from photos	text and layout in images	artistic look
Resolution	native 4MP	up to 4K	up to 3840 px (GPT Image 2.5)	up to 2048 px
Editing from a sentence	yes (Prompt to Edit)	yes	yes	yes (Edit Model)
Trained on licensed data	yes, per Adobe	not disclosed	not disclosed	not disclosed
Vector	yes	no	no	no

If legal safety and Photoshop are what matter most, pick Firefly. If you need to keep a face or product from a photo across new scenes, pick Nano Banana Pro: it holds up to 14 references. For images with long text, GPT Image is stronger; for art, Midjourney.

How to write prompts for Adobe Firefly

Firefly understands plain descriptive sentences. The formula that works: subject + action + setting + light + photography style:

Commercial product photo of an unbranded frosted glass perfume bottle
with a brass cap on a slab of travertine stone, pampas grass and roses
around it, soft sunlight with crisp shadows, warm neutral palette.

• Describe the light and lens: “soft window light,” “85mm lens” — Image 5 is strong on light and texture.
• For edits, say what to keep: “don’t change the window, the table or the floor.”
• For vectors, add “flat vector illustration, no gradients.”
• Other brands and real people are off-limits in Firefly — part of Adobe’s safety approach.

For more phrasing ideas, see our AI image prompts.

How to use Adobe Firefly in AI2Face

1. Open the AI2Face generator and pick Adobe Firefly in the model list.
2. Write a prompt, or attach a photo with “+” if you want an edit.
3. Choose the aspect ratio and generate. The cost in credits is shown before you start.

The same generator has Nano Banana Pro, GPT Image and Midjourney, so it’s easy to compare models on one prompt.

Limitations

• Strict filters. Firefly refuses recognizable people, other brands and characters — good for commercial work, bad for fan art.
• Content Credentials. Adobe attaches provenance metadata to files; platforms such as LinkedIn and Instagram can show an AI label based on it.
• Complex pricing on Adobe’s side: a free daily limit, generative credits and separate terms for partner models and video.

Bottom line

Adobe Firefly is the best pick when legal safety and working inside Photoshop matter: it’s trained on licensed content, and Image 5 generates at 4MP and edits photos from a sentence. For portraits from your own photo, complex multi-reference scenes and art, other models are stronger — Nano Banana Pro and Midjourney. Try Firefly and its alternatives on the same prompt in AI2Face and choose by results, not by brand.