What Is Generative Fill: How AI Paints In and Extends Images
You brush over an unwanted object — and it vanishes, as if it were never there. You drag the crop frame outward — and the empty space fills with a believable continuation of the scene. This is Generative Fill, one of the most striking AI features in image editors. Let's unpack how it works, how it differs from extending the frame, and where the limits of its powers lie.
What Generative Fill is
Generative Fill is a Photoshop tool built on the Adobe Firefly AI engine that, from a text prompt, adds, replaces or removes content inside a selected area of an image. Want to remove a passerby from a photo, swap an overcast sky for a sunset, or add a vase to an empty table? All of it comes down to a selection and a few words.
The real beauty is that the tool works non-destructively: the result appears on a separate generative layer while the original stays untouched. Don't like a variation? Delete the layer and the photo is exactly as it was.
In short
Generative Fill = "paint in what I ask for, right here." You select an area, type a prompt (or leave it empty so the AI follows the surroundings), and get three variations on a new layer.
How it works: selection and prompt
The process is dead simple and comes down to a few steps.
Select an area
With any selection tool, outline the region you want to change or fill.
Click Generative Fill
The button appears in the contextual taskbar right below your selection.
Type a prompt
Describe what should appear. Leave the field empty and the AI fills the area based on the surroundings (handy for removing an object).
Pick from three variations
Photoshop offers three versions on a separate layer. Not right? Generate more.
Under the hood is a so-called diffusion model: the AI starts from random speckles, like static, and gradually "develops" the missing piece out of them, guided by your prompt and the pixels around the selection so that light, shadow and perspective match the rest of the image.
Generative Expand: extending the frame
Generative Fill has a twin — Generative Expand. Where Fill works inside the picture, Expand does the opposite: it extends the image beyond its borders. Need to turn a vertical photo into a horizontal one? Not enough headroom above your subject? Grab the Crop tool, drag the handles outward past the original frame, hit Generate — and the AI paints in what could plausibly lie beyond the edge.
This is especially handy when you need to change the aspect ratio for a different platform: say, widening a square AI generation into a wide banner. By the way, on why generations are square by default, we wrote in what format AI images use.
Inpainting versus outpainting
Behind these two tools are two classic terms from the world of AI graphics. Knowing the difference is useful, because every editor uses it, not just Photoshop.
| Concept | What it does | Tool | Example |
|---|---|---|---|
| Inpainting | Paints INSIDE the frame | Generative Fill | Remove an object, replace a sky |
| Outpainting | Rebuilds BEYOND the frame | Generative Expand | Extend a background, change format |
Put simply: inpainting is "fill the hole inside," and outpainting is "show what was past the edge." Both rest on the same idea — the model guesses the missing pixels from context.
The Firefly engine and what's new in 2026
Both Generative Fill and Generative Expand run on the same AI engine — Adobe Firefly. It has an important practical advantage: Adobe trains Firefly on licensed and public-domain content and even promises to cover you legally if a copyright dispute ever comes up (what lawyers call IP indemnification) when you use the result commercially. For businesses and designers, that takes the copyright headache off the table.
In late January 2026 a new model shipped — Firefly Fill & Expand. The headline upgrade: generation now runs at up to 2K (2048×2048 pixels), double the previous size. In practice that means you can paint in far larger areas before softness and stretched detail set in, and the transitions and grain look more natural.
Finished editing? Save it the right way
Generative Fill hands you a layered PSD. To share it or post it online, export to PNG or JPG — and FormatZ will convert it into whatever format you need, right in your browser.
Convert PNG to JPGWhere the limits are
Generative Fill is impressive, but it's no magic wand. It pays to know where it stumbles.
- Fine detail and text. The AI can still mangle letters, fingers and intricate patterns — inspect at 100%.
- Large areas. The bigger the region you fill, the higher the odds of "hallucinations"; it's better to paint in pieces.
- Exact matches. The result doesn't always match the prompt on the first try — that's normal, generate variations.
- File format. The layered working file is a PSD; for the final export PNG (quality, transparency) or JPG (lightweight for the web).
If you want to refresh the format differences before exporting, see PNG vs JPG and what is PNG. Every available conversion is on the formats page.
