Generative Fill in Adobe Photoshop

By Michael Peters · Published March 17, 2026 · Last updated March 24, 2026

← Back to Blog

What is Generative Fill?

Generative Fill is an AI-powered feature in Adobe Photoshop, introduced in 2023 as part of the Adobe Firefly integration. It allows users to add, remove, or replace image content using a text prompt, with the AI generating new pixels that match the surrounding image in terms of lighting, perspective, style, and colour. The results are placed on a non-destructive generative layer, preserving the original image data.

How It Works

Generative Fill is powered by Adobe Firefly, Adobe's family of generative AI models trained exclusively on licensed, royalty-free content. This makes it safe for commercial use without the copyright concerns associated with some other AI image tools.

To use Generative Fill:

1. Make a selection in Photoshop using any selection tool (lasso, rectangular marquee, subject selection, etc.)
2. Click Generative Fill in the contextual task bar that appears
3. Enter a text prompt describing what you want to appear in the selection (or leave it blank to simply remove content)
4. Click Generate — Photoshop produces three variations to choose from
5. Select the best variation; you can regenerate as many times as needed

Key Use Cases

Removing Objects

Select any object in a photo and leave the prompt empty. Photoshop fills the area with contextually appropriate background content, effectively removing the object. This is significantly more powerful than Content-Aware Fill for complex backgrounds.

Extending Images

Generative Fill can extend the canvas beyond the original image boundaries. Select the empty canvas area and generate content to seamlessly extend the scene — useful for changing aspect ratios or creating wider compositions.

Adding Objects

Describe an object in a prompt and Photoshop places it in the selected area, matching the lighting and perspective of the existing image. A prompt of "a red umbrella" will insert a plausible umbrella that fits naturally into the scene.

Replacing Backgrounds

Use Select Subject to isolate a foreground subject, invert the selection, and generate a new background. The AI respects the lighting on the subject when creating the background scene.

Non-Destructive Workflow

All Generative Fill content is placed on a dedicated generative layer. The original image is never modified. You can hide, delete, or modify the generative layer at any time, and regenerate different variations independently. This makes it suitable for use in professional production workflows where change history matters.

Limitations

• Results can be inconsistent for very specific or detailed prompts
• Generating requires an internet connection (processing happens in Adobe's cloud)
• Complex structural elements (hands, text, specific faces) can produce imperfect results
• Very large selections may produce lower-quality results than smaller, focused generations

Generative Expand

A related feature, Generative Expand, works on the Crop tool. Extend the canvas in any direction and Generative Expand fills the new area with AI-generated content that matches the existing image. This makes it easy to reformat images for different aspect ratios without cropping.

How Generative Fill Actually Works

The underlying technology is Adobe's Firefly image model, which was trained on Adobe Stock images, openly licensed content, and public domain works. The training dataset was deliberately chosen to avoid copyright-encumbered material, which is the specific commercial differentiation Adobe makes against models trained on scraped web content.

When you make a selection and submit a prompt, Photoshop sends the selection bounds, the surrounding image context (a region beyond the selection edge), and your text prompt to Adobe's cloud inference servers. The model performs a diffusion-based inpainting pass — essentially starting with noise and progressively denoising it to produce content that is coherent with the surrounding pixels. The context region is critical: it is what allows the model to match lighting direction, colour temperature, texture, and perspective. Photoshop requests three independent samples from the model and presents them as variations in the Properties panel.

The result lands on a dedicated generative layer within a layer group. That layer contains the generated pixels at the selection's native resolution. Crucially, nothing about the original image is written to — Photoshop's non-destructive layer architecture means you can discard or regenerate at any time.

Generative Fill vs Content-Aware Fill

Content-Aware Fill has been in Photoshop since CS5, and it remains a capable tool for many object removal tasks. The key difference is that Content-Aware Fill works entirely locally: it samples from other parts of the same image and tiles or blends those samples into the selected area. This works well when the background is relatively uniform — grass, sky, pavement, sand — but struggles with complex, structured backgrounds like architecture, crowds, or detailed foliage at close range.

Generative Fill does not sample from elsewhere in the image. It synthesises new content from scratch, informed by the context but not constrained to replicate what is already there. That makes it capable of producing believable content in areas where Content-Aware Fill would produce smearing or visible repetition patterns. The trade-off is that you need an internet connection and a Creative Cloud subscription, and the operation takes several seconds rather than milliseconds. For a quick object removal on a simple background, Content-Aware Fill is often faster and requires no cloud round-trip.

Practical Workflow Examples

One of the most practical applications is aspect ratio conversion for multi-platform publishing. A photograph captured in portrait format needs to work as a landscape banner. Extend the canvas horizontally on both sides, select the empty areas, and let Generative Expand fill them with plausible scene content. What used to require a skilled retoucher and significant time now takes about thirty seconds. The quality is not always perfect — you may need to try two or three generations — but it gets you to a usable result far faster than starting from scratch.

For product photography retouching, the background replacement workflow is particularly effective. Isolate the product with Select Subject, invert the selection, and generate a new background with a simple prompt like "white studio gradient" or "polished concrete floor". The AI adapts the generated background to match the product's ground shadow and ambient light, producing a composite that reads as convincingly lit rather than obviously pasted.

Sky replacement is another strong use case, though Photoshop also has a dedicated Sky Replacement tool for this. Generative Fill gives more control through the text prompt, allowing you to specify mood as well as content — "overcast morning sky" versus "dramatic sunset" produces meaningfully different results.

Limitations and Best Practices

Generative Fill is genuinely impressive, but it has consistent weak spots that experienced practitioners learn to work around. Hands and faces remain the most problematic subjects — the model frequently produces subtly wrong anatomy that is convincing at a glance but fails under scrutiny. For any image where a person is clearly visible, treat generated content near faces or hands as needing manual cleanup rather than a finished result.

Text is reliably poor. If your selection includes or borders any typographic element, the generated content will likely produce nonsense characters or blurred approximations. Work around this by removing text before generation and re-adding it as a Photoshop text layer afterwards.

The quality of your selection matters more than the quality of your prompt. A rough, feathered selection on a complex edge will produce visible seams regardless of how good the model is. Invest time in clean selections using Refine Edge or Select and Mask, and the generated content will integrate far more naturally. Feathering the selection by 2–5 pixels before generating typically helps blend the generated edge into the surrounding image.

Prompt specificity follows a curve of diminishing returns. A simple, clear prompt of three to five words often outperforms an elaborate descriptive sentence. "Wooden floor" works better than "rustic aged oak hardwood flooring with visible grain pattern". The model reads the image context and fills in the detail — you are steering, not describing the pixel output.

Implications for Creative Professionals

The practical effect on professional retouching work is less dramatic than the marketing suggests, but it is real. Tasks that previously required a skilled retoucher and an hour of careful work can now be sketched out in minutes, leaving more time for the nuanced manual work that still matters. The workflow shifts from "can this be done?" to "is this result good enough, or does it need further refinement?"

For photographers, the technology has effectively raised the baseline for what clients expect from delivery images. Backgrounds that would once have been accepted as-is can now be quickly cleaned or replaced, and clients increasingly expect options that would previously have been charged as bespoke retouching work. Managing those expectations — and pricing accordingly — is the real professional challenge, not learning the tool itself.

From a copyright and intellectual property perspective, Adobe's commercial-use guarantee for Firefly-generated content removes one source of risk. Content generated by Firefly does not reproduce identifiable copyrighted works, and Adobe indemnifies enterprise customers against claims arising from Firefly outputs used within its products. This is a meaningfully different legal position from tools built on models trained on uncleared web content.

Photoshop Automation and Plugins

For developers, Generative Fill is accessible through Photoshop's UXP plugin API, allowing automated generation workflows within custom plugins. Read about automating tasks in Photoshop with Actions, Scripts and Plugins, or see our Photoshop development page for more information on what Mapsoft can build for you.