How to Use Generative AI in Photoshop
Learn how to use Photoshop’s Generative Fill and other generative AI tools to add, remove, or change image elements. Follow the steps and tips for getting more useful results.
Overview
Photoshop’s generative AI is chiefly used through Generative Fill and Generative Expand. Fill adds, replaces, or removes content inside a selection; Expand extends the canvas and synthesizes the newly exposed area. The practical workflow is to define the region precisely, optionally describe the intended result, generate alternatives on a separate layer, and inspect the chosen result before delivery. Adobe requires an internet connection, and current model access and credit rules depend on the selected model and plan. Generative AI features in Photoshop FAQ | Photoshop.
What generative AI in Photoshop does
Generative Fill uses an image selection and, optionally, a text prompt to synthesize pixels for that part of a composition. It can add a described object, replace selected content, or fill a removed object’s area using nearby visual context. Leaving the prompt blank asks Photoshop to infer a fill from the surrounding image. Each generation appears on a Generative Layer, allowing the user to switch among results, conceal the layer, or remove it without overwriting the original. Save in a layered format such as PSD to preserve editability. Generative AI features in Photoshop FAQ | Photoshop.
Generative Expand addresses a different task: increasing canvas dimensions and generating content beyond the original frame. Use Fill for edits within the image; use Expand for a wider crop or missing surroundings beyond an edge.
| Feature | Input | Typical use | Main control |
|---|---|---|---|
| Generative Fill | Selection in image | Add, replace, or remove an element | Selection boundary and optional prompt |
| Generative Expand | Enlarged crop/canvas | Extend background or reframe a photo | Crop handles, optional prompt |
How to use Generative Fill in Photoshop
Basic desktop workflow
- Open the image and preserve an untouched source, for example by keeping the original layer intact or saving a working copy.
- Use a selection tool—such as Lasso, Marquee, or a brush-based selection—to mark the region.
- Choose Generative Fill from the contextual task bar.
- Enter a concise description of the desired content. Leave the prompt blank when you want a context-based fill, especially for a simple removal or gap.
- Choose a model if the model picker is shown, then select Generate.
- Review the variations in the Properties panel. Choose one, or refine the selection or prompt and generate again.
- Inspect the edit at full size and normal viewing scale. Retouch locally if needed, and save as PSD or another layered format to retain the generated layer.
Adobe’s documented process is to open an image, define the area, select Generative Fill, choose an available model, enter a prompt or leave it blank, generate, and choose a variation. Interface options vary by version. Use Generative Fill in Photoshop on desktop | Photoshop. Each generation creates a new Generative Layer, according to Adobe’s Generative AI features in Photoshop FAQ | Photoshop.
Workflow diagram: controlled Generative Fill iteration. The procedure follows Adobe’s documented steps; inspection and refinement are practical quality-control measures, not guarantees of correctness. Use Generative Fill in Photoshop on desktop | Photoshop.
Common edits
- Remove a distraction: select the unwanted object and a small margin; leave the prompt blank or describe the intended replacement background.
- Add an object: select the intended location and describe the object and relevant appearance, such as “small ceramic vase, warm side light.” Adobe Learn recommends shaping the selection approximately to a distinct object silhouette, which gives spatial guidance as well as the text prompt. Adobe Learn - Learn Photoshop - Add to a scene with Generative Fill..
- Replace an element: select it and describe the replacement, for example “replace with a red wool coat.” Leave enough room for edges and shadows to blend.
- Fill contextually: a blank prompt asks Photoshop to use the surrounding image as context; it does not mean no generation occurs.
These are prompt patterns, not guaranteed results. The selection, visible context, selected model, and software version also influence the output.
How to use Generative Expand
- Select the Crop tool.
- Drag crop handles outward to create blank canvas.
- Choose Generative Expand from the contextual task bar.
- Optionally describe what should appear beyond the original frame. For a straightforward continuation, leave the prompt blank.
- Select Generate, compare variations in Properties, and retain the best match for the scene.
Adobe’s Enhance Detail for sharper Generative Fill variations | Photoshop specify the Crop-tool method and blank-prompt option. Expand can help when a portrait is tightly framed or a layout needs more background. Generated surroundings are plausible synthesis, not evidence of what was actually outside the photographed frame.
Prompt and selection strategy
A useful prompt names the object or material and, where relevant, its appearance or setting. Avoid asking the prompt to specify exact placement when the selection already defines the approximate location. For removal, a blank prompt often lets the model infer a replacement from nearby pixels; use a short description when the area needs a specific surface or background.
Selection quality can matter as much as wording. A loose selection may permit unwanted changes; one that hugs an object too tightly can omit shadow or transition areas. For a distinct shape, draw a selection that roughly matches its silhouette. Generate and inspect, then adjust one variable at a time so it is easier to evaluate the change. Adobe’s tutorial similarly recommends a shape-matched selection and reviewing generated variations. Adobe Learn - Learn Photoshop - Add to a scene with Generative Fill..
Access, models, and generative credits
Adobe’s desktop FAQ says generative features require an internet connection, are available where Adobe provides services except China, and support prompt input in more than 100 languages. Access and interface options can change by app version, region, and plan. Generative AI features in Photoshop FAQ | Photoshop.
Credit use is not a single universal rate. Adobe says most standard Generative Fill generations using Adobe Firefly models use one credit per generation; partner models and some other features have different or variable charges. Some plans include unlimited standard generations, while allotments and premium access vary. Check the in-app notice and current Generative credits FAQ | Creative Cloud.
The Photoshop FAQ describes Firefly output as 1024 × 1024 pixels for desktop and web. This is specific to the Firefly output described in that FAQ, not a universal specification for every model or Photoshop document. Check current model details when resolution matters. Generative AI features in Photoshop FAQ | Photoshop.
Limitations and quality control
Generative Fill produces a visual proposal, not a verified reconstruction. It may create inconsistent geometry, texture, shadows, perspective, or object details. A Macworld hands-on review of the 2023 beta found results ranging from realistic to conspicuously inaccurate, including inconsistencies in architectural details. That review evaluated an early beta, not current models; it nevertheless illustrates why outputs need inspection. Photoshop's Generative Fill AI tested on a Mac: Equal parts amazing and unreal | Macworld.
Check details that carry meaning: faces, hands, signs, logos, product labels, repeated structures, and documentary evidence. Compare the result with the source, inspect edges at high magnification, and do not present invented details as faithful records. Keep generated content on separate layers until approval, and verify that the edit is appropriate for the intended context.
A 2024 qualitative study analyzed 566 forum posts by creative professionals about Generative Fill. It reported uses including touch-ups, image expansion, and prototype imagery, alongside perceived productivity benefits and tensions involving human authorship, creative identity, and learned craft. The forum-post sample is selective and qualitative; it does not establish how prevalent those views are among all users or prove productivity gains. A Value-Oriented Investigation of Photoshop's Generative Fill.
Training claims, rights, and provenance
Adobe states that Firefly models are trained on licensed material, including Adobe Stock, and public-domain material whose copyright has expired, and that customer content is not used to train Firefly. These are Adobe’s published policy statements, not an independent audit of every model or partner model available in Photoshop. Approach to generative AI with Adobe Firefly - Adobe.
Adobe says qualifying generative edits automatically attach C2PA metadata recording information such as the application and model used and whether content was generated or modified. This embedded metadata does not itself add a visible label, and other platforms may handle or display it differently. Generative AI content labeling FAQ.
Development and practical conclusion
Adobe announced Generative Fill in Photoshop beta on May 23, 2023, integrating Firefly into a selection-based workflow for adding, extending, and removing image content. Adobe Unveils Future of Creative Cloud With Generative AI as a Creative Co-Pilot in Photoshop.
The reliable approach is to treat Photoshop generative AI as a localized editing aid: select deliberately, use a short prompt only when context is insufficient, compare alternatives, and inspect the pixels before accepting the result. Fill is for changes within a selection; Expand synthesizes beyond the canvas. Separate layers make revision easier, but do not guarantee factual accuracy, visual consistency, or rights clearance.