Photoshop Generative Fill: The Complete 2026 AI Guide

Photoshop Generative Fill no longer runs on Adobe Firefly alone: as of the August 2026 Photoshop update, you can switch Generative Fill to Google’s Nano Banana (Gemini 2.5 Flash Image) or Black Forest Labs’ FLUX.1 Kontext Pro from a dropdown in the Contextual Task Bar, each suited to different kinds of edits.

Contents hide

What Changed in Photoshop Generative Fill in 2026?

Until this update, Generative Fill only used Adobe’s own Firefly Image model to generate pixels inside a selection. Photoshop now exposes a model picker next to the prompt box, letting you route the same fill request to Firefly, Nano Banana, or FLUX.1 Kontext Pro without leaving the canvas. Adobe rolled this out through the Photoshop desktop app’s Creative Cloud update channel rather than a full version bump, so the feature can appear even if your version number hasn’t changed.

The change followed months of Firefly adding partner models for full-image generation, and Generative Fill is the first core Photoshop editing tool to inherit that same model choice rather than just Firefly’s standalone generation panel. Digital Camera World’s hands-on comparison found the practical differences between models show up most clearly on realistic composites and text-heavy generations, which lines up with how Adobe positions each model’s strengths.

Image expansion, the tool that extends a photo’s canvas and fills the new area, picked up the same choice: you can now expand with traditional Content-Aware Fill or route the extended area through generative AI. Adobe documents both changes in its official Photoshop release notes.

Which AI Models Can You Use in Generative Fill?

Photoshop’s Generative Fill dropdown currently offers three models, each with different strengths and a different credit cost. Firefly stays the default and is included in most Creative Cloud plans without an extra per-generation charge, while the two third-party models are billed against your Generative Credits balance.

  • Adobe Firefly Image Model – commercially safe training data, fastest generation, included in most plans
  • Nano Banana (Gemini 2.5 Flash Image) – strong at stylized elements, imaginative additions, and text inside images
  • FLUX.1 Kontext Pro – strong at contextual accuracy and matching existing lighting and perspective

How Does Nano Banana Perform in Photoshop Generative Fill?

Nano Banana, Google’s nickname for the Gemini 2.5 Flash Image model, handles stylized and imaginative generations well inside Photoshop Generative Fill, including adding text that renders legibly inside the image rather than as garbled shapes. It tends to be the better pick when a prompt calls for something that doesn’t already exist in the photo, like adding a fantastical object or a graphic element with readable type.

Where Nano Banana is weaker is matching the exact lighting and camera geometry of the surrounding photo for realistic composites; it favors creative interpretation over strict environmental accuracy. For product photography or real-estate retouching where the added content has to look like it was always there, that trade-off usually points you toward a different model.

Photoshop Generative Fill AI models producing colorful abstract art

How Does FLUX.1 Kontext Pro Compare to Nano Banana?

FLUX.1 Kontext Pro, built by Black Forest Labs, specializes in contextual accuracy: it reads the spatial relationships, shadows, and perspective already in your photo and tries to match them when generating new pixels. That makes it the stronger choice for object removal, background extension, and composites where the goal is invisibility rather than creative flair.

The trade-off is the opposite of Nano Banana’s: FLUX.1 Kontext Pro is more conservative with imaginative requests and less reliable at rendering legible text inside a generated area. Photographers doing environmental harmonization tend to prefer it, while designers adding stylized graphic elements tend to reach for Nano Banana instead.

Firefly vs Nano Banana vs FLUX.1 Kontext Pro: Which Should You Use?

There’s no single best model across every task; the right pick depends on what you’re generating and whether cost or realism matters more for that specific edit. The table below summarizes how the three compare on Photoshop’s Generative Fill as of the August 2026 update.

ModelBest forWeaknessCost
Adobe FireflyGeneral fills, commercially safe assetsLess imaginative than rivalsIncluded in most plans
Nano BananaStylized additions, in-image textWeaker environmental matching10 credits per generation
FLUX.1 Kontext ProObject removal, realistic compositesWeaker at imaginative or text requests10 credits per generation

How Should You Write Prompts Differently for Each Model?

Firefly and FLUX.1 Kontext Pro both respond well to short, literal prompts that describe what should physically be in the selection, like “wooden park bench, weathered gray paint, matching afternoon light.” Overloading either model with stylistic adjectives tends to produce muddier results than a plain physical description.

Nano Banana tolerates more creative and compound prompts, including requests that combine an object with in-image text, like “neon open sign, glowing pink letters reading OPEN 24/7.” Because Nano Banana leans on Gemini’s broader language understanding, it also handles longer, multi-clause prompts more reliably than Firefly does, though very long prompts still benefit from listing the most important detail first.

What Are Common Generative Fill Mistakes to Avoid?

The most common mistake is picking a model by habit instead of by task: defaulting to whichever model you used last, even when the current edit calls for the other one’s strengths, wastes credits on regenerations. Making a selection that’s too tight around the target object is a close second, since all three models need some surrounding context in the selection to blend edges convincingly.

  • Selecting only the object instead of including a margin of surrounding background
  • Using Nano Banana for realistic object removal instead of FLUX.1 Kontext Pro
  • Regenerating the same prompt repeatedly instead of adjusting wording after a bad result
  • Skipping Harmonize on composites that mix a generated element with existing photo layers

How Do You Switch AI Models in Generative Fill?

Make a selection on your canvas, open the Contextual Task Bar, and click into the Generative Fill prompt box. A small model-name dropdown sits next to the prompt field; clicking it lists Firefly alongside the available third-party models. Pick a model, enter your prompt, and generate as usual — Photoshop remembers your last-used model for the next fill in that session.

  1. Make a selection with the Lasso, Marquee, or Object Selection tool
  2. Open the Contextual Task Bar and click the Generative Fill prompt box
  3. Click the model dropdown next to the prompt field
  4. Choose Firefly, Nano Banana, or FLUX.1 Kontext Pro
  5. Enter your prompt and click Generate

How Many Credits Does Each Model Cost?

Generative Fill powered by Nano Banana or FLUX.1 Kontext Pro deducts 10 Generative Credits per generation from your Creative Cloud allotment, since both run as premium third-party models inside Firefly’s credit system. Firefly’s own model is included in most paid Creative Cloud plans without the same per-generation deduction, though heavy use can still hit plan-level Firefly limits.

If you’re on a plan with a small monthly credit allowance, it’s worth defaulting to Firefly for routine fills and reserving Nano Banana or FLUX.1 Kontext Pro for the specific shots where their strengths actually matter. Running ten test generations across all three models on a single image can burn through a month’s third-party credit budget quickly.

When Should You Use Content-Aware Fill Instead of Generative AI?

Content-Aware Fill, Photoshop’s older non-generative tool, is still the right choice when you’re removing something from a busy, textured background like grass, water, or fabric, since it samples real pixels from the image instead of inventing new ones. It’s also instant and free, with no credit cost and no risk of the AI introducing an object that wasn’t there before.

Reach for a generative model instead when the area you’re filling needs new detail that doesn’t exist elsewhere in the photo, such as extending a sky, adding an object, or filling a gap after moving a subject. For simple patches on repetitive textures, Content-Aware Fill is often faster and more predictable than any of the three AI models.

How Does the New Harmonize Tool Work With Generative Fill?

Harmonize is a separate Photoshop tool, also new in the 2026 release cycle, that automatically matches color, lighting, and shadow direction between a composited layer and its background. Used after Generative Fill, it can smooth over small mismatches left by whichever AI model you generated with, particularly useful when Nano Banana’s more stylized output needs to blend into a photorealistic scene.

Select the layer you want to blend, open Harmonize from the Contextual Task Bar, and let it analyze the composition; it then proposes a color and lighting match you can apply or adjust manually. It won’t fix a generation that’s structurally wrong, like an object floating at the wrong angle, but it closes the gap on color temperature and shadow direction reliably.

What Other AI Features Are New in Photoshop and Camera Raw?

The same update cycle added an AI-powered blemish removal tool to Camera Raw that treats blemishes and moles as separate categories, so you can clear up skin imperfections without accidentally erasing a mole you wanted to keep. Adobe also improved mask edge controls, added Render to DNG for exporting edited files back to a raw-compatible format, a bidirectional gradient tool, autofocus point display, and a new vectorscope for color grading work.

None of these require Generative Fill to use, but they land in the same update and reflect the same direction: giving editors more automated tools while keeping manual override available at every step. If you’re building out a broader AI toolkit alongside Photoshop, our roundup of AI tools for graphic designers covers complementary options.

How Does This Compare to Using Nano Banana or Firefly Outside Photoshop?

Running Nano Banana inside Photoshop’s Generative Fill gets you selection-aware editing: the model only touches the pixels inside your selection and blends against the surrounding image, which standalone image generators don’t do. If you want to compare Nano Banana’s raw output quality against other image generators outside of a selection-based workflow, our GPT Image 1.5 vs Nano Banana Pro comparison and our Nano Banana vs Seedream AI comparison both test the models head-to-head on full-image generation.

For logo and brand work specifically, Adobe Firefly’s dedicated vector and text-effect tools inside Illustrator and Photoshop still outperform general-purpose image models on precision, since they’re trained for that narrower task. Our guide to using Adobe Firefly for logo design covers that workflow separately from Generative Fill.

How Should Photographers vs Designers Use the New Models Differently?

Photographers doing retouching, object removal, or sky replacement work should default to FLUX.1 Kontext Pro or Firefly, since photorealism and environmental consistency matter more than creative flexibility for that kind of edit. A wedding or real-estate photographer removing a stray sign or extending a background wants the fix to be invisible, not stylistically interesting.

Designers building composite graphics, social posts, or mockups have more room to lean on Nano Banana, particularly for anything involving in-image text or a deliberately stylized look. A designer mocking up a storefront sign or a poster element benefits from Nano Banana’s stronger text rendering, even though the result is meant to look designed rather than photographed.

Is Switching AI Models in Photoshop Worth the Extra Credits?

Switching models is worth the extra credits when a Firefly generation isn’t giving you the result you need, not as a default habit for every fill. Start with Firefly since it’s already included in your plan, and escalate to Nano Banana for stylized or text-heavy requests, or FLUX.1 Kontext Pro for realistic composites, only when Firefly’s output falls short.

Track how many third-party generations a given project actually needs before committing to a workflow that leans on Nano Banana or FLUX.1 Kontext Pro by default. A freelancer billing a client for retouching time can usually absorb a handful of premium generations per project, but a high-volume studio processing hundreds of images a week should budget credit usage the same way it budgets stock-photo licensing costs.

Does Photoshop Label Images Edited With These AI Models?

Yes. Photoshop attaches Content Credentials metadata to files edited with any of the three Generative Fill models, recording that generative AI was used as part of the edit history. That metadata travels with the file when exported to formats that support it and can be inspected later to confirm whether, and how, AI tools touched an image.

This matters more for Nano Banana and FLUX.1 Kontext Pro than for Firefly, since publications, stock agencies, and some clients now require disclosure of which model generated specific content, not just whether AI was involved at all. If you work with outlets that have their own AI-disclosure policies, check whether they require the model name specifically before submitting a Generative Fill edit made with a third-party model.

How Big Is Adobe Firefly’s AI Ecosystem?

Adobe Firefly crossed 24 billion cumulative generated assets in 2025, according to Digital Camera World, which called it the fastest ramp recorded for any commercially licensed AI image product. That count includes every Generative Fill edit, text effect, and vector generation across Adobe’s apps, not just standalone image generations, so it isn’t directly comparable to a raw image-count figure from a tool like Midjourney.

The scale matters for Generative Fill specifically because Adobe built its credit and model-routing system to handle that volume, which is part of why adding third-party models required a dedicated credit tier rather than unlimited access. It also explains why Firefly, not Nano Banana or FLUX.1 Kontext Pro, stays the default model: Adobe controls its cost and licensing terms directly, while the partner models run through external infrastructure billed per generation.

Frequently Asked Questions

Do I need a paid Creative Cloud plan to use third-party models in Photoshop Generative Fill?

Yes. Nano Banana and FLUX.1 Kontext Pro draw from your Generative Credits balance, which is tied to a paid Creative Cloud subscription; free or trial accounts have limited or no access to the third-party models.

Does Photoshop Generative Fill still work offline?

No. Generative Fill requires an internet connection for every model, including Firefly, since generation happens on Adobe’s or the partner model’s servers rather than on your device.

Can I use Nano Banana or FLUX.1 Kontext Pro in Photoshop’s Generative Expand?

Image expansion currently offers a choice between traditional Content-Aware Fill and Adobe Firefly’s generative AI; third-party model support for expansion specifically has not rolled out as broadly as it has for Generative Fill selections.

Is content generated with Nano Banana or FLUX.1 Kontext Pro safe for commercial use?

Adobe’s Firefly model carries Adobe’s commercial-safe indemnification since it trains only on licensed and public-domain content. Third-party models like Nano Banana and FLUX.1 Kontext Pro don’t carry that same Adobe indemnification, so check each provider’s own commercial-use terms before shipping generated content in paid work.

Why is my Generative Fill result inconsistent with the rest of the photo?

Mismatched lighting or color usually means the model prioritized the prompt over environmental matching, which is more common with Nano Banana than FLUX.1 Kontext Pro. Try regenerating with FLUX.1 Kontext Pro for better contextual accuracy, or run the result through Photoshop’s Harmonize tool to correct color and shadow direction after the fact.

How much does a Nano Banana generation cost compared to Firefly?

Nano Banana and FLUX.1 Kontext Pro each cost 10 Generative Credits per generation, while Adobe’s own Firefly model is included in most Creative Cloud plans without that same per-generation credit deduction.

Subscribe for Newsletter

Chat Channel
F in WA @