Using Photoshop AI Models: Nano Banana & Flux in a Practical Test

By Stefan Petri Last updated on November 14, 2026 Shown in: Adobe Photoshop CC 2026

Do you know this? You have a good portrait, but the outfit doesn't fit the campaign, and with selection tools and layer masks you would be working for hours. In this tutorial, I'll show you how to use external AI models like Google Nano Banana and Flux directly from the interface in the beta version of Photoshop. Using the example of a business outfit that can be transformed into a Christmas costume with a text command, you'll see where these models are strong, where they fail, and how to refine the results.

Play video from YouTube

The video is only loaded from YouTube when you play it – data (including your IP address) is then transmitted to YouTube/Google. More in the privacy policy

Key Takeaways

• External AI models can be selected directly in Photoshop. You don't have to switch to another program for this.
• Nano Banana (Google Gemini) preserves the person's face, pose, and details significantly better than the standard version of Firefly.
• Flux is a usable alternative, but it often changes the background and face somewhat more.
• For generative expand, only Firefly works in the example shown. The choice of model is limited here.
• If the first prompt fails, a second attempt, a different selection, or a different model usually helps.
• A soft brush with a mask removes edges and small offsets that remain visible after generating.

Step 1: Install the Photoshop Beta

Before you get started, you need the beta version of Photoshop. The features for external AI models are already included there, even though they are not yet available in the full version. To do this, open the Creative Cloud desktop app, switch to the Apps section, and go to the Beta tab. There you will find Photoshop Beta. If the app is already installed, it says "Open"; otherwise you can install it with one click. The interface is in German and largely corresponds to the full version, except that some new AI tools are already integrated.

Step 2: Prepare the image and enter the first prompt

Open your photo in Photoshop. In the example, it is a woman in a business look. Select the person or the area you want to change, and then use Generative Fill. Enter the command, for example "Christmas suit" or "Christmas costume". German prompts work without any problem, so you don't have to switch to English. Make sure your command is as specific as necessary, but worded as openly as possible. A Santa hat, for example, can be added in the same prompt.

Step 3: Select the AI model

This is where the real advantage of this feature lies. Besides the prompt, you can choose the model. By default, Firefly Image 3 is set. In the test, this model completely changed the woman's face, even though only the costume was supposed to be adjusted. The results for three variations were also rather unusable, because the person was barely recognizable.

Therefore, switch to Google Nano Banana, which you can recognize by the G for Google. The difference is clear: face, eyes, and posture are preserved, and only the costume is laid over the person. The processing time is also noticeably shorter.

Step 4: Compare results with Flux

Next, try Flux if you want another option. In the test, the woman's face remained more similar than with Firefly, but the background also changed. The eyebrows and the hairstyle also differed slightly. For portraits, Flux is therefore a usable alternative, but you should check the result carefully before accepting it.

Step 5: Work around the problem with cutting out

A drawback became apparent when the result with Nano Banana was to be cut out: the person could not be cleanly separated from the background. One solution is to select the entire area instead of just the person. To do this, hold down the Ctrl key (Cmd on a Mac) and select the desired area with the selection tool. In the example, this resulted in the background staying stable. However, the person looked slightly blurrier in the foreground, and an additional Christmas tree appeared. This is not a perfect result, but it shows how you control with the selection what gets changed.

Step 6: Enlarge the Image with Generative Expand

Does the landscape-format image not fit your layout? Then use the Crop tool and drag the canvas in the desired direction, for example into a square or portrait format. Next, select the new, empty area and choose Generative Expand. You can also enter a prompt, such as a computer in the foreground. If you leave the field empty, the AI decides on its own how the image could continue. In the example, it sensibly continued the scene with the book and paper and adopted the look of the original.

Please note, however: in the test, only Firefly was available for Generative Expand. Choosing an external model is not possible at this point.

Step 7: Add Elements and Generate Text

With Generative Fill, you can also insert new elements into an existing image. Select the area, enter the prompt, and choose a model. If you don't like the result, simply switch the model.

AI models are often still weak when it comes to text. In the example, an airplane was supposed to pull a banner with the text "Heute Freibier" ("Free beer today"). Firefly and Flux failed, and Gemini delivered the text best. However, Gemini displayed only a single variant. In the result, the area around the banner was not clean, and the edges of the selection were visible.

Step 8: Post-Processing with a Mask and Soft Brush

If edges or transitions are visible, a layer mask will help. Choose a soft brush with a large diameter and mask the untidy areas. This way the edges disappear, and the result looks more harmonious. In the example, the text was easy to read afterwards and the transition was hardly noticeable.

Step 9: Improve a Photo with Multiple Attempts

The example with the photo of the daughter shows that not every prompt works right away. The first attempt with Nano Banana and a rectangular selection failed. Only the second prompt led to a usable result. If necessary, choose a different selection, for example the Lasso tool instead of the rectangle. The result was good, but stones in the background were slightly shifted. With a mask and the soft brush, this error can be corrected. The shoes and the posture were preserved, even though the leg was realigned.

Conclusion for Your Workflow

The combination of external models and Photoshop tools saves time. In the example, the redesign with a text command and one or two attempts took a few minutes, whereas working manually would have taken about an hour. Nevertheless, you should check every result critically, especially faces, text, and backgrounds.

Summary – Using AI Models in Photoshop: Nano Banana & Flux in a Practical Test

You have seen how to use external AI models directly in the beta version of Photoshop. Google Nano Banana proved the most suitable in the test when the face and person are meant to be preserved. Flux also delivers usable results, but changes the background. Firefly remains the only option for Generative Expand. Plan for more post-processing with text elements, and don't hesitate to try out several prompts or selection tools. The second part of the tutorial covers AI upscaling and other AI features.

Trainer About Stefan Petri Together with his brother Matthias, Stefan Petri runs the popular specialist forum PSD-Tutorials.de and the e-learning platform TutKit.com, which focuses on the training and further education of digital professional skills. All courses by Stefan Petri →