Photoshop will soon gain in-app support for using third-party AI tools to create, design, and edit within the longstanding software. On Thursday, September 25, Adobe announced support for third-party AI models for Generative Fill inside Photoshop Beta, including Google Gemini 2.5 (better known as Nano Banana) and Black Forest Lab’s Flux.1 Kontex [pro].
Third-party AI model support has previously launched inside Adobe Firefly, but the move is the first time Adobe is bringing the option to select different AI models to a Creative Cloud app. The update gives users the ability to choose Nanao Banana or Flux along with Adobe’s Firefly AI models when using Generative Fill.
I sat down with Adobe’s Joel Baer, Director of Product Management for Adobe’s generative AI efforts, for a demo of the new capabilities ahead of the launch, just a few weeks after I tried out Nano Banana inside the Gemini chatbot. Mixing Photoshop with Nano Banana takes the AI out of the chatbot and gives it Photoshop tools.
The third-party AI support is coming only to Generative Fill, but the tool includes support for masking, layers, and localized edits. That brings more control and fine-tuning than what’s possible with using Nano Banana inside the Gemini chatbot or inside Google Photos.
For example, users can highlight the entire canvas to generate an entirely new image, or use a brush to select only part of an existing image to edit.
Nano Banana’s ability to remix multiple images and the updated capabilities to generate AI photos that look more like real people are intact inside Photoshop as well. During the demo, Adobe placed two different images on the canvas of a drum set and a woman, and the AI combined the two to place the woman playing behind the drum set.
Like with the existing Generative Fill tool, the AI generation becomes a new layer, not part of the existing image itself. That makes it easy to delete or edit the generative layer. Inside the Properties panel of the generative layer, users can also change the AI model to try the same prompt with a different model, allowing users to switch from Adobe Firefly to Nano Banana, Flux, and back. Previous results are saved, so editors can go back and choose the best result.
Generative AI isn’t without its surprises. For example, in one demo, Adobe asked the AI to change a photo’s lighting from day to night. It did a fairly decent job at tweaking the lighting, except it added a car. The car was properly lit and even had a light trail, but it was driving on a sidewalk about to hit a street sign. Adobe’s editing tools can be further used to take out those glitches, using tools like the remove brush or masking out part of the generative layer.
Different AI training models vary in strengths and weaknesses, allowing users to choose a model means creators can choose the model that produces the best results. Adobe recommends Nano Banana for stylized elements, graphic details, and scene additions. Adobe suggests trying Flux for its contextual accuracy, perspective, and environmental harmony.
Unlike Firefly models, third-party AI models may use content scraped from the web
Adobe's safe-for-commercial-use AI label, however, doesn’t extend to those third-party partner models. Adobe trained Firefly using licensed images and public domain works, and the company thus says that it’s safe to use for commercial projects. Adobe doesn’t make any such claims for third-party models, where the user is responsible for vetting the ethics and safe use cases.
Google Gemini is trained on “publicly available data,” according to the company. That includes videos from YouTube.
Black Forest Labs doesn’t disclose where training data for Flux comes from, but the company’s Flux was also previously used on X’s Grok, an AI with more lax rules, including the ability to generate images of political figures and licensed characters. Grok has since switched to its own Aurora image generation model. Black Forest Labs indicates that Flux’s training data set went through pre-screening to eliminate NSFW (not safe for work) images.
Adobe says using the partner models does not require a separate subscription to Gemini or Flux. However, using a third-party AI model is a Premium feature, which requires a specific subscription plan. Adobe’s Creative Cloud All Apps Pro plan includes premium generative credits, and Firefly plans also include third-party partner model support. The partner models are not part of the single app Photoshop plan or Lightroom-Photoshop plans without also adding a Firefly subscription.
However, Adobe is temporarily allowing all Photoshop users to try out the third-party model capabilities without a premium subscription and without deducting from generative credits. Standard plan subscribers can generate up to 100 times without a subscription during the trial period. Creative Cloud Pro subscribers are limited to 500 generations per day per model.
The support for non-Adobe AI models is coming first to Photoshop Beta, which is separate from the fully-fledged Photoshop and can be downloaded from the Creative Cloud desktop app. The update with third-party AI support is available beginning September 25.
You may also like
Browse the best photo editing software, or learn how to spot an AI-generated image.
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.
