Adobe’s showing off experimental features that may shape the future of Photoshop

Editor’s note: The Adobe Max Sneaks presentation begins at 5:30 pm ET / 2:30 pm PT, and you can watch via livestream. We’ll be attending live and updating this story with new details throughout the event.

This week at Adobe Max, we’re getting a look at some of the experimental tools the company is developing during its annual Sneaks presentation. The session gives Adobe’s researchers a chance to show off works in progress, rather than announce finished products. These projects could eventually become features in Adobe’s applications (and, indeed, several things we saw at last year’s show are now available in Photoshop), though there’s no promise they’ll ever actually ship, or end up looking like what’s shown off. Think of it as a little glimpse into what Adobe teams are working on. 

This year’s lineup includes 10 projects, ranging from experiments in sound editing, generated-video timing, character development, real-time video manipulation and vector design. Several are aimed at photographers, too, such as AI-powered brushes that account for a subject’s shape and lighting, a tool that adapts Photoshop edits across different images and a system that generates new viewpoints from a single photograph.

While most remain demonstrations, Adobe is, for the first time, offering early research previews to everyone of two projects, Deep Brush and Hands On, through its new Labs Playground. Users can experiment and provide feedback, rather than simply access finished tools.

Project Deep Brush: Editing that follows the scene

First up is Project Deep Brush, which aims to make brush editing respond to what’s in a photograph rather than simply sit on top of it. In its blog post about the feature, Adobe explained that it wants creatives to “think of every photo as a real-world scene to explore, rather than a set of pixels to adjust.” 

Adobe says the tool uses AI to infer properties such as shape, color and lighting from a single image, separating the image into what it calls “physically decomposed layers.” Those layers then provide controls for edits that respect the scene’s geometry, shadows and illumination.

For photographers, potential uses include changing hair color or applying makeup while retaining the effects of facial shape and lighting. Adobe also describes placing graffiti, signage, a logo or texture on a curved or angled surface — such as a bottle, garment or building — with the edit automatically following that surface’s contours and lighting.

In the demo during Sneaks, the project manager used the brush in a few different ways. First, he painted graffiti on a concrete wall, which automatically adjusted to the texture and angle of the wall. The tool even let him paint instantly beneath the hands of an athlete on the wall. Then, he showed off rippled fabric, which is a more complicated surface. Despite that, the Deep Brush automatically blended with the folds, looking like it was part of the fabric. The tool also worked with text for this example.

Deep Brush also works by painting light directly into a scene, sort of like the recently released Relight, though with a brush. The presenter was able to adjust the color temperature of the light and make other adjustments, and selectively brush light where he wanted it.

Rather than treating a brushstroke simply as color applied over pixels, Deep Brush explores how that stroke could interact with the surfaces and illumination represented in the image.

Project Rinse Repeat: Copy the retouching, not the pixels

Project Rinse Repeat could be particularly useful for photographers working through repetitive editing projects. Adobe frames it as a more intelligent, adaptive version of copying and pasting. It takes an edit made in Photoshop and adapts it to another target, using the context of that target to decide which adjustments apply and how they should change.

In Sneaks, the product manager for the feature showed the process of creating a shadow for a product photo, but then there were five more products that also needed a shadow. With today’s tools, such a task would require lots of repetitive work and manual adjustments. With Project Rinse Repeat, though, she was able to copy the shadow she made for the first product and paste it onto the others, which was automatically adjusted based on the scene’s context and each product’s specific details.

Instead of simply repeating a fixed sequence of operations, as the traditional copy-and-paste method does, Rinse Repeat selects and adapts your edits to suit the new subject. Importantly, it performs those changes using native Photoshop tools, layers and masks. She said they are not generative, and the results remain non-destructive and editable, so users can inspect and refine them rather than being left with a flattened result.

As a photographer who doesn’t use generative AI, this is the experimental tool that gets me the most excited. I copy edits to other photos regularly, and I could see this saving me a lot of time and effort.

Project Hands On: Drag instead of prompting

Project Hands On is designed to let you make edits, including ones using generative AI, without having to describe them in text to a chat assistant. It analyzes an image and creates controls directly on the canvas, allowing users to manipulate specific details by dragging. Adobe gives the example of control points on a mug’s handle that let users reshape it directly.

The researchers found shapes particularly suited to this approach because they can be difficult to describe in words. Adobe says that potential applications include adjusting a product’s proportions, helping photographers fine-tune specific details or refining an AI-generated image that needs a targeted correction.

In the Sneaks demo, the project manager showed how to change a rectangular perfume bottle into a diamond shape and shorten a loveseat into a chair. It also works on photographs to adjust how a subject is posed.

While Photoshop already offers direct manipulation through tools such as Puppet Warp, Hands On creates content-aware controls from its analysis of the image rather than relying on a mesh and manually placed pins. It also stems from the existing ‘Edit Object’ tool in Adobe Firefly.

Project Sight Span: Move the camera after the shot

Project Sight Span turns a single image into a navigable environment, allowing users to explore new viewpoints and render still images or camera-controlled videos. Adobe suggests uses such as trying alternative angles for a product photograph or investigating possible views of a location before visiting it.

The system first creates a lightweight 3D scene for exploring compositions, then uses a separate image-generation step to render the selected view. Crucially, it must infer geometry and details that aren’t visible in the original photograph, so those new angles contain generated interpretations rather than additional photographic records of the scene.

While the final generated image doesn’t get me too excited, as an educator, I do see opportunities to use this as a learning tool. I could take my students’ images that would benefit from a different angle and use this to show them what it would actually look like instead of trying to just describe it. It could give them a clearer view of how changing camera angles can drastically change the impact of a particular image.

Project Scene Sweep: One video edit, multiple shots

For photographers who also shoot video, Project Scene Sweep aims to keep edits consistent across cuts. It identifies people, objects and locations throughout a sequence, allowing changes to follow their subjects across different shots.

Users can guide changes with a text prompt or an edited frame. Adobe’s examples include removing a distracting prop, replacing a logo or changing wardrobe across a sequence while preserving the original performance and pacing.

Project Custom Filters: Describe your own effect

Alongside the Sneaks projects, Adobe is offering an experimental preview of Project Custom Filters, which lets users create image filters by describing the effect they want. The company specifies that the tool makes it much easier to get a custom look, rather than relying on plugins, coding or enlisting someone else to build the filter. 

With the tool, you use the AI Assistant to describe what you want. It will then build a filter with controls specific to that, and apply it to the image. Adobe gave examples such as “create a lens distortion filter” or “do a painterly effect.” Users can then refine the result through further instructions and adjustments.

Adobe says this isn’t a static AI-generated image: the filter runs in real time on the image’s actual pixels and works non-destructively, leaving the original untouched. Users can still modify any of the changes to the image, just as you can with other non-destructive filters. 

Filters can also be shared, loaded onto other images using a reference number and remixed through descriptions and controls. It remains an early experiment, though, it is available for testing in the Labs Playground now. .

Which projects can you try?

Adobe’s new Labs Playground will include research previews of Deep Brush and Hands On, alongside Project Custom Filters. The Deep Brush and Hands On previews will be available when the Sneaks presentation begins at 5:30 pm ET / 2:30 pm PT. Adobe is inviting feedback on these early tools, which users can share by emailing [email protected]. 

As a reminder, these are still experimental tools at this point. Adobe hasn’t provided release dates for any of the features mentioned, and they may not ever make their way to the general public.