The use of artificial intelligence in Adobe Photoshop is being increasingly expanded by developers. In the past, the Adobe Photoshop-AI model “Firefly” took centre stage. Two further models have been available in the beta version of the programme for a few weeks now:

• Flux.kontext Pro from Black Forrest Labs
• Nano Banana from Google

Both models have so far been able to prove their efficiency, which is impressively demonstrated by many examples on the Internet. Changing people’s clothes or manipulating the scene view are just two examples. In Photoshop, they are activated via the options in the context-sensitive taskbar. This works after selecting the entire image or an area. In contrast to Firefly, Photoshop only produces one result with the two new models.
Apart from the familiar use cases, the AI provides support when working on projects in the areas of graphic design, matte paint or concept art.

Transforming scenes in Photoshop

The use of stock photos in productions is a normal process. The same motif at a different time of year or day is usually not available. This is where Nano Banana from Google can help. It allows the scene to be quickly adapted to the production management’s specifications. This is achieved with a simple prompt. In the example, a night scene is created from an existing recording:

Turn in a night scene

The method allows the uncomplicated creation of material for virtual TV sets or documentaries.

Transforming Text in Photoshop

“Flux.kontext Pro” is known for changing text in photos without manipulating the font or changing the character of the overall impression. The artist then corrects errors in product designs with just a few mouse clicks. A time-consuming subsequent photo shoot is not necessary. Alternatively, product names with different international designations can be created for motion graphics. These can be used flexibly later in production.

Both steps described with light and fonts can be combined. In practice, it has been shown that changes to the text with the Black Forrest AI should be performed in the first step. Then the lighting of the scene is changed with Google Nano Banana. The results are usually more realistic in this order and errors are avoided.

Photoshop-AI: Quick merging of motifs and objects

Matte painters and concept artists save a lot of time by using AI. Buildings can be inserted into landscapes or objects integrated into existing scenes in no time at all using text input – a single prompt is often enough.
However, the results are not always satisfactory, as the nano-banana model often automatically generates architectural suggestions. Significantly better results are achieved if you use a reference image. This is loaded into a new layer while the selection tool is used to define the target area in the background. If the background layer is active, it is then inserted via the context bar with the prompt:

Place the castle from Layer [Layer name]

If the first attempt fails, you can modify the prompt and possibly add further instructions such as “set on the mountain top”. The AI will then position the object better.

Inline prompts are available when inserting with Nano Banana. These are text layers with the instructions for working with Nano Banana. They are supplemented by graphics such as arrows and frames in a separate layer. These elements help with the exact placement of the object. Only one instruction is then required in the prompt for generative filling:

Complete the prompt in red

In some cases, Nano Banana inserts the prompt text and the corresponding graphic into the result. Then it helps to extend the instruction with “remove the red prompts and elements”.

Context-related fill and replace with Firefly

In Photoshop Beta 27.1.0, Adobe has added the new Firefly model “Fill and Replace (Beta)”. A reference image can be loaded and placed via the context-sensitive taskbar. The function can remove the background in the motif if desired. Free-standing objects with clear edges are ideal for insertion. A clear background usually ensures good results for architectural motifs. Transitions on the ground often need to be reworked, which is easy to do with the existing mask. Any remaining errors can then be corrected with a brush.

Conclusion

Numerous image manipulations are possible with the new AI models in Photoshop Beta. However, the results are not always optimal. In these cases, it helps to work in iterations and approach the final result step by step. This method leads to success in most cases. A reminder: depending on your subscription, it can become costly quickly.

The new technology is a useful tool for quick results. However, Photoshop-Ai cannot replace creativity or the experience of the artist. Rather, it helps to build up a project and can make the work process more efficient.

The post Photoshop AI in practice first appeared on DIGITAL PRODUCTION and was written by Jörn-Erik Burkert.

Boris FX Continuum 2026 continues the developer’s long-running effort to automate routine postproduction work through machine learning. The 2026 release strengthens AI masking, compositing, and keying features while speeding up overall processing across multiple effects.

Masking and isolation

The PixelChooser, Continuum’s core masking and isolation framework, has been overhauled with AI segmentation tools. The new Face ML model automatically detects and isolates facial regions such as eyes, lips, teeth, and surrounding skin. By defining specific facial areas, users can target effects precisely or exclude them entirely, reducing the need for manual rotoscoping.

Complementing this, Matte Refine ML analyses existing mattes to restore or enhance fine edge detail. Elements such as hair strands and semi-transparent surfaces are now easier to preserve in composites. The PixelChooser also gains six additional matte and mask modes—Pre, Post, Both, Stacked Composite, Alpha Masked, and Alpha Composite—offering flexible control over whether masks are applied before, after, or alongside effects. Both Face ML and Matte Refine ML are available inside Beauty Studio, within the PixelChooser itself, and as standalone filters in the Key & Blend category.

Smarter face detection

Face Detection via Mocha adds automatic recognition and tracking of multiple faces within a clip. The feature integrates Mocha’s planar tracking, allowing individual faces to be followed accurately across frames. Once detected, Continuum effects can be selectively applied to specific tracked regions, streamlining retouching or compositing tasks that previously required separate mask layers.

Updated Primatte Studio

The Primatte Studio keying module has been rebuilt for Continuum 2026. According to Boris FX, it now employs a new 3D keying algorithm combined with integrated AI Denoise for cleaner source plates. Expanded Light Wrap and Spill Suppression controls offer greater realism when blending foregrounds into new backgrounds. The new Colour Link function automatically adjusts the hue and luminance of the keyed subject to match the underlying background input, reducing manual colour correction work. The redesigned custom toolbar centralises keying parameters, creating a more efficient, single-pane workspace for editors and compositors.

A persponal Note: We do love Ben’s tutorals but for this one: Turn on the automatic german translation, and enjoy a very hearty giggle – the confidences of the synthetic voice (that sounds nothing like Ben) talks about primates is… top-not unintentional comedy!

AI-powered transitions

Two new machine-learning transitions extend Continuum’s effects library. Jump Cut Fixer ML uses AI-based optical-flow analysis to generate seamless dissolves between two adjacent shots. By reconstructing motion between frames, the tool can mask timing gaps or abrupt cuts, preserving narrative flow without additional editing. Built-in colour adjustment parameters and interactive on-screen widgets enable further control over the dissolve timing and strength.

Depth Wipe ML, the second new transition, uses AI-generated depth maps to move beyond flat 2D wipes. Instead of a simple horizontal or vertical reveal, the effect creates a transition that passes dynamically through the 3D space of the image, wiping from foreground to background. The effect includes more than fifteen presets for rapid setup and provides controls for animation, blending, and dissolve shaping.

Enhanced legacy tools

Continuum 2026 revisits several of its longstanding effects. The Scanline effect recreates the appearance of vintage CRT displays, now adding adjustable parameters for glow, noise, and vignette intensity to deliver more authentic analogue texture. Lens Blur expands to support full HDR workflows and introduces a Depth Mask ML option that allows defocus or rack-focus effects to respond accurately to depth information.

The Colour Link filter, also available as a standalone effect, gains a Diffuse Colour Displacement texture option that improves colour blending and realism when integrating keyed subjects into complex backgrounds. Particle Illusion benefits from smoother animation interpolation, a reverse-animation toggle, and a newly curated emitter library designed for modern compositing environments. Meanwhile, Lens Flare has been optimised for performance and now renders up to four times faster than before. Title Studio improves type handling with expanded font-family support and more stable character rendering across different hosts.

Faster AI models

Continuum’s existing AI-based effects have been refactored for speed and quality. Retimer ML can now accommodate mixed frame-rate sources within After Effects, avoiding frame mismatch issues in complex sequences. UpRes ML incorporates NVIDIA Super Scale models, offering higher-quality image enlargements for up-conversion tasks. Depth Map ML produces cleaner, more consistent mattes suitable for grading or selective blurring operations. Motion Blur ML introduces adaptive sampling and synchronisation with Beat Reactor, enabling rhythm-driven motion-blur intensities. Frame Fixer ML receives refinements for handling trimmed or incomplete frames in Avid Media Composer projects, improving overall frame reconstruction reliability.

Continuum honoured with an Engineering Emmy

In October 2025, the Television Academy presented Boris FX Continuum with an Engineering, Science & Technology Emmy Award. Founder and CEO Boris Yamnitsky accepted the award together with long-time team members Peter McAuley, Jason Clement, and Mike Escola. The accolade recognises Continuum’s three-decade contribution to visual-effects and compositing workflows, marking it as one of the industry’s most influential non-linear-editing plug-in suites.

Yamnitsky described the award as both humbling and affirming, recalling the product’s beginnings as code written in his parents’ basement and its early adoption within emerging NLE systems such as Media 100 and Adobe Premiere. The Emmy recognition also highlights Boris FX’s commitment to broad platform support. Continuum has long maintained compatibility across Avid, Premiere Pro, DaVinci Resolve, Vegas, Filmora, and Final Cut Pro, reflecting the company’s belief that every editor deserves access to the same toolset regardless of host application.

For the record

Boris FX presents the 2026 release as a continuation of this Emmy-recognised legacy, extending Continuum’s compositing foundation with new AI-driven technologies. As always, postproduction teams are advised to test AI-masking accuracy and stability within their own workflows before integrating the update into production pipelines.

borisfx.com

The post Continuum 2026: AI masks, fixes, and faces first appeared on DIGITAL PRODUCTION and was written by Bela Beier.