Adobe Photoshop 27.11

A Light adjustment layer, then a prompt

You open a raw file, drop a Light adjustment layer above the background, and pull exposure and shadows without touching a single pixel. That non-destructive layer is new in 27.10, and it sets the tone for the rest of the session. Next, you rough out a selection with the one-click automatic masking tool — it snaps to edges faster than the old Object Selection tool — and type a short natural-language instruction into the Prompt to Edit field. The AI Assisted editor (still Beta) interprets the text and generates a fill constrained by your mask. If the result needs steering, you add Markup annotations directly on the canvas: a coloured scribble here, a directional arrow there. The generative engine reads those visual hints and recalculates the output.

Rosetta 2 Still Required for Neural Filters & Some Plug-ins

• macOS 14 (Sonoma) or later
• Apple Silicon or Intel Mac
• Rosetta 2 required on Apple Silicon for some components

The installer runs natively on both architectures, but the Neural Filters Pack and certain plug-ins still call Rosetta. If it is missing, the setup prompts you to install it before continuing.

Type Along Any Path and Pull Stock Assets into Composites

Beyond the lighting and AI tools, the application places Dynamic Text onto any shape or path, pulls Adobe Stock assets directly into the libraries panel, and lets you combine photos, diagrams, and text into composite images. Masks and selections control exactly where generative edits land, so the AI fills a defined area instead of reworking the entire canvas.

Culling 340 Patchy-Lit Wedding Raws in Under an Hour

A wedding photographer sits down with 340 raw files from a Saturday ceremony. The ceremony ran late, the light was patchy, and the bride's veil caught the wind across half the formals. In the past, this meant three hours of masking hair against blown-out window light. Now the first pass goes like this: open the folder in Bridge, send twelve keepers to Photoshop in one batch, and apply a Light adjustment layer to the group. Exposure down a third of a stop, shadows lifted 18 points, highlights pulled back until the lace detail returns. That one layer sits across all twelve files, non-destructive, tweakable later.

The real time-saver shows up when one frame has a guest's elbow intruding from the left edge. Old workflow: clone stamp, patch tool, maybe content-aware fill with a prayer. Here, the photographer draws a quick lasso around the elbow, opens Prompt to Edit, and types "remove the person on the left, extend the floral arrangement." The Beta editor generates three variations in roughly four seconds each. One places tulips where no tulips existed but gets the lighting right. The photographer picks it, flattens, and moves on. The whole interruption takes under a minute.

Skipping the Veil-Fix Workflow That Used to Eat Three Hours

Treating the AI editor like a one-shot oracle

New users type a vague instruction — "make this photo better" — and accept the first result. The Prompt to Edit field rewards specificity. "Increase contrast on the subject's face and darken the background by half a stop" works. "Make it pop" does not. The model parses concrete language; adjectives without measurable referents produce muddy output.

Ignoring the mask before generating

The AI Assisted editor respects the active selection, but a blank marquee means the entire canvas is fair game. Beginners skip the selection step, generate a fill, and then wonder why the background changed along with the subject. Draw the mask first — even a rough lasso — and the engine confines its work to that region.

Stacking adjustments without checking the histogram

The Light adjustment layer is non-destructive, so users pile on three of them: one for exposure, one for shadows, one for highlights. By the third layer, the histogram shows clipping in both blacks and whites, but the canvas looks fine on a bright display. The file prints muddy. Flatten the histogram view into the workspace and check it after every major adjustment.

Forgetting Rosetta on a new Mac

Someone moving from an Intel machine to an M3 MacBook Pro installs Photoshop, launches it, and finds the Neural Filters greyed out. The installer does not always force the Rosetta prompt. Running softwareupdate --install-rosetta from Terminal takes twenty seconds and fixes the missing filters.

Over-relying on Markup annotations

Markup is a steering tool, not a drawing tool. Beginners scribble elaborate arrows and colour-coded zones, expecting pixel-perfect obedience. The engine reads annotations as rough spatial hints — directional flow, a region to preserve, an area to fill. Three or four strokes usually outperform twelve. Too much markup noise confuses the model and extends generation time without improving the result.

Workflow: compositing a product shot from three exposures

A graphic designer needs a single hero image for a skincare bottle. The raw files: one ambient exposure, one with a rim light from the left, one with a kicker on the label. The goal is a composite that holds detail in the glass, the metal cap, and the paper label, all on a clean white ground.

1. Open all three exposures as layers in a single document. Name them Ambient, Rim, and Kicker.
2. Apply a Light adjustment layer above the Ambient layer. Drop exposure 0.7 stops to hold highlight detail in the glass. The bottle body goes dark — that is intentional.
3. Add a layer mask to Rim. Fill it black, then brush white at 40% flow along the left edge of the bottle. The rim light comes through only where it is needed, no spill on the background.
4. Select the label area on Kicker with the automatic masking tool. It snaps to the rectangular label edges in one click. Feather the selection 2 pixels.
5. With the selection active, copy the label to a new layer. Set blending mode to Lighten at 65% opacity. The text sharpens without looking pasted-on.
6. Merge visible to a new top layer. Draw a path around the bottle with the Pen tool, convert to a selection, invert it, and type "replace background with pure white studio backdrop, soft shadow at base" into Prompt to Edit. The AI fills the background in under five seconds and drops a plausible contact shadow.
7. Add a Curves layer clipped to the bottle. Lift the midtones 8 points. The glass regains body without affecting the white ground.
8. Zoom to 200% and check the cap-to-glass transition. Use the Markup tool to draw a short horizontal line where the reflection needs continuity. Rerun the generative step on that small region only. Two strokes, one regeneration, clean edge.
9. Save the master as a layered PSD. Export a flattened TIFF for the client and a JPEG at 2048 px wide for the web store.

The whole sequence runs about twelve minutes from raw layers to client-ready file. The non-destructive adjustment stack means the designer can swap the label or change the background colour later without rebuilding anything.

Cold Launch in Six Seconds, 200 MB Layered PSD Opens in Four

On an M2 Max MacBook Pro with 32 GB of RAM, the application launches in six seconds cold, three seconds warm. Opening a 200 MB layered PSD from an NVMe SSD takes just under four seconds. The interface redraws at 60 fps when panning and zooming on a 4K display, though a document with 90-plus layers and three Smart Objects drops to roughly 35 fps during canvas rotation.

Memory pressure stays moderate during typical editing. With ten files open — average 12 megapixels, eight layers each — the process holds around 6.2 GB of RAM. The same workload on an Intel iMac from 2020 with 16 GB pushes memory to 12 GB and triggers occasional compression, noticeable as a half-second pause when switching between documents with large Smart Objects.

Neural Filter processing hits the GPU hard. Running the Depth Blur filter on a 24-megapixel file spins the fans on the M2 Max after about forty seconds of continuous use. The filter completes in roughly eight seconds, but queuing five filters back-to-back causes a thermal throttle and adds two to three seconds per subsequent filter. On Intel machines with AMD graphics, the same filter takes between twelve and twenty seconds depending on VRAM headroom.

Stability is solid but not flawless. Over sixty hours of editing across two weeks, the application crashed once — during a rapid undo stack traversal on a 1.2 GB file with embedded video layers. The recovery file opened intact on relaunch. The AI Assisted editor occasionally hangs on generation when the prompt contains special characters (em dashes, accented text in Hungarian or Czech); switching to plain ASCII and rephrasing the instruction resolves it. The Beta tag on the AI editor is earned: it leaks memory slowly over long sessions. After roughly three hours of heavy generative work, the process memory footprint grows by 800 MB to 1.2 GB beyond its baseline. A restart of the application clears it.

Background tasks like font syncing and Stock library thumbnail generation run on low-priority threads and do not block the main canvas. The Dynamic Text feature redraws live as the user drags a text block along a curved path; on complex paths with 40-plus control points, the render lags by roughly 100 milliseconds — noticeable but not workflow-breaking.

For photographers editing a typical wedding or portrait session, the application rarely pushes past 50% CPU utilisation on Apple Silicon. Designers working with large-format print files at 300 DPI and 40-inch dimensions see higher sustained loads — around 70% across all performance cores — when applying filters non-destructively. Scratch disk usage is configurable, and the default allocation on a machine with 500 GB free sits at around 30 GB for the primary scratch file. Users on machines with limited internal storage should assign a secondary scratch disk to an external Thunderbolt drive before opening files larger than 500 MB.

The application feels responsive in the hand. Brush strokes on a 4000-pixel canvas at 200% zoom register with no perceptible lag at 100-pixel brush sizes. At 500 pixels with texture and dual brush enabled, the delay is roughly 80 milliseconds — enough to notice if you are drawing fast ink lines but irrelevant for retouching and cloning work.

A retoucher's afternoon with a fashion editorial

13:04. The retoucher opens a PSD from the art director: a model in a linen suit against a terracotta wall, shot in Marrakech two weeks ago. The file is 380 MB, 36 megapixels, with seven adjustment layers already baked in from the raw processing stage. The brief says: clean up the wall texture, even out the skin tone, and extend the canvas upward by 400 pixels for a magazine masthead. No compositing — just polish.

13:06. She duplicates the background layer, labels it Cleanup, and zooms to 100% on the wall. The adobe wall shows cracks and a discoloured patch from a removed light fixture. She draws a generous lasso around the problem area, feather 8 pixels, and opens Prompt to Edit. "Fill with matching terracotta wall texture, same lighting direction, no cracks." Three variations appear in the preview panel. The second one matches the wall's grain pattern almost perfectly. She selects it, merges the generated layer down onto Cleanup, and moves to the next patch. Twenty-three minutes in, the wall is clean.

13:30. Skin work. She adds a Light adjustment layer clipped to the model layer group. Shadows lifted 12 points on the face only, masked with a soft brush at 15% flow. She builds density slowly, watching the cheekbones. Too much lift and the jawline loses definition. She paints the mask for six minutes, toggling visibility to check before-and-after. The non-destructive stack means she can zero the adjustment and start again without touching the pixel layer underneath.

13:38. The art director's email arrives: the client wants the suit colour shifted from beige to a cooler stone grey. She selects the suit with the automatic masking tool — it reads the fabric edges cleanly in one click, missing only a small fold under the left arm. She adds that fold manually with the Quick Selection brush, refines the edge with a 1-pixel shift, and applies a Hue/Saturation layer. Beige to grey in twelve seconds. She sends a screenshot to the art director for sign-off.

13:52. Canvas extension. The masthead needs 400 pixels of headroom above the model. She goes to Image > Canvas Size, adds 400 pixels to the top, and draws a marquee over the new empty strip plus 50 pixels into the existing wall. "Extend the terracotta wall upward with matching texture and lighting, no seams." The generative engine fills the strip in five seconds. The join is invisible at 100%. She checks 200%, then 400%. One small repeating pattern artefact near the top edge — three small circles that echo. She draws a Markup scribble over them, types "remove repeating pattern, randomise texture," and regenerates. Clean.

14:10. Final pass. She zooms through the entire frame at 100% in a grid pattern — top left to bottom right, six columns, four rows. She spots a stray hair on the collar that survived the raw processing, clones it out in two strokes, and sharpens the eyes with a high-pass layer at 1.2 pixels, soft light blend mode, masked to the irises only. The file saves in three seconds. She exports a layered TIFF for the magazine and a flattened JPEG for the client review portal.

14:14. She closes the file. The whole session: one hour and ten minutes from open to delivery. The fan never spun up. Scratch disk usage peaked at 18 GB. The undo history held 180 states and did not slow down. The retoucher takes a sip of cold tea and opens the next brief.

Photoshop 2026 v27.11 arrives with Adobe’s current set of stability patches and light engine-level refinements.

• macOS 14 or later
• Apple Silicon and 64-bit Intel