Product

The full editing surface.

What Editmamei does, broken out: the pillars of the toolkit, every capability category, and prompts that work today. For the elevator pitch, see the home page.

What it is

Photoshop, but the AI does the menus.

Editmamei gives your AI assistant a working photographer's toolkit inside Photoshop: documents, layers, selections, adjustments, filters. Your AI calls them as building blocks in service of what you actually want done, instead of you hunting through menus.

1

The Photoshop you have, driven by language

Editmamei drives the real Adobe Photoshop you already installed: the full program, not a hosted web app or a cloud copy. Your AI talks to the desktop, and your project files live there.

2

Non-destructive by default

Adjustment layers, masks, groups. Editmamei builds the kind of layer stack a working editor builds. Everything stays editable, maskable, removable. Nothing bakes into pixels unless you ask.

3

Recipes that reproduce

A template is a reproducible aesthetic recipe. Apply it later to a different image and the AI reads the recipe’s reasoning to recreate the look on the new file. Editing decisions stop being one-shots. The whole template system, authoring and applying alike, is a Pro feature.

What it does

The full editing surface, exposed as tools.

Eight categories, each shown as one real Photoshop tool run on the same photo, so the AI plans accurately and you can audit what it ran. Hover or tap a result to see the original.

Documents — source photo opened in Photoshop
In HEICRAWPSD Out PSDJPEGPNG

Documents

Open PSD, JPEG, PNG, TIFF, DNG, HEIC, and the standard raw formats; save layered PSDs; export JPEG and PNG. Camera metadata (make, model, lens, ISO, focal length, GPS) and ACR develop settings surface to the AI before it edits.

Layers — document with a layer stack
Layers
  • Saturation +42
  • Deepen sky
  • Background blur
  • Background

Layers

Create, duplicate, delete, rename, reorder, group, merge, flatten. Set opacity, blend mode, visibility, locking. The complete layer tree returns as JSON, so the AI always knows the document structure. Move, scale, rotate, and fit-to-document transforms are included too.

Smart selections — Select Subject marquee on the subject Select Subject

Smart selections

Magic Wand, plus rectangle and feather. Every selection returns area, edge complexity, and pixel counts, so the AI can verify before committing to a mask or adjustment. Pro adds Sensei-backed Select Subject and Select Sky.

Non-destructive adjustments — original photo Non-destructive adjustments — Hue / Saturation applied
Hue / Saturation

Non-destructive adjustments

Curves, Levels, Hue/Saturation, Brightness/Contrast as adjustment layers: editable, maskable, removable. An active selection at call time becomes the new layer’s mask automatically.

Filters & styles — original photo Filters & styles — Gaussian Blur applied
Gaussian Blur

Filters & styles

Gaussian Blur, Motion Blur, Sharpen, Smart Sharpen, Reduce Noise, High Pass, Shadows/Highlights, Add Noise. Drop shadow, stroke, outer glow as full layer styles. Auto-rasterization handles text and Smart Object inputs cleanly.

Masks — original photo Masks — Select Sky → mask applied
Select Sky → mask
Masks — black and white layer mask
Layer mask

Masks

Layer masks across the lifecycle: create from the active selection (or reveal-all), apply, delete. Discrete tools, predictable behavior.

Templates — original photo Templates — Saved recipe applied
Saved recipe

Templates

A reproducible aesthetic recipe: capture a finished edit, then apply it to new images later, where the AI re-derives each value for the new photo and self-judges against the recipe's exit criteria. The whole template system (create, save, apply, verify, recall) is a Pro feature.

Histogram RGB · per-channel
Red Green Blue
Per-channel histogram of the City-Street photograph, read back from the document so the AI can verify tonal range quantitatively.

Visual verification

Downscaled JPEG previews return inline so the AI sees what the document actually looks like and confirms operations actually changed pixels instead of trusting a success message. 256-bin per-channel histograms with mean / stdev / median back that up quantitatively.

What you say to it

Three prompts that work today.

The AI reasons about intent. Editmamei handles the choreography.

  1. Example 1 · Landscape grade in one sentence

    "Open E:\Photos\beach.jpg. Build a non-destructive editing stack with Curves and Hue/Saturation adjustment layers, group them as 'grade', and warm the midtones slightly. Save the layered PSD next to the original and export a 2400px sRGB JPEG."

    What happens

    Roughly ten distinct tool calls. Each verifiable, each undoable. The AI reasons about intent; Editmamei handles the Photoshop choreography.

  2. Example 2 · Portrait retouch with feedback (Pro)

    "Open this portrait. Use Select Subject to isolate the person, feather the selection 2 pixels, and add a Curves adjustment layer clipped to that selection that gently warms the skin tones. Show me the before and after."

    What happens

    The AI can call photoshop_get_preview at any step to see what the document looks like and adjust. Selection feedback tells it whether Select Subject actually worked or needs refinement. Select Subject is a Pro tool; Community covers Magic Wand, rectangle, and feather selections.

  3. Example 3 · Batch processing with a template (Pro)

    "Apply my 'warm coastal' template to every image in E:\Photos\shells-raw\, exporting flattened JPEGs to E:\Photos\shells-web\ at 2000px square."

    What happens

    Templates are a Pro feature. The recipe captures a complete editing approach, and the AI works through the folder image by image, re-deriving each value for the photo in front of it. Templates are how editing decisions become repeatable instead of one-shots.

One photo or a hundred?

Pick the AI client to fit the sitting.

Editmamei works the same in every supported AI client. The client you choose changes how long a sitting lasts before things start to feel slow.

One photo, one sitting

Claude Desktop

The easiest setup and the most familiar if you've used Claude on the web. A great fit for a hero shot to grade, a portrait to retouch, a quick template to apply. Open the app, ask for the edit, watch Photoshop do it.

Cursor users: the experience is the same. Use whichever is already in your workflow.

A whole set, repeating templates

Claude Code

A larger working memory for the conversation keeps it fast across hundreds of edits in one sitting. Real-estate batches, wedding selects, multi-image template authoring. A one-time terminal install gets it going; after that it works the same way Desktop does.

If you find a Desktop session getting sluggish in the middle of a batch, this is the fix.

Not sure? Start in Desktop. The first time you feel a session getting slow, switch to Code. The FAQ and the installation guide cover both clients in detail.