photo-watermark

name

photo-watermark

description

Use whenever a user wants to add a visible watermark (logo image OR text) to a photo. Supports configurable position (9-grid keywords like 'bottom-right', 'center'), opacity, scale, and text color. Auto-pulls logo and company name from the IMM agent's brand cache at /mnt/user-data/uploads/.imm_brand.json if available, otherwise asks for a logo file or text string. Independent of the photo-metadata-inspector — does NOT touch metadata; if user wants both metadata removal AND watermarking, the inspector handles that flow. Triggers on phrases like "add my logo to this photo", "watermark this image", "put a copyright on this", "brand this photo", "add a watermark", "stamp this with my logo".

Photo Watermark

Adds a visible logo or text watermark to a photo. Standalone — works whether or not the photo has been scrubbed. Pulls brand identity from the shared cache so you usually don't have to specify anything.

When this fires

• "Add my logo to this photo"
• "Watermark this image"
• "Put a copyright on this"
• "Brand this photo before I post it"
• "Stamp this with my logo"
• "Add 'Premier Equipment' to the corner"
• "Add a watermark to all these listing photos"

The flow

Step 1 — Resolve the watermark source

In priority order:

1. Inline logo provided in the user's message (uploaded file) → use that
2. Inline text provided ("watermark this with 'Sold by Premier'") → use that
3. Brand cache at /mnt/user-data/uploads/.imm_brand.json → use the logo from there
4. Brand cache has only a company name (no logo) → use text © <company_name>
5. Nothing available → ask the user:

   "What watermark would you like to apply? Drop a logo file (PNG/SVG) or tell me what text to add (e.g. '© Premier Equipment')."

Step 2 — Apply the watermark

import sys
sys.path.insert(0, 'scripts')
from watermark import apply_watermark

apply_watermark(
    input_path="/mnt/user-data/uploads/photo.jpg",
    output_path="/mnt/user-data/outputs/photo_watermarked.jpg",
    logo_path="/tmp/premier_logo.png",     # OR
    text="© Premier Equipment",            # OR both
    position="bottom-right",               # see grid below
    opacity=0.65,                          # 0.0–1.0; default 0.65
    scale=0.18,                            # logo width as fraction of photo width
    padding=24,                            # px from photo edge
    text_color="#FFFFFF",                  # any hex
    text_shadow=True,                      # legibility on busy backgrounds
)

Step 3 — Confirm placement

Show the watermarked photo. If it looks off, offer to redo with:

• A different position (top-left / top-center / top-right / middle-left / center / middle-right / bottom-left / bottom-center / bottom-right)
• Different opacity (more transparent if it dominates, more opaque if it disappears)
• Different scale (smaller if the logo overwhelms, larger if it's barely visible)

Position grid

top-left      top-center      top-right
middle-left   center          middle-right
bottom-left   bottom-center   bottom-right

Default: bottom-right. Most photography conventions place the watermark there — it's far from the focal point but always visible.

Recommended settings by photo type

Photo type	Position	Opacity	Scale
Equipment listing (product clearly framed)	bottom-right	0.65	0.18
Hero shot for marketing	bottom-right	0.55	0.20
Detail / close-up (limited corners)	top-right or center	0.45	0.15
Inspection / forensic photo	bottom-center	0.85	0.22
Social media (square crop)	bottom-right	0.70	0.16

Critical rules

• Does NOT touch metadata. If the user wants both metadata removal AND a watermark, route to photo-metadata-inspector which can call this skill as its option 2.
• Uses brand cache by default. Same cache as the IMM spec agent — /mnt/user-data/uploads/.imm_brand.json. Don't re-prompt for the logo if it's there.
• Inline overrides win. If the user pastes "use logo XYZ.png" or "watermark text 'For Sale'", that takes precedence over the cache.
• Always show the result. Don't just save the file silently — present_files() it and ask if placement looks right.
• One photo or many. If the user drops a folder or multiple photos, batch-watermark all of them with the same settings.

File map

• scripts/watermark.py — the engine (logo overlay + text overlay with full controls)

Anti-patterns

• ❌ Watermarking a photo that still has metadata baked in — direct them to the inspector first if privacy matters
• ❌ Picking a position that covers the main subject — always default bottom-right unless told otherwise
• ❌ 100% opacity (no transparency) — looks amateurish, suggests the watermark wasn't intentional
• ❌ Asking for the logo when the brand cache has one — silently use it; only ask if cache is empty