photo-rebrand

name

photo-rebrand

description

Use ONLY when a user needs to remove a competitor's watermark from a listing photo in a LEGITIMATE co-broker or authorized-resale arrangement (industry-standard practice in used equipment dealing where Broker A has the seller and Broker B has the buyer, photos shared between brokers as normal operations). Removes pixel-baked watermarks via OpenCV inpainting (Telea/NS algorithms), expanded region masking, and morphological refinement. REQUIRES explicit user confirmation of the co-broker basis before any modification. Every operation writes an audit log entry to .rebrand_log.jsonl with timestamp, photo hash, basis, and co-broker contact for paper-trail purposes. Best results when chained with photo-watermark to apply the user's own brand over the cleaned region. Triggers on phrases like "rebrand this listing photo", "remove the watermark", "co-broker rebrand", "swap the watermark", "this is a shared listing", or any photo with a visible competitor watermark in an authorized resale context.

Photo Rebrand (Co-Broker Edition)

Removes pixel-baked watermarks from listing photos in legitimate co-brokerage arrangements. Requires explicit user confirmation of authorization before any modification. Every operation is logged for audit purposes.

When this fires

• "Rebrand this listing photo"
• "Remove the watermark — this is a co-broker deal"
• "Swap the watermark on this photo"
• "I'm sharing this listing with [other broker], rebrand it"
• After the inspector flags a watermark and the user says "yes, that's authorized — remove it"

CRITICAL: confirmation flow

This skill modifies pixel-level branding that has legal weight. Before any rebrand:

Step 1 — Explain what's happening

"This will remove the [Machinery Network / source] watermark from the photo. Watermarks can carry legal weight (copyright notices, trademark claims), so I need to confirm this is an authorized rebrand before proceeding."

Step 2 — Get explicit confirmation of basis

Ask the user to confirm ONE of these bases for the rebrand:

Please confirm the basis for this rebrand:

☐ Co-broker arrangement — The original broker (named: ____) and I are co-listing this machine. Photos shared as part of the co-broker agreement.

☐ Authorized resale — I have explicit permission from the original photo owner to re-list this photo under my brand.

☐ Stock OEM photo — This was an OEM brochure photo that got watermarked downstream; the OEM grants downstream reseller use.

☐ Other (please specify): _______________

I also need the contact/name of the co-broker (for the audit log): _______________

Step 3 — Acknowledge and proceed

Only after the user provides:

• The basis (one of the checkboxes)
• The co-broker name (if applicable)

…call rebrand_photo() with confirmed=True.

Step 4 — Display the log entry

After rebrand, show the user the audit entry that was written:

✓ Rebrand complete.
  Timestamp: 2026-05-28 02:15:43
  Source: ynnjkrazj0jevf413gosylm8b4le.webp
  SHA256: 3f8a1c... (truncated)
  Basis: Co-broker arrangement
  Co-broker: Machinery Network
  Region: bottom-right
  Output: /mnt/user-data/outputs/photo_rebranded.jpg

  Log location: /mnt/user-data/uploads/.rebrand_log.jsonl

The technical flow

import sys
sys.path.insert(0, 'scripts')
from rebrand import rebrand_photo

# Typical use: chained after the inspector flagged a watermark
entry = rebrand_photo(
    input_path="/mnt/user-data/uploads/photo.jpg",
    output_path="/mnt/user-data/outputs/photo_rebranded.jpg",
    region=(y0, y1, x0, x1),   # from inspector's watermark_detector
    region_label="bottom-right",
    region_padding=0.5,         # expand 50% on each side for full coverage
    basis="Co-broker arrangement",
    co_broker="Machinery Network",
    mask_mode="auto",           # auto-detect bright vs dark watermark
    inpaint_radius=5,           # 3-7 range; bigger = more blending
    inpaint_method="telea",     # "telea" or "ns" — try both if unhappy
    dilate_px=6,                # mask growth for edge blending
    confirmed=True,             # REQUIRED — must be True
)

Quality expectations — be honest with the user

OpenCV inpainting is good but not perfect on real-world watermarks. Set expectations:

Watermark style	Background	Quality
Solid logo over plain area (sky, blank wall)	Plain	Excellent — invisible removal
Solid logo over busy content	Mixed	Good — minor blur in cleaned area
Translucent text over busy content	Hard	Visible smudge — needs cover
Tiled / full-image watermarks	N/A	Not removable with this tool

Standard practice: chain rebrand with photo-watermark to apply the user's own logo over the cleaned region. The user's logo hides any inpaint imperfection, and the end-state is what they wanted anyway — their brand on the listing photo.

# After rebrand, immediately invoke photo-watermark:
from watermark import apply_watermark

apply_watermark(
    input_path="/mnt/user-data/outputs/photo_rebranded.jpg",
    output_path="/mnt/user-data/outputs/photo_final.jpg",
    position="bottom-right",
    opacity=0.75,    # slightly higher than default to cover any residue
    scale=0.22,      # slightly bigger to dominate the cleaned area
    use_brand_cache=True,  # pulls from .imm_brand.json
)

Audit log

Every rebrand writes a JSON line to .rebrand_log.jsonl:

{
  "timestamp": "2026-05-28T02:15:43.123456",
  "operation": "watermark_removal",
  "source_photo": "/mnt/user-data/uploads/listing.webp",
  "source_sha256": "3f8a1c4e...",
  "output_photo": "/mnt/user-data/outputs/listing_rebranded.jpg",
  "basis": "Co-broker arrangement",
  "co_broker": "Machinery Network",
  "region": "bottom-right",
  "inpaint_radius_px": 5
}

The SHA-256 of the source file ties the log entry to the EXACT input photo — useful if anyone ever questions the basis. Same photo + different basis = two log lines, both recoverable.

Anti-patterns

• ❌ Removing a watermark without user confirmation. Always require the explicit confirmation step.
• ❌ Removing watermarks from photos with no claimed authorization. The skill is for co-broker / authorized cases only. If the user can't name a basis, decline.
• ❌ Removing without logging. Every operation gets a log entry, no exceptions.
• ❌ Promising perfect removal. Inpainting has visible artifacts on complex backgrounds. Tell the user up-front, plan to cover with their own watermark.
• ❌ Bypassing the confirmation by passing confirmed=True without asking. The flag exists to enforce the conversation, not to skip it.

Companion skills

• photo-metadata-inspector — runs FIRST to detect the watermark region and provide the slice coordinates this skill needs
• photo-watermark — runs AFTER to apply the user's own brand over the cleaned area

The natural three-skill chain: inspect → rebrand → watermark.

File map

• scripts/rebrand.py — inpainting engine + audit logger
• references/legal_context.md — explains the legitimate co-broker basis for educational purposes