source-brand-assets

name

source-brand-assets

description

scrape a company's visual brand assets (logos + hero/product reference photos) from the public web into a structured `brand-assets/` folder. Use when starting creative work for a new brand and you need real reference imagery before generating product photos, end cards, or ad mocks. Provider-agnostic; uses web search + curl. Not for collecting written brand guidelines — use `source-brand-guidelines` for that.

source-brand-assets

Purpose

Bootstrap a creative project for any company by collecting real reference photographs of their product(s) and their official logo into a structured brand-assets/ folder. Downstream skills like create-product-images-higgsfield-product-photoshoot and create-product-images-nanobanana take these references as input and produce on-brand generated stills.

Captures the exact reference set required to keep generated imagery faithful to a real product (e.g. Amex Platinum metal card, a specific SKU, a packaged consumer good, a software UI). Without grounded references, downstream image generation drifts off-brand.

Inputs

• Company name (required) — e.g. American Express, Liquid Death, Nothing, Peloton.
• Product or campaign focus (required when the company has multiple SKUs) — e.g. Platinum card, Mountain Water sparkling, Phone (2). If omitted and the company has a flagship product, infer from the most prominent product on the homepage.
• Output directory (default: <project>/brand-assets/) — the skill writes into card-photos/ (or <product>-photos/) and logos/ subdirectories under this path.
• Minimum asset count (default: 3 product photos, 1 logo SVG + 1 high-res PNG). Aim for 4–6 product photos when available — one clean front-on hero plus 2–3 lifestyle/contextual shots.

Workflow

1. Create the folder structure

mkdir -p <out>/<product>-photos <out>/logos

Use card-photos/, bottle-photos/, device-photos/, etc. — whatever names the actual product type. For broad brand asset collection without a single SKU, use product-photos/.

2. Find logo files

The most reliable public source is Wikimedia Commons. Search:

"<Company> logo" site:commons.wikimedia.org

Look for the official 2018+ wordmark file. Fetch the file description page to extract both the SVG upload.wikimedia.org/wikipedia/commons/... URL and the highest-resolution PNG render (/thumb/.../<width>px-....svg.png, usually up to 3840px).

Fallbacks if Wikimedia is thin:

• brandfetch.com/<domain> — has SVG/PNG variants for most brands.
• logo.wine — clean SVG/PNG downloads.
• The company's own /press, /brand, or /newsroom page — these often link a brand-assets zip.

Download with curl -sSL -A "Mozilla/5.0" -o <name>.svg <url> — many sites 403 on default curl UA. Never scrape from sites that explicitly forbid asset redistribution; surface a note in the output README pointing to the official press kit instead.

3. Find product / hero photographs

The official press kit is the gold standard but most major brand newsroom pages 403 WebFetch requests. Plan B sources, in order of typical reliability:

1. Wikimedia Commons — search Category:<Company> and any product sub-categories (e.g. Category:Credit cards). Direct upload.wikimedia.org URLs are CDN-fetchable.
2. Specialist editorial — for finance, The Points Guy, NerdWallet, CNBC Select; for consumer tech, The Verge, MKBHD; for FMCG, Trendhunter, Dieline; for fashion, Hypebeast, Highsnobiety. These embed press kit images at their own CDN with no auth.
3. Image stock aggregators (last resort, watermark risk) — Shutterstock/Getty/Alamy/iStock surface preview URLs you can sometimes lift, but they will be small and watermarked. Use only for layout reference, never for downstream generation input.
4. Official press CDN by guess — many brands run their press images on runway-media-production.global.ssl.fastly.net, s7d2.scene7.com, or similar Scene7/Akamai endpoints discoverable in editorial article HTML.

Search pattern that works:

"<Company> <Product>" official product image press release
"<Company> <Product>" press kit hi-res
"<Company> <Product>" site:commons.wikimedia.org

For each candidate URL: curl -sSL -A "Mozilla/5.0" -o <name>.<ext> <url> and then Read the file to verify it's actually the right product (not a competitor, not a similar SKU, not a watermarked thumbnail).

4. Verify

Open every downloaded image with the Read tool and confirm:

• The product is the exact SKU requested (e.g. silver Platinum, not Centurion black; iPhone 15 Pro, not iPhone 15).
• No visible watermarks (Shutterstock, Getty, Alamy overlay).
• Resolution is usable: ≥ 1000px on the longest edge for hero references, ≥ 600px for lifestyle.
• For logos, both vector (SVG) and a large raster (PNG ≥ 2048px) are present.

Delete and replace anything that fails. Never ship a brand asset folder without visual verification — wrong-SKU references silently poison every downstream generation.

5. Write <out>/README.md

One section per subfolder listing every file with:

• Source URL (so the asset is reproducible/auditable)
• One-line description of what it depicts
• Era/edition note when relevant (e.g. "2025 mirror edition", "post-Sept refresh")

Include a usage-rights footer pointing to the official press kit URL — these scraped assets are for creative development, not redistribution.

Output

<out>/
  README.md                       # index + provenance + rights note
  logos/
    <company>-logo.svg            # current wordmark
    <company>-logo-<width>.png    # ≥ 2048px PNG render
    <company>-logo-old.svg        # optional legacy mark
  <product>-photos/
    <product>-<source>-<note>.png # 4–6 verified product photos
    ...

Plus, if running under test conventions, a manifest.json listing every file, its source URL, resolution, file size, and verification status (verified | watermarked | rejected).

Quality Checks

• Every product photo shows the same SKU as requested.
• No watermarks on any retained file.
• At least one clean hero product shot (subject isolated, neutral or product-press background) suitable as a primary reference image for downstream generation.
• At least one lifestyle/contextual shot showing the product in real-world use.
• Logo present in both SVG and ≥ 2048px PNG.
• README.md exists and lists every file with its source URL.
• No file > 20 MB without explicit reason (raw press TIFFs can be re-encoded to PNG).

Failure Modes

• Brand press site 403s WebFetch — common for Amex, Nike, Apple. Fall back to Wikimedia + specialist editorial CDNs. Do not waste cycles retrying the official site.
• Wikimedia category has no product shots — only logos / building exteriors. Pivot to editorial-CDN scrapes (e.g. runway-media-production...fastly.net/...amex-plat-mirror-consumer.jpg).
• All candidates are watermarked stock previews — surface the gap explicitly in README.md and recommend a manual press-kit request from the brand.
• Wrong SKU collected — easy to grab Centurion when asked for Platinum, or iPhone Pro Max when asked for Pro. Visual verification step is non-optional; do not skip even when filenames look right.
• Mirror / limited edition shipped as the default reference — flag the era explicitly in the README so downstream generation doesn't accidentally use a one-off design as the canonical product.
• No vector logo available — accept high-res PNG and note in README that an SVG is missing; do not try to vectorise a raster.

References

Workflow battle-tested on:

• Amex Platinum (Sept 2025 refresh) — Wikimedia commons had logo but no card photo; press kit 403'd; recovered hero shots from Upgraded Points + TPG + runway-media CDN.
• Coca-Cola, Peloton, Gooseworks ad runs in this repo (see sibling project folders).

Pair with:

• source-brand-guidelines — written voice/visual rules
• understand-brand-assets — turn collected assets into a reusable brand brief
• create-product-images-higgsfield-product-photoshoot — hero/end-card generation from these references
• create-product-images-nanobanana — lifestyle/everyday-setting generation from these references