Youtube branding

youtube-branding

The channel-identity lego-block of the marketing-guru family. Before you ideate and deploy videos (marketing-ideate → marketing-deploy), a channel needs a face: this skill turns a name + slogan + niche into a full, upload-ready brand kit.

Asset	Size	YouTube use
banner.png	2560×1440	Channel banner (art + exact wordmark in the safe area)
profile.png	1024×1024	Channel profile picture / avatar
logo.png	1024² transparent	Branding master (transparent PNG)
logo_512.png	512² transparent	Watermark to overlay ON videos
brand.md	—	Channel keywords + description (copy-paste)

Where it sits in the growth loop (see marketing-guru):

   [ youtube-branding ]  →  IDEATE → BACKLOG → DEPLOY → MEASURE → LEARN
      (channel setup,         (what to make & how it does, per channel)
       once / on rebrand)

How it works — division of labour

The mechanical pipeline lives in the studio package (studio/marketing/brand.py) and is driven by the studio brand <spec.json> CLI command. It:

• requests the art text-free from Nano Banana, then Pillow-overlays the wordmark in YouTube's safe area (never trust the image model to spell),
• cuts the logo to transparency with rembg, and cover-crops to upload sizes.

You (the agent) author the brand identity — palette, emblem concept, banner scene — into a brand-spec JSON. That creative judgement is the whole job; the CLI is just I/O.

Workflow

0. Setup (once)

cd /Users/dasein/dev/slope-studio
source .venv/bin/activate 2>/dev/null || { uv venv && source .venv/bin/activate && uv pip install -e ".[fal]"; }
grep -q FAL_KEY .env && echo "FAL_KEY ok" || echo "MISSING FAL_KEY — set it in .env (needed for real art)"

1. Gather inputs

You need: channel name, slogan, and the niche/aesthetic. If the aesthetic is unstated, infer one from the niche (don't over-ask) — pick a palette + a single repeatable motif/emblem that ties avatar↔logo.

2. Author a brand-spec JSON

Copy example-spec.json and rewrite every field. Rules that make the output good:

• style is appended to all three prompts and MUST end with no text, no letters, no words — the wordmark is added later by Pillow, so the art must be clean.
• logo_prompt → a flat, simple, iconic emblem on a plain solid background (rembg cuts that background out). Keep it legible at watermark size — few elements.
• profile_prompt → the same motif, centered, filling the square, readable when tiny.
• banner_prompt → wide 16:9 cinematic art with empty negative space across the centre (push detail to the left/right edges) so the wordmark has room.
• palette → primary (deep bg), accent (title glow + rule), highlight (slogan), text (title fill), each [r,g,b].
• name / slogan are the exact wordmark text; keywords / description are the channel copy. slug is the output folder under runs/_brand/.

3. (optional, free) Dry-run the wiring

# validates the spec + full Pillow/rembg pipeline with offline placeholder art, $0.
# ⚠ writes to runs/_brand/<slug>/ and OVERWRITES — set a THROWAWAY "slug" for the test.
studio brand myspec.json --provider stub

4. Generate the real kit (~$0.12)

studio brand myspec.json

Three Nano Banana stills @ $0.039 = ~$0.117. Output lands in runs/_brand/<slug>/.

5. Eyeball every asset and iterate

Always view banner.png, profile.png, logo.png — verify:

• banner wordmark is centered, legible, not colliding with busy art;
• logo cut is clean (no halo) and transparent (logo.png corner alpha = 0);
• the motif is the same emblem on logo and profile.

Re-roll a weak asset by tweaking its prompt in the spec and re-running (a full run re-rolls all three stills; accept the small re-spend, or dry-run with stub first).

6. Deliver

Report the asset paths, the keywords, and the description, and where each goes on YouTube (banner / picture / Settings→Basic info). Offer the video watermark overlay:

# overlay logo_512.png bottom-right at ~10% width on a finished video
ffmpeg -i in.mp4 -i runs/_brand/<slug>/logo_512.png -filter_complex \
  "[1]scale=iw*0.10:-1[wm];[0][wm]overlay=W-w-40:H-h-40" -c:a copy out.mp4

Notes & gotchas

• Never bake channel text into the AI image — image models misspell. name/slogan are always Pillow-overlaid; that's why style must forbid text.
• Custom banner/profile upload anytime; a custom video thumbnail (different thing, see film-maker studio thumbnail) needs a verified YouTube channel.
• Logo background removal is rembg (already a dep) — it cuts a clear foreground subject from a plain background, so author logo_prompt with a clean subject on a flat bg.
• Fonts auto-pick the boldest macOS face (Avenir Next Condensed / Futura / Helvetica), with graceful fallback.
• This makes static brand art, not video. Produce videos with film-maker; decide what to post and run the loop with marketing-guru.