agentleFS
Sign inStart free
aiagentwithdhruvaiagentwithdhruv/skills/recreate-thumbnails/SKILL.md

recreate-thumbnails

Skill16 starsChanged 9 months agoMITView on GitHub

Face-swap YouTube thumbnails to feature Your Name using AI. Use when user asks to recreate thumbnails, face swap images, generate YouTube thumbnails, or create thumbnail variations.

  • Reads credentials

3 other projects use a version of this, 3 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

1 rule. 1 is common, 0 are distinctive, and 0 are very specific to this project.

  • 1

    Common

    Most of the 3 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Common:

    Must explicitly request 16:9 format in prompt

    line 102

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Recreate YouTube Thumbnails

Goal

Face-swap YouTube thumbnails to feature Your Name using Gemini image model. The system analyzes face direction, matches reference photos by pose, and generates variations.

Scripts

  • ./scripts/recreate_thumbnails.py - Main generation script
  • ./scripts/analyze_face_directions.py - Reference photo analyzer

Quick Start

# From YouTube video (auto-downloads thumbnail)
python3 ./scripts/recreate_thumbnails.py --youtube "https://youtube.com/watch?v=VIDEO_ID"

# From local thumbnail
python3 ./scripts/recreate_thumbnails.py --source ".tmp/thumbnails/source.jpg"

# Edit pass on generated thumbnail
python3 ./scripts/recreate_thumbnails.py --edit ".tmp/thumbnails/recreated_v3.png" \
  --prompt "Change colors to teal. Change 'AI GOLD RUSH' to 'AGENTIC FLOWS'."

Full Workflow

Step 1: Build Reference Photo Bank (One-time)

# Drop 30-40 photos of Nick into raw folder
mkdir -p .tmp/reference_photos/raw

# Analyze and rename with face direction metadata
python3 ./scripts/analyze_face_directions.py

Creates files like:

  • nick_yawL30_pitchU10.jpg — looking 30° left, 10° up
  • nick_yawR45_pitch0.jpg — looking 45° right, level

Step 2: Generate Thumbnails

# From YouTube URL (analyzes face, finds best reference, generates 3 variations)
python3 ./scripts/recreate_thumbnails.py --youtube "VIDEO_URL"

# Custom variation count
python3 ./scripts/recreate_thumbnails.py --source "thumbnail.jpg" -n 5

# Skip direction matching
python3 ./scripts/recreate_thumbnails.py --source "thumbnail.jpg" --no-match

Step 3: Edit & Refine

# Single edit
python3 ./scripts/recreate_thumbnails.py --edit ".tmp/thumbnails/recreated_v3.png" \
  --prompt "Change colors to teal brand colors."

# Chain multiple edits
python3 ./scripts/recreate_thumbnails.py --edit ".tmp/thumbnails/edited_1.png" \
  --prompt "Make text bigger. Change background to white."

CLI Reference

FlagDescription
--youtube, -yYouTube video URL
--source, -sSource thumbnail path or URL
--edit, -eImage to edit (enables edit mode)
--prompt, -pEdit instructions (required for edit mode)
--variations, -nNumber of variations (default: 3)
--refsNumber of reference photos (default: 2)
--no-matchSkip face direction matching

Output Organization

.tmp/thumbnails/
├── 20251205/              # Date folder
│   ├── 104016_1.png       # Variation 1
│   ├── 104016_2.png       # Variation 2
│   ├── 104016_3.png       # Variation 3
│   └── 104532_edited.png  # Edit pass

API Details

  • Model: gemini-3-pro-image-preview (Nano Banana Pro)
  • Cost: ~$0.14-0.24 per generation/edit
  • Latency: 10-60+ seconds per image
  • Output: ~1376x768 (close to 16:9)

Learnings

  • 2 reference photos is optimal (1 loses likeness, 3+ causes regeneration)
  • Must explicitly request 16:9 format in prompt
  • Label images in prompt: "IMAGE 1: Reference, IMAGE 2: Thumbnail"
  • "100% exact duplicate except face" instruction works well
  • Edit passes work for text, colors, graphs, backgrounds

Environment

NANO_BANANA_API_KEY=your_key

Schema

Inputs

NameTypeRequiredDescription
youtube_urlstringNoYouTube video URL (auto-downloads thumbnail)
sourcefile_pathNoSource thumbnail image path
variationsintegerNoNumber of variations (default: 3)
edit_imagefile_pathNoImage to edit (enables edit mode)
promptstringNoEdit instructions for edit mode

Outputs

NameTypeDescription
thumbnailsarrayGenerated thumbnail file paths in .tmp/thumbnails/

Credentials

NameSource
NANO_BANANA_API_KEY.env

Composable With

Skills that chain well with this one: cross-niche-outliers, youtube-outliers

Cost

$0.14-0.24 per generation

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free