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.pyCreates files like:
nick_yawL30_pitchU10.jpg— looking 30° left, 10° upnick_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-matchStep 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
| Flag | Description |
|---|---|
--youtube, -y | YouTube video URL |
--source, -s | Source thumbnail path or URL |
--edit, -e | Image to edit (enables edit mode) |
--prompt, -p | Edit instructions (required for edit mode) |
--variations, -n | Number of variations (default: 3) |
--refs | Number of reference photos (default: 2) |
--no-match | Skip 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 passAPI 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_keySchema
Inputs
| Name | Type | Required | Description |
|---|---|---|---|
youtube_url | string | No | YouTube video URL (auto-downloads thumbnail) |
source | file_path | No | Source thumbnail image path |
variations | integer | No | Number of variations (default: 3) |
edit_image | file_path | No | Image to edit (enables edit mode) |
prompt | string | No | Edit instructions for edit mode |
Outputs
| Name | Type | Description |
|---|---|---|
thumbnails | array | Generated thumbnail file paths in.tmp/thumbnails/ |
Credentials
| Name | Source |
|---|---|
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