version: "1.0.0" name: bokeh-effect description: >- Apply synthetic depth-of-field bokeh to photos using MiDaS monocular depth estimation. Makes phone photos look like they were shot with a fast prime lens. No depth sensor required — works from a single image. metadata: { "openclaw": { "emoji": "📷", "requires": { "bins": ["uv"] }, }, }

portrait — Synthetic Bokeh via Depth Estimation

Estimates per-pixel depth from each image using Intel MiDaS, then applies variable lens blur: near objects stay sharp, far objects blur progressively.

The skill directory (where this SKILL.md lives) is referred to as $SKILL_DIR below.

When to Use

Use this skill when the user wants to:

• Add a portrait-mode / DSLR-style shallow depth-of-field effect
• Make phone photos look more professional
• Blur distracting backgrounds without manual masking

Setup (first run only)

cd "$SKILL_DIR" && uv sync

MiDaS weights are downloaded from torch.hub on first use (~400 MB for DPT_Large).

Agent Workflow

1. Ask the user

Before I process the images, I need to know:
 
🌀 Blur strength  (1–50, default: 15)
  Stronger = more pronounced bokeh.
  Subtle effect: 5–10  |  Natural: 12–18  |  Heavy: 20+
 
🧠 Model
  - DPT_Large   — most accurate depth, slower  (default)
  - DPT_Hybrid  — good balance
  - MiDaS_small — fastest, lower depth quality
 
⚙️  Device
  - auto   — GPU if available, else CPU (default)
  - cpu    — force CPU (~5s/image, fine for small batches)
 
📁 Input / output directories  (default: ./input and ./output)

2. Edit config.json

3. Run

cd "$SKILL_DIR" && uv run python scripts/bokeh.py --config config.json

4. Report results

Config Reference

Common Invocations

# Default settings
cd "$SKILL_DIR" && uv run python scripts/bokeh.py
 
# Stronger blur
cd "$SKILL_DIR" && uv run python scripts/bokeh.py --blur-strength 25
 
# Fast mode (lower quality depth)
cd "$SKILL_DIR" && uv run python scripts/bokeh.py --model MiDaS_small
 
# Force CPU
cd "$SKILL_DIR" && uv run python scripts/bokeh.py --device cpu

Output

Processed images written to output_dir with the same filenames as input.

Error Handling

• Missing input_dir → exits with clear message
• Individual image failures → logged; other images continue
• Depth model download failure → shows torch.hub error with fix suggestion