Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Enhance images and videos using HitPaw's AI enhancement API
.claude/skills/leoyeai-hitpaw-image-enhancer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 196% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 108% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 260% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 58% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 130% | 0% |
A powerful OpenClaw skill that integrates HitPaw's state-of-the-art AI enhancement technology for both images and videos. Enhance, upscale, restore, and denoise with multiple AI models.
Based on the official HitPaw API Documentation, this skill leverages industrial-grade AI models developed in-house by HitPaw's expert R&D team.
According to the Image API Introduction, our image processing services offer world-class capabilities designed to handle a wide variety of restoration scenarios:
The Image API offers two classes of AI models to suit different needs:
As detailed in the Available Models documentation:
| Model | Multiplier | Description | Best For | |-------|------------|-------------|----------| | general_2x / general_4x | 2x / 4x | General Enhance Model | General photos, landscapes | | face_2x / face_4x | 2x / 4x | Portrait Model (Clear) | Soft/beauty style portrait enhancement | | face_v2_2x / face_v2_4x | 2x / 4x | Portrait Model (Natural) | Natural/realistic portrait enhancement | | high_fidelity_2x / high_fidelity_4x | 2x / 4x | High Fidelity Model | Professional photography, conservatively upscaling high-quality sources | | sharpen_denoise_1x | 1x | Sharp Denoise Model | Aggressive denoising with sharpening | | detail_denoise_1x | 1x | Detail Denoise Model | Gentle denoising with texture preservation |
Powered by Stable Diffusion technology:
| Model | Multiplier | Description | Best For | |-------|------------|-------------|----------| | generative_portrait_1x/2x/4x | 1x/2x/4x | Generative Portrait Model | Extremely low-quality portraits, "re-imagines" details | | generative_general_1x/2x/4x | 1x/2x/4x | Generative Enhance Model | Heavily compressed or very low-resolution general images |
Technical Highlights:
bash# General photo upscaling (landscape, architecture) enhance-image -u landscape.jpg -m general_4x -o hd_landscape.jpg # Portrait beautification (soft skin) enhance-image -u selfie.jpg -m face_4x -o portrait_beautified.jpg # Professional archival restoration (natural look) enhance-image -u old_photo.png -m face_v2_2x -o restored.png --keep-exif # Denoise grainy low-light photo enhance-image -u night_photo.jpg -m sharpen_denoise_1x -o clean.jpg # Generative reconstruction for severely degraded image enhance-image -u blurry_face.jpg -m generative_portrait_2x -o ai_face.jpg
According to the Video API Introduction, our video processing services provide industrial-grade solutions for restoring and upscaling video content:
From the Video Models Documentation:
| Model | Description | Use Case | |-------|-------------|----------| | ultrahd_restore_2x | Ultra HD Model | High-definition upscale; natural-looking 1080p→4K | | general_restore_1x / 2x / 4x | General Restore Model | General video restoration, de-noising, de-blurring | | portrait_restore_1x / 2x | Portrait Restore Model | Multi-face restoration with temporal stability | | face_soft_2x | Video Face Soft Model | Facial beautification with consistent appearance | | generative_1x | Generative Video Model | Extreme restoration of heavily degraded footage |
Technical Highlights:
bash# Convert old 720p footage to 4K enhance-video -u old_clip.mp4 -m ultrahd_restore_2x -r 3840x2160 -o 4k_remastered.mp4 # Restore grainy, noisy home video enhance-video -u home_movie.avi -m general_restore_2x -r 1920x1080 -o cleaned.mp4 # Beautify faces in vlog/interview enhance-video -u interview.mp4 -m face_soft_2x -r 1920x1080 -o soft_faces.mp4 # Stabilize and restore old family footage with multiple faces enhance-video -u family_reunion.mov -m portrait_restore_2x -r 1920x1080 -o restored.mp4 # Generative AI restoration for severely degraded source enhance-video -u heavily_compressed.mp4 -m generative_1x -r 1920x1080 -o regenerated.mp4
Industry-Leading Quality: Professional-grade output suitable for commercial photography, archival restoration, and broadcast-quality video remastering
Unparalleled Fidelity: Strictly retains original details and subject identity, ensuring outputs remain true to inputs
Comprehensive Model Catalog: 16 specialized models covering virtually every restoration scenario
Scalable Performance: Optimized for low-latency, high-throughput workloads
| Scenario | Recommended Model | |----------|-------------------| | General photo upscale | general_2x or general_4x | | Portrait beautification | face_2x or face_4x | | Portrait natural look | face_v2_2x or face_v2_4x | | Professional archival | high_fidelity_2x / high_fidelity_4x | | Grainy low-light | sharpen_denoise_1x | | Subtle denoise | detail_denoise_1x | | Severely degraded | generative_portrait_* or generative_general_* |
| Scenario | Recommended Model | |----------|-------------------| | SD → 4K upscale | ultrahd_restore_2x | | General cleanup | general_restore_2x | | Interview/vlog beautification | face_soft_2x | | Old home movies (multiple faces) | portrait_restore_2x | | Severely compressed/ degraded | generative_1x |
bashclawhub install hitpaw-image-enhancer
Set your HitPaw API key:
bashexport HITPAW_API_KEY="your_api_key_here"
Or create a .env file in your OpenClaw workspace:
HITPAW_API_KEY=your_api_key_hereGet your API key at: https://playground.hitpaw.com/
Test the API directly in the browser: HitPaw Playground →
> Note: Place actual before/after screenshots in the images/ folder. See images/README.md for guidelines.
| Scenario | Before | After | |----------|--------|-------| | General Upscale (2x) | !Before | !After | | Face Enhancement | !Before | !After | | Generative Portrait | !Before | !After |
| Scenario | Original Frame | Enhanced Frame | |----------|----------------|----------------| | General Restoration | !Original | !Enhanced | | Portrait Restoration | !Before | !After |
enhance-image| Option | Type | Default | Description | |--------|------|---------|-------------| | --url, -u | string | required | URL of the image to enhance | | --output, -o | string | output.jpg | Output file path | | --model, -m | string | general_2x | Image model (see below) | | --extension, -e | string | .jpg | Output extension (.jpg, .png, .webp) | | --dpi | number | original | Target DPI for metadata | | --keep-exif | boolean | false | Preserve EXIF data from original | | --poll-interval | number | 5 | Polling interval in seconds | | --timeout | number | 300 | Maximum wait time in seconds |
| Model | Multiplier | Best For | DPI Support | |-------|------------|----------|-------------| | general_2x / general_4x | 2x / 4x | General photos, landscapes | ✅ | | face_2x / face_4x | 2x / 4x | Portrait & face enhancement | ✅ | | face_v2_2x / face_v2_4x | 2x / 4x | Improved face model | ✅ | | high_fidelity_2x / high_fidelity_4x | 2x / 4x | High quality preservation | ✅ | | sharpen_denoise_1x | 1x | Denoise & sharpen | ✅ | | detail_denoise_1x | 1x | Detail preservation | ✅ | | generative_* (1x/2x/4x) | — | AI generative fill | ❌ |
bash# Simple 2x upscale with general model enhance-image -u photo.jpg -o enhanced.jpg -m general_2x # Face enhancement 4x enhance-image -u portrait.jpg -m face_4x -o portrait_4x.jpg --keep-exif # High fidelity with custom DPI enhance-image -u old-photo.png -m high_fidelity_2x -dpi 300 -o hd.png # Batch processing for img in *.jpg; do enhance-image -u "$img" -o "upscaled/$img" -m general_4x done
enhance-video--resolution or -r). Must be in WIDTHxHEIGHT format (e.g., 1920x1080).--timeout to extend if needed.| Option | Type | Default | Description | |--------|------|---------|-------------| | --url, -u | string | required | URL of the video to enhance | | --output, -o | string | output.mp4 | Output file path | | --model, -m | string | general_restore_2x | Video model (see below) | | --resolution, -r | string | required | Target resolution in WxH (e.g., 1920x1080) | | --original-resolution | string | — | Original resolution (e.g., 1280x720) - optional | | --extension, -e | string | .mp4 | Output extension (.mp4, .mov, .avi) | | --fps | number | — | Target FPS (preserves original if omitted) | | --keep-audio | boolean | true | Preserve audio track | | --poll-interval | number | 10 | Polling interval in seconds | | --timeout | number | 600 | Maximum wait time in seconds |
| Model | Description | Use Case | |-------|-------------|----------| | general_restore_1x / 2x / 4x | General video restoration | General upscaling | | face_soft_2x | Face-softening enhancement | Portrait videos | | portrait_restore_1x / 2x | Portrait restoration | Face-focused content | | ultrahd_restore_2x | Ultra HD upscaling | Highest quality upscale | | generative_1x | Generative fill | AI-powered restoration |
bash# Upscale to 1080p using general_restore_2x enhance-video -u input.mp4 -o output_1080p.mp4 -m general_restore_2x -r 1920x1080 # Upscale to 4K with specific original resolution enhance-video -u clip.mov -o 4k.mov -m general_restore_4x -r 3840x2160 --original-resolution 1920x1080 # Denoise with portrait model enhance-video -u portrait_video.avi -m portrait_restore_2x -r 1920x1080 -o clean_portrait.mp4 # Add color to B&W (if generative model supports) enhance-video -u bw_vintage.mp4 -m generative_1x -r 1920x1080 -o colorized.mp4
Coin costs depend on video length, model, and resolution. Approximate rates:
Always check current rates at: https://playground.hitpaw.com/
Common errors and solutions:
| Error | Cause | Fix | |-------|-------|-----| | Invalid API key | Wrong or expired key | Update HITPAW_API_KEY | | Insufficient coins | Account balance too low | Top up at HitPaw Playground | | Unsupported model | Model name typo or not available | Check model table above | | Invalid extension | Output format not supported | Use .jpg/.png/.webp for images; .mp4/.mov/.avi for videos | | Invalid video URL | URL not publicly accessible | Ensure video is reachable via HTTPS | | Input/target resolution over limit | Exceeds 36 MP total pixels (e.g., 7680x4320 = ~33 MP) | Reduce resolution | | Video duration over limit | Video longer than 1 hour | Trim video first | | Rate limit exceeded | Too many requests | Wait and retry with exponential backoff | | Video processing failed | Corrupt video or unsupported codec | Try different input format or re-encode |
This skill implements the official HitPaw API as documented:
https://api-base.hitpaw.comPOST /api/photo-enhancerPOST /api/video-enhancerPOST /api/task-statusBoth endpoints return a job_id. Use the status endpoint to poll until COMPLETED, then download from res_url.
For longer videos, increase --timeout as needed (e.g., --timeout 3600 for 1 hour).
For videos, resolution is required. Choose based on your needs:
resolution to original dimensions (use --original-resolution for better quality).Max output: 36 megapixels total (width × height ≤ 36,000,000 pixels). Examples: 3840×2160 = 8.3 MP ✅, 7680×4320 = 33.2 MP ✅, 8192×4608 = 37.7 MP ❌.
By default, enhance-video keeps the audio track (--keep-audio, default true). Use --no-keep-audio to strip audio.
This skill is an unofficial integration with HitPaw API. You must have a valid API key and comply with HitPaw's terms. The skill author is not responsible for any charges incurred.
MIT © HitPaw-Official
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 7,754 | 2,951 | -62% | 1 | 1 | 0% | 1,838 | 5,443 | +196% | 0 | 0 | — |
case-02 | fail→pass | 10,782 | 3,160 | -71% | 1 | 1 | 0% | 2,709 | 5,633 | +108% | 0 | 0 | — |
case-03 | fail→pass | 6,628 | 2,916 | -56% | 1 | 1 | 0% | 1,535 | 5,525 | +260% | 0 | 0 | — |
case-04 | pass→pass | 4,351 | 4,755 | +9% | 1 | 1 | 0% | 1,037 | 5,842 | +463% | 0 | 0 | — |
case-05 | pass→fail | 7,821 | 4,020 | -49% | 1 | 1 | 0% | 1,435 | 5,745 | +300% | 0 | 0 | — |
case-06 | pass→fail | 8,072 | 7,854 | -3% | 1 | 1 | 0% | 2,050 | 6,362 | +210% | 0 | 0 | — |
case-07 | fail→pass | 16,384 | 3,176 | -81% | 1 | 1 | 0% | 3,539 | 5,593 | +58% | 0 | 0 | — |
case-08 | fail→pass | 11,058 | 2,925 | -74% | 1 | 1 | 0% | 2,409 | 5,537 | +130% | 0 | 0 | — |
case-09 | fail→pass | 7,038 | 2,778 | -61% | 1 | 1 | 0% | 1,295 | 5,454 | +321% | 0 | 0 | — |
case-10 | fail→pass | 8,866 | 2,792 | -69% | 1 | 1 | 0% | 1,997 | 5,535 | +177% | 0 | 0 | — |
case-11 | fail→pass | 9,152 | 2,571 | -72% | 1 | 1 | 0% | 2,171 | 5,380 | +148% | 0 | 0 | — |
case-12 | fail→pass | 5,266 | 2,381 | -55% | 1 | 1 | 0% | 1,167 | 5,342 | +358% | 0 | 0 | — |
case-13 | fail→pass | 2,070 | 1,848 | -11% | 1 | 1 | 0% | 436 | 5,198 | +1092% | 0 | 0 | — |
case-14 | pass→pass | 14,161 | 2,297 | -84% | 1 | 1 | 0% | 2,803 | 5,430 | +94% | 0 | 0 | — |
case-15 | fail→pass | 8,157 | 1,761 | -78% | 1 | 1 | 0% | 1,810 | 5,246 | +190% | 0 | 0 | — |
case-16 | fail→pass | 10,415 | 3,067 | -71% | 1 | 1 | 0% | 2,207 | 5,478 | +148% | 0 | 0 | — |
case-17 | pass→pass | 4,379 | 1,288 | -71% | 1 | 1 | 0% | 964 | 5,032 | +422% | 0 | 0 | — |
case-18 | pass→pass | 6,051 | 1,457 | -76% | 1 | 1 | 0% | 1,135 | 5,137 | +353% | 0 | 0 | — |
case-19 | pass→pass | 9,840 | 1,978 | -80% | 1 | 1 | 0% | 2,168 | 5,277 | +143% | 0 | 0 | — |
case-20 | fail→pass | 8,847 | 4,931 | -44% | 1 | 1 | 0% | 1,789 | 5,865 | +228% | 0 | 0 | — |
case-21 | fail→pass | 9,960 | 2,177 | -78% | 1 | 1 | 0% | 1,912 | 5,342 | +179% | 0 | 0 | — |
case-22 | fail→pass | 5,880 | 2,189 | -63% | 1 | 1 | 0% | 1,181 | 5,217 | +342% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +59 percentage points is the difference between those two pass rates over the 22 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.