Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Generate videos using Flyworks (a.k.a HiFly) Digital Humans. Create talking photo videos from images, use public avatars with TTS, or clone voices for custom audio.
.claude/skills/freestylefly-flyworks-avatar-video/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-14 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 61% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 14% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 52% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 18% | 0% |
This skill allows you to generate videos using Flyworks (a.k.a HiFly 飞影数字人) Digital Humans. Available features:
For detailed documentation, see the references/ folder:
This skill works with a default free-tier token, but it has limitations:
To remove limitations:
export HIFLY_API_TOKEN="your_token_here"scripts/hifly_client.pyThe main entry point for all operations.
bash# List available public avatars python scripts/hifly_client.py list_public_avatars # List available public voices python scripts/hifly_client.py list_public_voices # Create a video with a public avatar (TTS) python scripts/hifly_client.py create_video --type tts --text "Hello world" --avatar "avatar_id_or_alias" --voice "voice_id_or_alias" # Create a video with a public avatar (Audio URL or File) python scripts/hifly_client.py create_video --audio "https://... or path/to/audio.mp3" --avatar "avatar_id_or_alias" # Create a talked photo video using bundled assets python scripts/hifly_client.py create_talking_photo --image assets/avatar.png --title "Bundled Avatar" # Clone a voice using bundled assets python scripts/hifly_client.py clone_voice --audio assets/voice.MP3 --title "Bundled Voice" # Check status of generated tasks python scripts/hifly_client.py check_task --id "TASK_ID" # Manage local aliases (saved in memory.json) python scripts/hifly_client.py manage_memory add my_avatar "av_12345" python scripts/hifly_client.py manage_memory list
bash# First find a voice and avatar python scripts/hifly_client.py list_public_avatars python scripts/hifly_client.py list_public_voices # Generate python scripts/hifly_client.py create_video --type tts --text "Welcome to our service." --avatar "av_public_01" --voice "voice_public_01"
bash# Create the avatar from an image URL python scripts/hifly_client.py create_talking_photo --image "https://mysite.com/photo.jpg" --title "CEO Photo" # Output will give you an Avatar ID, e.g., av_custom_99 # Save it to memory python scripts/hifly_client.py manage_memory add ceo av_custom_99 # Generate video using the new avatar python scripts/hifly_client.py create_video --type tts --text "Here is the quarterly report." --avatar ceo --voice "voice_public_01"
When assisting users with video generation, follow these guidelines:
Video generation requires both text AND a voice. If the user provides text but no voice:
manage_memory list to see if the user has saved any voice aliases.list_public_voicesclone_voice --audio [file] --title [name]For a prompt like "Create a talking photo video from my photo saying 'this is my AI twin'":
clone_voicebash python scripts/hifly_client.py create_talking_photo \ --image user_photo.jpg \ --text "this is my AI twin" \ --voice SELECTED_VOICE_ID \ --title "My AI Twin"
After creating avatars or cloning voices, offer to save them:
bashpython scripts/hifly_client.py manage_memory add my_avatar AVATAR_ID --kind avatar python scripts/hifly_client.py manage_memory add my_voice VOICE_ID --kind voice
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-14 | fail→pass | 8,812 | 1,925 | -78% | 1 | 1 | 0% | 1,383 | 1,746 | +26% | 0 | 0 | — |
case-01 | fail→fail | 9,099 | 5,566 | -39% | 1 | 1 | 0% | 1,395 | 1,676 | +20% | 0 | 0 | — |
case-02 | fail→fail | 12,531 | 5,973 | -52% | 1 | 1 | 0% | 2,432 | 1,752 | -28% | 0 | 0 | — |
case-03 | fail→fail | 2,032 | 7,722 | +280% | 1 | 1 | 0% | 316 | 1,600 | +406% | 0 | 0 | — |
case-04 | fail→pass | 6,408 | 2,064 | -68% | 1 | 1 | 0% | 1,058 | 1,700 | +61% | 0 | 0 | — |
case-13 | fail→pass | 9,100 | 2,606 | -71% | 1 | 1 | 0% | 1,582 | 1,800 | +14% | 0 | 0 | — |
case-05 | pass→pass | 12,745 | 3,067 | -76% | 1 | 1 | 0% | 1,931 | 1,863 | -4% | 0 | 0 | — |
case-06 | pass→pass | 8,869 | 1,943 | -78% | 1 | 1 | 0% | 1,277 | 1,674 | +31% | 0 | 0 | — |
case-07 | fail→pass | 5,056 | 2,741 | -46% | 1 | 1 | 0% | 1,055 | 1,608 | +52% | 0 | 0 | — |
case-08 | fail→pass | 8,897 | 2,627 | -70% | 1 | 1 | 0% | 1,497 | 1,772 | +18% | 0 | 0 | — |
case-09 | fail→pass | 7,019 | 2,308 | -67% | 1 | 1 | 0% | 1,553 | 1,745 | +12% | 0 | 0 | — |
case-10 | pass→pass | 7,914 | 4,291 | -46% | 1 | 1 | 0% | 1,425 | 2,130 | +49% | 0 | 0 | — |
case-11 | fail→pass | 6,087 | 2,215 | -64% | 1 | 1 | 0% | 1,272 | 1,727 | +36% | 0 | 0 | — |
case-12 | fail→pass | 11,739 | 2,153 | -82% | 1 | 1 | 0% | 1,935 | 1,707 | -12% | 0 | 0 | — |
case-15 | fail→pass | 8,745 | 2,834 | -68% | 1 | 1 | 0% | 1,718 | 1,918 | +12% | 0 | 0 | — |
case-16 | fail→pass | 5,335 | 1,299 | -76% | 1 | 1 | 0% | 909 | 1,571 | +73% | 0 | 0 | — |
case-17 | fail→pass | 8,656 | 1,825 | -79% | 1 | 1 | 0% | 1,341 | 1,574 | +17% | 0 | 0 | — |
case-18 | fail→pass | 10,002 | 4,588 | -54% | 1 | 1 | 0% | 1,722 | 2,167 | +26% | 0 | 0 | — |
case-19 | pass→pass | 6,063 | 10,393 | +71% | 1 | 1 | 0% | 992 | 3,308 | +233% | 0 | 0 | — |
case-20 | fail→pass | 11,635 | 11,892 | +2% | 1 | 1 | 0% | 2,666 | 3,036 | +14% | 0 | 0 | — |
case-21 | fail→pass | 12,441 | 5,686 | -54% | 1 | 1 | 0% | 2,032 | 2,639 | +30% | 0 | 0 | — |
case-22 | fail→pass | 7,595 | 4,355 | -43% | 1 | 1 | 0% | 1,155 | 2,108 | +83% | 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, and 19 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +68 percentage points is the difference between those two pass rates over the 19 comparable cases. 1 case got worse with the skill loaded, and it is 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.