Skip to main content

One post tagged with "videos"

View All Tags

Reusable Video Characters with LiteLLM

Sameer Kankute
Sameer Kankute
SWE @ LiteLLM
Krrish Dholakia
Krrish Dholakia
CEO, LiteLLM

Upload a video character once, reference it across unlimited generations. LiteLLM now handles character management with full router support.

What's New​

Four new endpoints for video character operations:

  • Create character - Upload a video to create a reusable asset
  • Get character - Retrieve character metadata
  • Edit video - Modify generated videos
  • Extend video - Continue clips with character consistency

Available from: LiteLLM v1.83.0+

Quick Example​

import litellm

# Create character from video
character = litellm.avideo_create_character(
    name="Luna",
    video=open("luna.mp4", "rb"),
    custom_llm_provider="openai",
    model="sora-2"
)
print(f"Character: {character.id}")

# Use in generation
video = litellm.avideo(
    model="sora-2",
    prompt="Luna dances through a magical forest.",
    characters=[{"id": character.id}],
    seconds="8"
)

# Get character info
fetched = litellm.avideo_get_character(
    character_id=character.id,
    custom_llm_provider="openai"
)

# Edit with character preserved
edited = litellm.avideo_edit(
    video_id=video.id,
    prompt="Add warm golden lighting"
)

# Extend sequence
extended = litellm.avideo_extension(
    video_id=video.id,
    prompt="Luna waves goodbye",
    seconds="5"
)

Via Proxy​

# Create character
curl -X POST "http://localhost:4000/v1/videos/characters" \
  -H "Authorization: Bearer sk-litellm-key" \
  -F "video=@luna.mp4" \
  -F "name=Luna"

# Get character
curl -X GET "http://localhost:4000/v1/videos/characters/char_xyz123" \
  -H "Authorization: Bearer sk-litellm-key"

Key Features​

✅ Full Router Support - Load balance across multiple model deployments
✅ Character Encoding - Automatic provider/model tracking in character IDs
✅ Error Handling - Proper HTTP status checks before response parsing
✅ Backward Compatible - External providers receive NotImplementedError, not instantiation errors
✅ Multi-Deployment - Router picks optimal deployment when target_model_names is set

Best Practices​

Character uploads:

  • 2-4 seconds optimal
  • Match target resolution (16:9, 9:16, or 1:1)
  • 720p-1080p
  • Clear character isolation

Prompting:

✅ "Luna the fox dances through a cosmic forest, stars trailing her movement"
❌ "A character that looks like Luna"

Always mention character name verbatim in prompt.

Implementation Notes​

All four handler methods now include:

  • response.raise_for_status() - Proper error detection before model parsing
  • Router-first dispatch - Consistent with avideo_edit/extension
  • Async support - Full async/await pattern

What's Inside​

  • 8 handler methods (sync + async pairs)
  • Character transformation classes
  • SDK functions + Router wiring
  • Full test coverage
  • Comprehensive error handling

Common Issues​

Character doesn't appear?

  • Include character ID in characters array
  • Use character name in prompt (exact match)
  • Ensure character occupies meaningful screen space

Distorted character?

  • Character video aspect ratio must match target resolution
  • Upload again with matching dimensions

Want to edit with character?

  • Use avideo_edit (currently no character support in extensions)
  • Edit preserves original composition

Next Steps​

Resources: