LiblibAI Image & Video Generation

Generate images (Seedream4.5) and videos (Kling) via LiblibAI's API.

Prerequisites

Environment variables must be set:

• LIB_ACCESS_KEY — API access key
• LIB_SECRET_KEY — API secret key

Usage

Run the CLI script at scripts/liblib_client.py:

# Generate image
python3 scripts/liblib_client.py image "a cute cat wearing a hat" --width 2048 --height 2048

# Generate video from text
python3 scripts/liblib_client.py text2video "a rocket launching into space" --model kling-v2-6 --duration 5

# Generate video from image
python3 scripts/liblib_client.py img2video "the cat turns its head" --start-frame https://example.com/cat.jpg

# Check task status
python3 scripts/liblib_client.py status <generateUuid>

Image Generation (Seedream4.5)

• Endpoint: POST /api/generate/seedreamV4
• Model: doubao-seedream-4-5-251128
• Default size: 2048×2048. For 4.5, min total pixels = 3,686,400 (e.g. 2560×1440)
• Supports reference images (1-14), prompt expansion, and sequential image generation
• Options: --width, --height, --count (1-15), --ref-images, --prompt-magic

Video Generation (Kling)

Text-to-Video

• Endpoint: POST /api/generate/video/kling/text2video
• Models: kling-v2-6 (latest, supports sound), kling-v2-1-master, kling-v2-5-turbo, etc.
• Options: --model, --aspect (16:9/9:16/1:1), --duration (5/10s), --mode (std/pro), --sound (on/off)

Image-to-Video

• Endpoint: POST /api/generate/video/kling/img2video
• Provide --start-frame image URL; optionally --end-frame (v1-6 only)
• For kling-v2-6: uses images array instead of startFrame/endFrame

Async Pattern

All generation tasks are async:

• Submit task → get generateUuid
• Poll POST /api/generate/status with { "generateUuid": "..." }
• Result contains images[].imageUrl or videos[].videoUrl

Notes

• QPS limit: 1 request/second for task submission
• Max concurrent tasks: 5
• Image URLs in results expire after 7 days
• For kling-v2-5-turbo and kling-v2-6, mode must be "pro" (default)
• Sound generation only supported on kling-v2-6+