term-llm
Docs → Guides → Video generation

Media

Video generation

Generate videos with Venice AI using text-to-video or image-to-video models.

On this page

Generate videos using Venice AI’s native video API.

term-llm video "a corgi surfing at sunset"

By default, videos are:

  • Saved to ~/Pictures/term-llm/ with timestamped filenames
  • Quoted before queueing so you can see the estimated cost
  • Polled until completion and written as .mp4

Video Flags

Flag Short Description
--input -i Input image for image-to-video
--reference -r Reference image(s) for models that support reference_image_urls (repeatable)
--output -o Custom output path
--model Venice video model override
--duration Video duration (5s, 10s)
--aspect-ratio Aspect ratio, e.g. 16:9, 9:16
--resolution Output resolution (480p, 720p, 1080p)
--negative-prompt Negative prompt
--audio Request audio for models that support it
--quote-only Quote the job and exit
--no-wait Queue the job and exit without polling
--json Emit machine-readable JSON to stdout
--poll-interval Poll interval while waiting
--timeout Maximum wait time
--debug -d Show debug information

Video Examples

# Text-to-video
term-llm video "a neon train passing through Tokyo at night"
term-llm video "a corgi surfing at sunset" --model kling-v3-pro-text-to-video

# Image-to-video
term-llm video "make Romeo blink and wag his tail" -i romeo.png
term-llm video "cute dog, influencer reacts" -i romeo.png --aspect-ratio 9:16 --duration 10s

# Multi-reference image-to-video (model support varies)
term-llm video "keep Romeo's face consistent while a Japanese influencer reacts" \
  -i romeo.png \
  -r romeo-closeup.png \
  -r romeo-profile.jpg \
  --model kling-o3-pro-image-to-video

# Planning and batch workflows
term-llm video "astronaut on mars" --quote-only
term-llm video "astronaut on mars" --quote-only --json
term-llm video "cyberpunk city" --no-wait --json

JSON Output

--json prints a single structured object to stdout, which is useful for scripts. Human progress lines stay on stderr when JSON mode is off.

Example:

{
  "provider": "venice",
  "prompt": "astronaut on mars",
  "model": "longcat-distilled-text-to-video",
  "duration": "5s",
  "resolution": "720p",
  "status": "queued",
  "quote": {
    "amount": 0.09
  },
  "job": {
    "queue_id": "123e4567-e89b-12d3-a456-426614174000"
  }
}

Credentials

term-llm video currently uses Venice AI and reads credentials from VENICE_API_KEY or image.venice.api_key in your config.

Defaults

If you do not specify a model, term-llm picks a cheap Venice default:

  • longcat-distilled-text-to-video for text-to-video
  • longcat-distilled-image-to-video for image-to-video

Model-specific inputs

Venice exposes advanced fields like reference_image_urls, scene_image_urls, end_image_url, and element-based references, but support is model-dependent. This command currently exposes the portable subset:

  • primary --input image
  • repeatable --reference images

If a chosen Venice model rejects reference images, the API will return that error directly instead of term-llm pretending otherwise.