Sjinn Video Generation

SJinn Video Generation

Use this skill when the user wants to create a video from text or reference media, animate an image, use an end frame, or download a generated video through SJinn.

Prerequisites

Before generating, quickly verify the CLI and login:

sjinn --version
sjinn auth whoami

The CLI must be 0.1.18 or newer for MiniMax H3 and Wan3. If either command fails or the CLI version is older, use sjinn-setup first.

Workflow

1. Choose a model and options. For model-specific options, see references/models.md.
2. Prefer JSON output so results are easy to parse.
3. Use --async for long-running jobs unless the user explicitly wants to wait.
4. Use --download [path] only when the user asks to save the result locally.

Commands

Text to video:

sjinn video generate --prompt "a cinematic dolly shot through a neon studio" --model veo3 --aspect 16:9 --async --json

Image to video from a URL:

sjinn video generate --prompt "slow camera push in, soft evening light" --image "https://example.com/frame.png" --model kling3 --duration 5 --json

Image to video from a local file:

sjinn video generate --prompt "animate this character turning toward camera" --image-urls "./character.png,./style-ref.png" --model gemini-omni-video --aspect 16:9 --duration 6 --async --json

Seedance 2.0 with multiple local references:

sjinn video generate --prompt "two consistent characters walking through a research lab" --model seedance2 --media-urls "./character-a.png,./character-b.png,./lab-reference.mp4" --duration 5 --mode fast --resolution 480p --async --json

Seedance 2.5 with image, video, and audio references:

sjinn video generate --model seedance2.5 --prompt "Use @Image 1 for the presenter and @Image 2 for the showroom styling, follow the movement in @Video 1, and use @Audio 1 as the soundtrack." --image-urls "./presenter.jpg,./showroom.png" --video-urls "./camera-motion.mp4" --audio-urls "./music-bed.wav" --duration 8 --aspect 16:9 --resolution 720p --async --json

For Seedance 2.5, use --image-urls, --video-urls, and --audio-urls for reference inputs. Reference markers map to each option's comma-separated order and count independently by media type: the first image is @Image 1, the first video is @Video 1, and the first audio file is @Audio 1. Only reference resources provided in the command. Do not combine them with --image, --media-urls, or --mode.

MiniMax H3 with a reference image (2K output):

sjinn video generate --model minimax-h3 --prompt "the character turns as the camera moves closer" --image-urls "./character.png" --duration 6 --async --json

Wan3 with a reference image and no prompt:

sjinn video generate --model wan3 --image-urls "./scene.png" --duration 10 --resolution 720p --async --json

For MiniMax H3 and Wan3, use --image-urls, --video-urls, and --audio-urls for references, including a single image. Do not use --image, --mode, --multi-shot, --end-image, or --media-urls. Only Wan3 can omit --prompt, and only when at least one reference is provided. Check references/models.md for reference counts and duration limits before submitting.

End-frame video:

sjinn video generate --prompt "transition from morning to sunset" --image "./start.png" --end-image "./end.png" --model veo3 --json

Output Handling

On success, return the task ID, primary URL, any additional URLs, model, and elapsed time. If status is created, tell the user to query the task with sjinn-task-status.

On error, summarize the CLI error plainly and suggest sjinn-setup only when the error indicates the CLI is missing or unauthenticated.