Video

Generate video from text or from an image. The platform offers two engines: fal is the main one (many models), and Higgsfield is a DoP-style image-to-video option.

What you can do

Three ways to generate video:

• Text → video (t2v) — describe a scene, get a clip.
• Image → video (i2v) — animate a still image.
• First frame → last frame — interpolate motion between two stills.

Some models also support reference-to-video, extend/continue, and edit. Which modes you get depends on the model you pick:

Choosing a model

fal_generate_video is the workhorse: one entry point to many top models. Pick a model with model_id, and pass that model's parameters in input. Each family supports different modes and fields:

higgsfield_generate_video — DoP-style image-to-video. Models: higgsfield-ai/dop/lite / dop/standard / dop/turbo (each with a /first-last-frame variant). It has no pure text-to-video — for text → video, weave an image first (see below).

Interface

fal_generate_video — submit a job. Inputs: model_id (an endpoint id) and input (that model's full parameter object, passed through). Returns a request_id.

fal_get_video_status — poll a job. Pass both model_id and request_id. Returns IN_QUEUE / IN_PROGRESS, or COMPLETED with the video URL.

higgsfield_generate_video — submit. Inputs: prompt, image_url (required), model_id, params (model-specific, passed through). Returns a request_id.

higgsfield_get_video_status — poll. Pass request_id only. Returns in_progress, completed (with the video URL), or failed / nsfw.

Using it in a Skill

Video generation is asynchronous — two steps:

1. Submit with *_generate_video → get a request_id.
2. Poll *_get_video_status until it's COMPLETED → get the video URL.

Generation can take a few minutes — keep polling until it completes. For fal, pass both model_id and request_id when polling; for Higgsfield, only request_id.

Text → video on an image-only engine (or whenever you want art-directed framing): generate an image first, then feed it as the start frame.

# Text → video (Veo 3.1, with audio, 8 seconds)
fal_generate_video(
  model_id="fal-ai/veo3.1",
  input={ "prompt": "A cinematic drone shot over a misty mountain lake at dawn",
          "duration": "8s", "resolution": "1080p", "generate_audio": true }
)

# Image → video (Kling — note start_image_url, not image_url)
fal_generate_video(
  model_id="fal-ai/kling-video/v3/pro/image-to-video",
  input={ "prompt": "the character slowly turns and smiles",
          "start_image_url": "https://…", "duration": "5" }
)

Weave for image → video:

1. google_generate_image / openai_generate_image(prompt)        → image
2. fal_generate_video(model_id=".../image-to-video",
     input={ prompt: motion, image_url: <image> })              → request_id
   (or higgsfield_generate_video with image_url)
3. poll *_get_video_status until COMPLETED                      → video URL