MODEL PROFILE · fal (hosted endpoint)
LongCat Video Distilled I2V 720p
An endpoint-specific profile for animating an input image. Distinguish its documented controls from other LongCat request types and from untested quality claims.
Source reviewed: · Video model · image-to-video
Provider documentation summary. No hands-on results or independent ranking are claimed.
Documented facts
The following fields come from the official provider source, reviewed on the date above.
- Exact hosted identity
- fal-ai/longcat-video/distilled/image-to-video/720p. Public API documentation exists; successful inference was not tested.
- Inputs and output
- Required image_url, optional prompt; provider describes 720p/30fps video. Output includes video, prompt and seed.
- Timing defaults
- num_frames defaults to 162 and fps to 30. A nominal 5.4 seconds follows by division; this is not a verified maximum duration.
- Sampling controls
- num_inference_steps and num_refine_inference_steps each default to 12. Seed and prompt expansion are also exposed.
- Export controls
- X264 MP4 is default; VP9 WebM, ProRes4444 MOV and GIF are listed. Quality and write-mode controls are separate settings.
- Unknown commercial details
- This review establishes no price, launch date, downloadable distilled checkpoint or license. Provider labeling is not a license audit.
Read the endpoint input section, not every type on the page
The page’s generic About line describes video-to-video, while the endpoint title and required image_url identify image-to-video. Its Other types section also contains a different request with negative_prompt, guidance_scale and aspect_ratio. Do not assume those fields are accepted by this route. The uncertainty is in the documentation; no request was made to resolve it. Endpoint schema and other request types.
The practical decision
Use this route as an evaluation candidate when a retained still image must become a moving shot. Define the motion you need and the details that must survive it before comparing providers. A product silhouette, face, logo or background boundary can fail at different moments, so inspecting only the first frame is insufficient. Choose a small collection of existing inputs that exposes those weaknesses, then inspect matching moments across saved results.
What to record in your own comparison
- Record the exact route, reference image, prompt, seed and every non-default setting. A neighboring endpoint name is not evidence that two results used the same model configuration.
- Inspect the first, middle and final frames, then play the full clip. Look for identity drift, edge deformation, incomplete actions and unintended camera movement. These are proposed checks, not observed defects in this model.
- Compare matched dimensions and playback timing. An export container does not establish native generation quality, and a larger encoded file does not prove greater visual fidelity.
- Keep generation settings separate from delivery settings. When costing an accepted shot, include retries and any downstream processing once a current price is verified. No cost or throughput advantage is asserted here.
Limitations and unknowns
No generated samples, measured speed, comparative ranking or endpoint availability test. Native audio, maximum frame count, aspect-ratio handling, exact model developer attribution and self-hosting terms remain unverified for this distilled hosted route. Do not import limits or licensing from the base LongCat model.
Compare your saved results
Use the DualView editor to inspect files you already have. The accepted-output calculator uses your own rates and counts. Neither link runs a model or purchases generation.
Related reporting
- H3 Max versus H3: separate video endpoints and dated costs
- Inspect existing video outputs in DualView