HeyGen: HeyGen Video

HeyGen Video is a general-purpose video generation model from HeyGen, listed on api14 under the model id heygen/heygen-video-1. It sits in the video category and is built for rendering short clips rather than for text or chat tasks. If you are evaluating it, the main thing to understand is that it is a generation endpoint: you give it source material, and it returns a video file.

On the input side, the model accepts text, images, video, and audio, so you can drive a clip from a written prompt, a reference still, existing footage, or an audio track. The only output modality is video. There is no published context length for this model, so token-based planning does not apply here. The output details are more useful: clips can be rendered at 480p, 768p, or 2K, in aspect ratios of 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16, and durations run from 5 to 15 seconds in one-second steps. A first_frame image can be supplied to anchor the opening shot. Audio generation is not enabled, so the model does not synthesize dialogue, ambience, or sound effects on its own.

Those capabilities point to a fairly specific set of jobs. Short vertical clips at 9:16 for social feeds, square 1:1 cuts for in-feed placements, and widescreen 16:9 or 21:9 for landing pages and presentations all fall inside the supported aspect ratios. Because a first_frame image is accepted, you can hold a consistent look across a series of clips by reusing the same starting still. Because video and audio are accepted as inputs, you can also work from existing footage or a supplied track rather than starting from text alone. The 5 to 15 second range suits teasers, product shots, and loopable background segments rather than long-form narrative.

Calling it through api14 works the same way as any other model on the gateway: use the OpenAI-compatible endpoint and pass heygen/heygen-video-1 as the model id. The model exposes no supported request parameters, so there are no sampling or generation flags to tune in the request body. In practice that means your control over the result comes from the inputs you provide (prompt text, reference image, source video, audio) and from the output settings the model defines, such as resolution, aspect ratio, duration, and the first frame.

Two limits are worth weighing before you commit. There is no context length to plan around, and the model does not generate audio, so any project that depends on synthesized speech, ambience, or sound effects will need those elements produced elsewhere and combined afterwards. If your task is text generation, reasoning, or anything outside video output, a different model on api14 will fit better. Within short-form video generation, though, the combination of multiple input modalities, a first-frame anchor, and a wide spread of aspect ratios and durations covers most common clip requirements.