Create video from text or images, guide shots with keyframes, extend existing clips, and replace the soundtrack with supplied audio.
Describe the scene, action, and camera movement to generate video from text. Build the brief around a specific moment, such as a product entering the frame or a character performing a clear gesture, rather than a loose list of visual adjectives.
The fal I2V task accepts a first frame, last frame, or both. The canvas follows the first image when present, otherwise the last. Omitting both images falls back to text-to-video with a default 16:9 canvas.
fal extend-video accepts a 1.625–60s source up to 50 MB and adds 1–15s. Return the full video or only the continuation. Formats include MP4, MOV, WebM, M4V, and GIF; format alone does not establish full source compatibility.
For T2V/I2V, target_audio_url replaces the soundtrack: longer audio is trimmed; shorter audio gains silence. Audio must be at least 2 seconds and at most 15 MB. This does not establish native audio or motion guidance; extension uses separate audio references.
Text-to-video lists 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16. Image tasks follow the supplied keyframe canvas; extension follows the source by default. For text and image generation, 1080P is refinement from native 768P, not native 1080P generation.
| Dimension | H3 Max Turbo | Veo 3.1 | Wan 3.0 Prime |
|---|---|---|---|
| Input formats | Text and image; extension by task | Text and image; visual references | Text, image, video, audio; task-specific combinations |
| Core direction | Video generation, keyframe guidance, and extension | Video generation with audio and visual guidance | Multi-shot video with multimodal references |
| Audio capabilities | Uploaded audio can replace the soundtrack | Native audio generation | Native dialogue, music, and effects |
| Video duration limits | Up to 15s requested per generation; extension adds 1–15s | Gemini API: 4, 6, or 8s initially; references and 1080p/4K require 8s | Model Studio: up to 30 seconds |
| Output resolution | fal T2V/I2V: 480P or 768P; 1080P refinement from 768P. Extension has separate settings. | Gemini API: 720p; 1080p/4K at 8s; extension at 720p | Model Studio: 480P, 720P, 1080P |
| Suggested uses | Keyframe-led shots and continuation of existing clips | Scenes planned around sound, visual references, and continuation | Multi-shot stories and reference-led scene development |
H3 Max Turbo sources: Text-to-video API · Image-to-video API · Video extension API
Describe a shot in text or start from an existing image and specify what happens next. This suits motion proposals developed from static concepts; begin with a focused subject and clear action.
Describe the action or camera movement that should follow the end of a clip. This suits content that needs to extend a moment; inspect the transition and visual changes at the join.
Try a different soundtrack on an existing video to compare rhythm and mood. This suits a workflow where the visuals are established before sound is explored; check how the replacement fits the edit.
Explore video workflows for short motion studies, follow-on scenes, and different creative moods.
The fal extension page specifies a source lasting 1.625-60 seconds and no larger than 50 MB. Its upload form lists MP4, MOV, WebM, M4V, and GIF. Meeting the format list alone is insufficient; the source must also satisfy the duration and size limits.
The extend-video task can return either the full extended video or only the continuation. Choose according to how the result will be used. If you request only the new portion, retain the source clip so the transition can still be checked in context.
Yes. The fal image-to-video specification permits end_image_url by itself for end-only keyframe generation, and the output canvas follows that image. When a first image is also supplied, the first image controls the canvas. The requested motion should connect naturally to the chosen visual destination.
The text-to-video and image-to-video specifications provide disabled, balanced, and quality prompt-expansion modes. Disabled skips expansion. Use that choice when you want the submitted wording to remain the basis of the request, and assess the visual result rather than inferring greater fidelity from the setting alone.
No. Text-to-video and image-to-video use target_audio_url for supplied soundtrack replacement. The extension page describes audio references that replace the source soundtrack as reference material. Apply each task’s own input conditions; neither description alone establishes precise lip synchronization or beat-driven motion.
