FLUX 3 Video Creation
FLUX 3 is Black Forest Labs' preview video model, supporting the creation of videos with synchronized audio from text, image keyframes, or existing clips. The platform model ID is flux-3, and generation, editing, and upscaling tasks are submitted through a unified video API.
Supported Capabilities
Text-to-video is suitable for starting from a shot brief; image keyframes can constrain the visuals; video continuation can extend existing clips. Draft mode is used to view a quick preview first, then reproduce full quality using your own draft task. Editing and upscaling are provided as separate operations through the same API.
Key Specifications
- Generation modes
- t2v、i2v、v2v、draft_enhance
- Standard generation duration
- 5–20 whole seconds or auto; video continuation is 5–15 seconds
- Output tiers
- hd、fhd、qhd、uhd; draft supports hd only
- Image keyframes
- 1–10 images, with keyframes assignable by time
- Audio
- Generation modes support synchronized audio, and generate_audio can explicitly be set to false
- Delivery
- Asynchronous platform task ID, video link, actual duration, dimensions, and frame rate
Use Cases
You can create short shots for product pitches and constrain visuals with product images; you can also explore scenes and camera movements based on a storyboard description. Choose video continuation when an existing clip needs to be extended, editing when visuals need to be modified, and upscaling when the final video size needs to be increased. Text, subject details, motion continuity, and audio should still be manually reviewed.
Operations and Billing
Submit action=generate, edit, or upscale to POST /flux/videos; omitting action is equivalent to generate. Generation and editing are billed by actual output seconds, while upscaling is measured by actual output MP·seconds and adjusted according to the output frame rate relative to 24 fps. 1 MP is 1024×1024 pixels. Unit prices vary by operation; refer to the current rules on the pricing page for specific Credits.
Get Started
Use model=flux-3 and an appropriate mode for generation; image-to-video also requires keyframes, and continuation also requires start_video. After setting async=true, obtain the platform task_id and query results through POST /flux/tasks, or use callback_url to receive completion notifications. CLI, MCP, and Skill use the same public interface and task ID.
View the Video API Integration GuideUsage Boundaries
FLUX 3 is still in preview, and supported scope and limitations are subject to the current API documentation. Draft cache is temporary and cannot be guaranteed to be permanently reproducible; draft_enhance uses the current caller's draft_task_id and cannot override the original prompt and assets. Editing or upscaling does not accept generation-specific parameters such as mode and duration. Please use materials you have the right to process, and check the final video's actual dimensions, duration, and content.
Frequently Asked Questions
Why do generation, editing, and upscaling use the same endpoint?
Use action to select the operation, then provide the parameters for that operation. Within generation, use mode to select text, image, continuation, or draft enhancement.
How do I submit image-to-video generation?
Select action=generate, mode=i2v, and provide keyframes. You can submit an image list or keyframes arranged by seconds.
How do I enhance a draft?
Use the returned draft_task_id and submit with mode=draft_enhance. The draft must belong to the current caller, and the temporary cache must still be valid.
Is upscaling charged based on the input dimensions?
Billing is based on the actual output MP·seconds of the delivered video; measurement considers actual duration, width, height, and frame rate.
How do I get the finished video for an asynchronous request?
Set async=true to obtain task_id, then query the completed result through /flux/tasks. The video link and actual media information are in the completion response.