Black Forest Labs

FLUX.3 Video

FLUX.3 Video infers text-to-video, timestamped image-to-video, or video continuation from the supplied media. It supports automatic or explicit duration, multiple aspect ratios, optional synchronized audio, HD draft previews, and HD or FHD final output. The separate Draft Enhance model reproduces a saved preview at full quality from its private cache.

How to use this model

  1. 1

    Write only a prompt for text-to-video, add timed keyframes for image-to-video, or add one source clip for continuation; the task is inferred automatically.

  2. 2

    For image-to-video, add 1–10 images and place each one at a unique time on the timeline.

  3. 3

    Use automatic duration when the last keyframe should determine the result, or select an explicit duration for a fixed timeline.

  4. 4

    Enable draft mode for a lower-cost HD preview, then use Enhance draft on the saved gallery result for final HD or FHD quality.

3 Draft EnhanceAvailable

FLUX 3 Draft Enhance

Use Draft Enhance only after generating a FLUX.3 draft preview. Select the durable private cache saved with an eligible gallery result and choose HD or Full HD; the provider replays the original generation from that cache instead of asking for the prompt and source media again.

Try now

Properties

  • Required input: one owned FLUX.3 saved-draft cache.
  • Output resolution: HD or Full HD, priced from the source task and actual draft duration.

How to use this model

  1. 1

    Select one saved draft owned by your account.

  2. 2

    Choose HD to skip the final upscaling pass or Full HD for the finished result.

  3. 3

    Open the model directly or use Enhance draft on an eligible gallery card.

  4. 4

    If a draft is absent from the picker, its durable cache is unavailable and it cannot be enhanced.

Parameters

Saved Draft

draft_cache_id
Required

Resolution

resolution
Optional
Default: fhd

  • HD

  • Full HD

Recommended3 VideoAvailable

FLUX 3 Video

The active FLUX.3 Video version accepts a prompt for new generation, 1–10 timestamped images for a guided timeline, or a source video for continuation. The interface infers the task from those inputs, while saved-draft enhancement is exposed as its own model.

Try now

Properties

  • Inferred tasks: text-to-video, timestamped image-to-video, and video continuation.
  • Generation supports 5–20 second explicit duration or automatic duration, HD/FHD output, and optional synchronized audio.

How to use this model

  1. 1

    Provide either no source media, timed image keyframes, or one source video so the intended task is unambiguous.

  2. 2

    Keep keyframe times ordered, unique, and within the displayed timeline.

  3. 3

    Select synchronized audio only when the prompt describes the intended sound.

  4. 4

    Enable Draft Preview for a lower-cost HD result, then open that result with the separate Draft Enhance model if it is worth finishing.

Parameters

Task

task
Optional
Default: t2v

  • Text to Video

  • Image to Video

  • Video Continuation

Prompt

prompt
Optional

Automatic Duration

auto_duration
Optional
Default: true

Duration

duration
Optional
Default: 10Range: 5 – 20

Aspect Ratio

aspect_ratio
Optional
Default: auto

  • Automatic

  • 21:9

  • 2:1

  • 16:9

  • 4:3

  • 1:1

  • 3:4

  • 9:16

Resolution

resolution
Optional
Default: hd

  • HD

  • Full HD

Synchronized Audio

generate_audio
Optional
Default: true

Draft Preview

draft
Optional
Default: false

Keyframes

keyframes
Optional

Source Video

start_video
Optional