Seedance 2.5 is live — 30-second cinematic video with native audio & real-person references
FLUX 3 Video API Pricing: Draft, HD, FHD, and Continuation
2026/08/23

FLUX 3 Video API Pricing: Draft, HD, FHD, and Continuation

Calculate FLUX 3 video costs across Draft, HD, FHD, keyframe, and continuation modes, including the draft-to-final workflow and 5–20s totals.

FLUX 3 Video costs from $0.056 to $0.488 per second on reAPI, depending on whether the request is a Draft, a normal text/keyframe generation, or a video continuation at HD or FHD. A ten-second job can therefore cost $0.56, $1.57, $2.67, $3.78, or $4.88 even though every request carries the same model id.[1]

The mode is the price. Treating FLUX 3 as one per-second number produces bad estimates, particularly for continuation.

The six FLUX 3 rates

Public reAPI rates checked August 23, 2026:

Request typeResolutionPrice per second10-second cost
Draft from text/keyframesHD$0.056$0.56
Text or keyframesHD$0.157$1.57
Text or keyframesFHD$0.267$2.67
Continuation DraftHD$0.111$1.11
Video continuationHD$0.378$3.78
Video continuationFHD$0.488$4.88

FLUX 3 accepts integer durations from 5 to 20 seconds. The bill is simply the tier rate multiplied by requested duration.[2]

Tier5 seconds10 seconds20 seconds
Draft$0.28$0.56$1.12
Text/keyframes HD$0.79$1.57$3.14
Text/keyframes FHD$1.34$2.67$5.34
Continuation Draft$0.56$1.11$2.22
Continuation HD$1.89$3.78$7.56
Continuation FHD$2.44$4.88$9.76

Rounded table totals use the displayed public rate. The final credit charge follows the live rate card.

Why continuation costs more

A text request asks the model to create a clip from a prompt. A continuation request also has to read the source video's picture and audio, establish its state, and carry both forward. FLUX 3 places that job in a separate tier.[2]

At HD, continuation is about 2.4 times the standard text/keyframe rate. Starting from a video solely because it feels more controlled can therefore be expensive. Use continuation when the source contains motion, timing, or audio continuity that a still keyframe cannot express.

If only the initial composition matters, send one image as the first keyframe. It stays in the standard tier.

Keyframes do not add a surcharge

One to ten image_urls can guide a FLUX 3 generation. One image is the start frame; two are start and end; three or more become evenly spaced middle frames in array order. Keyframe jobs use the standard text/image tier rather than the continuation tier.[2]

That makes a ten-second FHD transition between two approved designs $2.67 at the current rate:

{
  "model": "flux-3-video",
  "prompt": "A clean mechanical transformation between the two product states",
  "image_urls": [
    "https://files.example.com/start.png",
    "https://files.example.com/end.png"
  ],
  "duration": 10,
  "resolution": "fhd",
  "audio": true
}

The order is semantic. Renaming the files or sorting the URLs later can change the intended timeline.

Draft-to-final is two bills, not one discounted final

draft: true renders a low-quality HD preview. If it works, submit a second request with draft_from_task_id to finalize that exact Draft at normal quality. The finalization is billed at the corresponding normal tier.[2]

For a ten-second text-to-video job:

Draft HD                  $0.56
Finalize same draft, HD   $1.57
Total                     $2.13

Going directly to HD would be $1.57, so Draft does not save money on a first-try success. It saves money when several bad ideas can be rejected before full rendering.

Suppose four concepts are tested and one is finalized:

4 Drafts × $0.56          $2.24
1 HD final                $1.57
Total                     $3.81

4 direct HD attempts      $6.28

The Draft loop saves $2.47 in that example. It works because three ideas die cheaply.

There is an API constraint worth designing around: finalization takes the completed Draft task id, and only resolution may change. Sending a new prompt, duration, or media with draft_from_task_id is rejected. If the prompt changes, it is a new generation rather than a finalization.

Silent video is not cheaper

FLUX 3 Video generates native audio by default. Setting audio: false returns a silent clip but does not select a lower rate.[2] Disable audio when the output should be silent or a separate sound mix is planned, not as a cost control.

A cost-aware routing rule

Use this order:

  1. Prompt only when no visual state has been approved.
  2. One or more keyframes when approved stills define the important moments.
  3. Continuation only when source motion or audio must carry into the new clip.
  4. Draft first when several concepts are likely to be discarded.
  5. Direct final when the prompt and references are already locked.

Break-even math for the Draft loop

Draft is worthwhile when the number of rejected ideas is high enough to cover its extra charge on the accepted idea.

For a ten-second HD job, let n be the number of concepts. One direct HD attempt is $1.57; one Draft is $0.56, and a single accepted concept still needs a $1.57 final.

direct workflow = n × $1.57
draft workflow  = n × $0.56 + $1.57

At one concept, direct costs $1.57 and Draft-to-final costs $2.13. At two, both routes are close: $3.14 direct versus $2.69 for two Drafts and one final. From two competing concepts onward, the Draft path is already cheaper if only one is finalized.

The math assumes a Draft reliably predicts whether the final composition and motion will work. It may not reveal small text, fine texture, or FHD-only defects. If those details determine acceptance, one early direct final should calibrate how much trust to place in Draft reviews.

Price a continuation before choosing it

Continuation should earn its higher rate by removing an edit or preserving something that cannot be captured in a still.

For a 20-second FHD continuation, the displayed total is $9.76. A new 20-second FHD text/keyframe generation is $5.34. The $4.42 difference buys access to the source clip's moving state and audio continuity.

Ask three questions before paying it:

  • Does the first new frame need to match a moving source, or would the source's last frame be enough?
  • Must the original soundtrack continue without a cut?
  • Is the source motion actually useful, or is it merely being attached as a visual reference?

If the last still plus a prompt can define the next shot, keyframe generation stays on the standard tier. If a person is mid-action or the audio phrase must continue naturally, continuation has information the still cannot carry.

Production preflight

Before submitting a batch, validate the same fields used by billing:

CheckWhy it matters
Duration is an integer 5–20Invalid values fail; omitted duration becomes 5
Draft uses HDDraft does not have an FHD tier
Keyframe array order is intentionalReordering changes the timeline
Video field is present only for continuationIt selects the expensive tier
Finalization contains no new prompt/mediaA completed Draft is finalized as-is
Audio choice is creative, not financialSilent output receives no discount

Store the resolved tier beside the task id. When a user disputes a continuation bill, the application should be able to show which input selected that tier rather than reconstructing the payload from memory.

The FLUX 3 model page carries the live six-tier rate card. The FLUX 3 API documentation covers the request modes, keyframe order, Draft finalization, and polling.

References

  1. reAPI FLUX 3 model and pricing page, accessed August 23, 2026.
  2. reAPI FLUX 3 API documentation, accessed August 23, 2026.
  3. Black Forest Labs, “FLUX 3 Video”, August 4, 2026.