PRACTICAL GUIDE · PROMPTTOVISUAL
FLUX 3 Video: API Modes, Audio, Drafts and Pricing
Understand BFL’s FLUX 3 Video API, keyframes, continuation, resolution and provider pricing. Independent guide; PromptToVisual does not host this model.
VIDEO OFF · INDEPENDENT GUIDE
Seedance availability · FLUX 3 API guide · Sora API migration · Prepare a still reference · All guides
What is FLUX 3 Video, and is it hosted here?
FLUX 3 Video is Black Forest Labs’ video-and-audio model. PromptToVisual does not host FLUX 3 Video and does not currently provide video generation. This guide helps you read the upstream API contract and choose an input mode. It contains no generated sample, measured latency, quality ranking or promise that another model here is equivalent.
What happened on August 4, 2026?
BFL’s August 4 release note introduced video with synchronized audio and the flux-3-video endpoint. The current note calls that release a preview, not API GA. We therefore do not label August 4 as verified general availability. The current documentation and pricing pages show an accessible API product, but access and a historical GA label are different claims.
Choose text, images or an existing clip
The flux-3-video route distinguishes t2v, i2v and v2v. Text-to-video and image-to-video support 5–20 seconds; image-to-video accepts a prompt plus 1–10 keyframe images. Video continuation takes a prompt and an existing clip, with a narrower documented 5–15-second output range. Do not treat continuation as a generic edit or assume the same duration limit for every mode.
How do keyframes, multiple scenes and audio fit together?
Use one still to establish an opening, or ordered keyframes to guide the sequence. BFL documents multiple scenes and camera angles in one generation. Audio can include speech, ambience and effects; generate_audio is enabled by default and can be disabled. This is documented capability, not our guarantee of identity continuity or lip-sync accuracy. Inspect cuts and sound as separate acceptance criteria.
HD and FHD are delivery classes, not exact round dimensions
Current BFL documentation gives 1280 × 704 for 16:9 HD and 1920 × 1088 for FHD. It now also lists QHD and UHD. HD comes directly from generation; the higher classes use the video upsampler. Do not describe every delivered class as native generation, or promise exactly 1920 × 1080 from the FHD label. Check the current dimension table for your aspect ratio.
Use a draft to choose a direction before a full render
BFL’s draft mode returns an HD preview and a draft_cache bundle. The documented draft_enhance mode uses that saved bundle to finish the selected generation; a fresh request with the same prompt may produce another take. Keep the bundle if you need to enhance a reviewed draft. We have not tested this reproduction behavior and do not offer a local enhancement button.
What are the current BFL API prices?
Checked September 27, 2026: BFL lists t2v/i2v at USD 0.17 per output second for HD and 0.29 for FHD; continuation is 0.41 and 0.53 respectively. HD drafts are 0.06 per second for t2v/i2v and 0.12 for continuation. Higher-resolution rates are listed separately upstream. These are BFL provider/API prices, not PromptToVisual retail prices or credit conversions.
Budget a workflow, not only one clip
At the listed rate, a ten-second HD t2v full render is USD 1.70 before any other separately charged work. Budget drafts, retries and final enhancement separately, and record the provider’s settled cost instead of assuming a preview estimate is the invoice. A low unit price says nothing about how many attempts meet your brief.
Implementation checks before switching an existing pipeline
Map input fields and supported media for each mode before sending a job. Handle submission, polling, terminal failures and downloads explicitly. Store the accepted output and its settings so a later request can be compared. Check availability and your account limits in BFL rather than sending an old provider payload unchanged. Review the official API reference as the next step.
Sources and next steps
Source check: September 27, 2026. Official product claims and provider contracts are distinguished. No PromptToVisual video generation or benchmark is claimed.
BFL video modes and resolution →
Use the official source links to evaluate the external service. PromptToVisual does not provide video generation.