FLUX 3 Video

FLUX 3 Video is Black Forest Labs' unified async video endpoint: text-to-video, image keyframes, continuation from an existing clip, and a draft → enhance workflow. OpenTryOn exposes it as flux3-video on video-generate (same BFL_API_KEY as FLUX 3 Image and FLUX.2). For the Max speed/quality tier (version=max), see FLUX 3 Max (flux3-max).

Prerequisites

BFL_API_KEY=your_bfl_api_key

CLI

# Text-to-video
opentryon video-generate --model flux3-video \
  --prompt "Runway walk under soft key light, camera tracking" \
  --aspect-ratio 9:16 --duration 8 --resolution fhd

# Image-to-video (optional end frame)
opentryon video-generate --model flux3-video \
  --prompt "She turns slowly toward camera" \
  --image data/model-1.jpg --end-image data/model-2.jpg

# Continue from an existing clip
opentryon video-generate --model flux3-video \
  --prompt "They step into a crowded night market" \
  --video outputs/prior_clip.mp4

# Draft preview, then full-quality enhance (use draft_cache URL or file from the draft result)
opentryon video-generate --model flux3-video \
  --prompt "Fox running through dawn mist" --draft
opentryon video-generate --model flux3-video \
  --draft-cache path/to/draft_cache.bin --resolution fhd

# Dry-run
opentryon video-generate --model flux3-video --prompt "test" --dry-run

Notable flags

Flag	Default	Notes
--aspect-ratio	auto	21:9 … 9:21 or auto
--duration	auto	Whole seconds 5–20 (auto); v2v caps at 15
--resolution	hd	hd, fhd, qhd, uhd
--safety-tolerance	2	0 (strictest) through 7
--draft	off	Fast preview; save draft_cache from the API result
--no-audio	audio on	Silent clip
--video	—	v2v continuation from local path or URL
--draft-cache	—	draft_enhance from a prior draft bundle
--version	latest	latest or max (Max is also available as --model flux3-max)

Out of scope here: FLUX Video Edit at POST /v1/flux-tools/video-edit-v1 (separate BFL tool; not wired in OpenTryOn yet).

Python

from tryon.api import Flux3VideoAdapter

adapter = Flux3VideoAdapter()

mp4 = adapter.generate_text_to_video(
    prompt="Editorial runway walk, soft window light",
    aspect_ratio="16:9",
    duration=8,
    resolution="fhd",
)
with open("outputs/flux3_t2v.mp4", "wb") as f:
    f.write(mp4)

mp4 = adapter.generate_image_to_video(
    image="data/model-1.jpg",
    prompt="Gentle turn, fabric motion",
    duration=10,
)

MCP

After registry changes, restart the MCP server. Tools are generated from the registry:

• video_generate_flux3_video

API summary

• Submit: POST https://api.bfl.ai/v1/flux-3-video with x-key
• Poll: polling_url until status is Ready
• Result: result.sample — signed MP4 URL (or base64)

See the BFL API reference for the full request schema.