Black Forest LabsBlack Forest Labs·🎬 Video Generation

Flux 3 Extend

anonymized
Try on Venice.ai β†—
Quick reference
Flux 3 Extend β€” TLDR
  • - 🎬 Continues an existing clip beyond its final frame as a video mode.
  • - 🧠 Part of FLUX 3, BFL's first multimodal image-video-audio model.
  • - πŸ”Š Audio generated with the frames: speech, lipsync, effects, ambience.
  • - πŸ“ Up to 20 seconds, HD or FHD (1920Γ—1088 16:9), 24 fps.
  • - ⚑ Draft mode returns fast, cheaper previews before full-quality renders.
  • - πŸ”§ Shares one video endpoint with text-to-video and image-to-video modes.
  • - πŸ†• Adds video and audio, absent from the image-only FLUX.2 family.
  • - 🏒 Black Forest Labs, Freiburg, Germany; released 2026.
πŸ’° Pricing
β€”
πŸ“… On Venice since
Aug 4, 2026
73 days ago
Provider

Black Forest Labs is a generative AI company based in Freiburg im Breisgau, Germany, founded by former members of Stability AI. The lab is best known for developing the Flux family of text-to-image models, which generate images from natural language prompts…

Read full profile β†’
9 models on Venice
4 video Β· 3 image Β· 2 inpaint
Since Nov 25, 2025

About this model

Flux 3 Extend is the video-continuation path of Black Forest Labs' FLUX 3 model. Rather than starting from scratch, it takes an existing clip as input and generates further footage from it. On BFL's API it is one mode among several exposed through a single video endpoint, alongside Flux 3 text-to-video, Flux 3 image-to-video, and keyframe-pinned generation as used by Flux 3 First Last Frame.

The headline shift versus prior Flux generations is modality. The FLUX.2 line β€” Flux 2 Pro, Flux 2 Max and its editing variant β€” is image-only, built for photorealistic stills, typography and multi-reference control. FLUX 3 is described by BFL as a single multimodal model trained across images, video and audio, producing video with synchronized sound in the same pass rather than dubbing it afterwards, including multilingual speech with lipsync, sound effects and ambience.

Practical limits are documented: clips up to 20 seconds at HD or FHD resolution (1920Γ—1088 for 16:9) at 24 fps in one request, with result URLs expiring roughly two hours after a job completes. A draft mode returns a quick, lower-cost preview that can later be rendered at full quality, which suits iterative extension work.

Extend pairs naturally with still-image siblings such as Flux 3 and Flux 3 when a project begins with generated or edited frames and then grows into longer, multi-shot sequences.

This About section is AI-generated from public sources (Claude Opus 5), with no human editing. It may contain inaccuracies β€” verify critical details against the sources listed above.

Data sources: Venice API Β· HuggingFace Β· Wikipedia β€” enrichment updated 7h ago