Black Forest Labs Launches FLUX 3: First End-to-End Jointly Trained Model for Image, 20-Second Video, and Audio

What happened

Black Forest Labs (BFL), creators of the FLUX model series, launched FLUX 3 on July 23, 2026 — a jointly trained multimodal model capable of generating images or combined video/audio clips up to 20 seconds from a single prompt. The model launched in limited access.

Context and impact

FLUX 3 is architecturally distinct from prior generative pipelines that assemble separate image, video, and audio models. End-to-end joint training across modalities is considered a meaningful step toward truly unified multimodal generation. BFL directly competes with Sora (OpenAI), Veo (Google), Runway, and Kling — but with a single unified model rather than multiple specialized ones.

Details

• Video: up to 20 seconds from a text prompt
• Training: end-to-end joint training (not an assembled pipeline of separate models)
• Modalities: image, video, audio — all in one model
• Availability: limited access at launch, broader access planned
• Black Forest Labs: German startup, creators of FLUX 1.x and FLUX 2.x series
• Competitors: Sora, Veo 3, Runway Gen-4, Kling