FLUX 3 generates video with audio — but you can't get it yet
Black Forest Labs launched FLUX 3, one model for image, video and audio up to 20 seconds. It's gated early access with no pricing. Don't rebuild your content pipeline yet.
Black Forest Labs — the lab behind the FLUX image models a lot of commerce teams already use for product shots — shipped FLUX 3, a single model that generates images, video, and native audio from one prompt. It's a real capability jump. It's also gated early access with no published price, no general API, and no open weights. That gap between "announced" and "usable" is the part worth planning around.
What actually happened
Per Black Forest Labs' announcement on July 23, 2026, FLUX 3 "jointly learns from images, videos, and audio within a unified architecture" rather than bolting separate modality models together. Concretely: video up to 20 seconds per generation, evaluated at 720p, with native audio generated alongside the video — including multilingual dialogue. It supports text-to-video, image-to-video, video-to-video, and keyframe-driven generation. The image side handles multiple aspect ratios and high-accuracy text rendering in several languages.
BFL also announced FLUX-mimic, a robotics collaboration with mimic robotics that puts the FLUX 3 backbone behind dexterous manipulation, with early testing at Audi.
The benchmark claim: FLUX 3 was preferred in 77% of matchups against Runway Gen-4.5 and 93% against Luma Ray 3.2. Worth flagging — those are BFL's own preliminary comparisons, not third-party evals. VentureBeat's coverage makes the constraint explicit: this is a limited release. API access and private weights are rolling out in phases, by application. No pricing has been published.
Why it matters for your business
A 20-second product video with usable audio, generated from a still and a prompt, is genuinely a different cost structure than hiring a crew. If that lands at a sane price, the economics of a small brand producing weekly video content change completely.
But "announced" is not "in your stack." We've watched too many teams reorganize a content workflow around a model in gated preview, then sit on a half-built pipeline for four months waiting on an access email. The generation model is the easiest piece to swap and the worst piece to build your process around.
Build the parts that don't depend on which model wins: a clean product-asset library with consistent lighting and angles, prompt and brand-voice templates under version control, a review-and-approval gate before anything publishes, and a generation step behind one interface so swapping FLUX 3 for Kling or Veo is a config change. Then when access opens and pricing lands, you're testing a provider — not rebuilding a workflow.
Key takeaways
- FLUX 3 launched July 23, 2026: one multimodal model for image, video, and native audio, up to 20 seconds per generation, evaluated at 720p
- It's gated early access by application — no general API, no published pricing, weights limited to selected partners
- The 77% and 93% win rates against Runway Gen-4.5 and Luma Ray 3.2 are BFL's own preliminary evals; test on your own catalog before believing them
- The operator move: build the asset library, prompt templates, and approval gate now; keep the generation model behind a swappable interface
Want AI content production that survives the next model launch? We build creative pipelines where the generation model is one swappable step — your assets, prompts, and approval gates stay yours. See what we've built or talk through your content workflow.
Sources: Black Forest Labs, VentureBeat.
- #ai-video
- #flux-3
- #content-production
- #product-photography
- #vendor-risk
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Agentic AI spending: $944B now, $3.35T by 2030 — read it right
A new PHD/WARC forecast puts agent-facilitated consumer spending at $944B in 2026, rising to $3.35T by 2030. That's 3.8% of consumer spend. Here's the operator read.
Read itInstacart buys Arpalus: who owns your shelf data?
Instacart acquired computer-vision startup Arpalus to turn 600,000 shoppers into a shelf-intelligence sensor network. If they run in your stores, they know your inventory better than you do.
Read it