Generative media expands from models into orchestration systems
Black Forest Labs is extending one system across images, video, audio and robotics while Runway automates selection among generative-media models; the announcements show capability generation and model routing emerging as separate system layers.
Artificial Intelligence··Morning
One model reaches across more media types
Black Forest Labs says FLUX 3 processes images, video and audio in one system, with its Video version producing clips up to 20 seconds with synchronized sound. Company tests preferred its outputs to Runway Gen-4.5 in 77% of comparisons and Luma Ray 3.2 in 93%. Video and Action remain in early access, and the same backbone is reportedly being adapted through FLUX-mimic for Audi production tasks. Neither the comparisons nor the robotics trial has independent results.[1]
Another layer chooses among models
Runway's Media Router generates nothing itself; it selects models by cost, latency and quality, with price caps and provider lists as constraints. It returns the choice and rationale, errors when nothing qualifies and can preview a decision without generating. Runway, which released Gen 4.5 in December 2025, is thus positioning itself as an orchestration layer and single integration point across providers as well as a model maker. Its quality score still rests on the company's own evaluation.[2]
Capability and routing solve different problems
FLUX 3 extends generation from images into video, audio and robotics, while Media Router organizes access to such capabilities by quality, speed and cost. One works at the model layer and the other at the selection layer. Because both rely on internal evaluation, neither establishes general superiority. The signal to watch is whether blinded common-prompt tests show routing preserving quality while lowering cost or latency, and independent FLUX 3 comparisons confirm the performance claims.[1], [2]
Related columns
For more information on this topic, you can read the related columns.
