Generative media is the creativity layer of AI. It needs creative judgement to decide what is worth making, and production control to make the same thing twice.
In practice, it’s iteration, not a single prompt: create rapidly, compare concepts side by side, and keep a human in the loop for approval before anything ships. From there the work moves into production pipelines, combining models and modalities. Training custom models when needed and knowledge of model capabilities is critical.
Generative media is infrastructure, not a headline. Image and video have been the standard output. However, the new unit of creation is navigable worlds. A key distinction is whether the output is delivered in realtime or in batch.
The operations people actually reach for are unglamorous and specific: object removal, multiple angles, zoom, outpainting and virtual try-on.
There is no best model. What works is the smallest combination of fast, focused models trained on the exact task, chained well.
Source: fal.ai, The State of Generative Media, Volume 2
https://fal.ai/gen-media-report-volume-2/