fal Assets is live.
Everything you generate on fal lives in one library now: images, video, audio, 3D. Semantically searchable, and instantly reusable as input across Sandbox, Playground, and Workflows.
No re-uploading. No filename hunting. Just "the angry one" or "the sunset
Grok Imagine Video 1.5 just got a major update on fal
Reference-to-Video: up to 7 reference images — every character stays consistent (720p)
Text-to-Video: now available on 1.5, up to native 1080p
Image-to-Video: native 1080p support
MiniMax H3 takes direction like a camera crew. A slow dolly into an eye that closes on cue, an eavesdrop played with just a head tilt, a tracking shot that matches her run. The trick is stating where the subject, camera, and background are each moving.
Stills from Seedream 5.0
Shaky one-handed phone camera, a real kitchen at dusk, and a hand-drawn creature living in it. MiniMax H3 generated the live-action and the animation in one pass, down to the faucet drips and the creature's tiny voice.
Text-to-video is getting weirdly good at "accidental" footage.
A late-night laundromat, flickering fluorescents, delayed autofocus, and a hand-drawn apparition drifting between the washers. MiniMax Hailuo-03 generated all of it from a single prompt. No reference images, no
