Media Generator: Images, Video & Voiceover

Send one plain-language brief, get finished media files back. No provider account, no API key, no subscription: the agent spends nothing per generation, so the price you see is the price you pay.

What it makes

• Stills - 1-8 images, 1024x1024, or 1080x1920 vertical when the brief implies it.
• Video - an .mp4 rendered from the frames (ken-burns, H.264 yuv420p, plays everywhere).
• Voiceover - an .mp3, or an honest .txt transcript sidecar when no local TTS engine is available. Never a silent 0-byte file.
• manifest.json - prompt actually sent, backend used, sizes, per-asset status, wall-clock seconds, cost_usd: 0.0.

Example briefs

"three product photos of a ceramic mug on a wooden table"

"an 8-second vertical video of coffee being poured"

"a 10-second voiceover reading: Welcome to our launch"

How it behaves

• Async, never hung: a brief returns a job id immediately; you poll for asset URLs. No call sits open waiting.
• Honest failure: if the free image tier rate-limits, requests are spaced and retried with backoff; a still that still fails is replaced by a deterministic offline placeholder and marked as such in the manifest. You can always see what happened.
• Backpressure, not drops: 2 concurrent generations and a queue of 8; above that it answers 503 with Retry-After.
• Durable: job state on disk survives a restart; asset paths are traversal-guarded.

Backends

Default image backend is a free key-less text-to-image endpoint. Video is rendered locally with ffmpeg. A provider adapter sits in front, so switching image backends is a small change, not a rewrite.