Video Generation Goes GA — On Every Plan

April 17, 2026

Partially superseded — April 23, 2026

On April 23, Video Lab moved to Runway Gen-4.5. Several of the operations described below now return a "temporarily unavailable" message with migration guidance. Read the Runway note for the current capability surface.

Video generation is now available on every plan, including Community. It runs on our own GPU, one render at a time per user, with a real queue you can see — not a fake progress bar.

What's new

• Available on every plan. Community can generate video clips for the first time.
• Honest progress. The status line shows the actual phase the render is in — queued, generating (with step count), upscaling, decoding — and a real ETA when you're waiting.
• Two quality modes. A fast distilled mode (~30 seconds on the GPU) for iteration, and a higher-fidelity production mode (~2 minutes) for finals.
• Hard caps surfaced clearly. 5 seconds per clip, 1280×768 max resolution. If you ask for longer, the assistant tells you it clamped instead of silently truncating.
• One render per user at a time. Additional requests wait in a real FIFO queue with a visible position.

When you'd notice it

• You're on Community and want to make a short social clip — you couldn't before; you can now, with the same progress visibility everyone else gets.
• You kick off a render, walk away, and come back to a status that actually tells you where things stand: "Generating frames — step 24 of 40, 1m 12s elapsed."
• You ask for a 30-second clip. Alfrada clamps it to 5 seconds, says so, and points you at the Polished Narrated Video playbook for stitching multiple clips together.

Heads up

During an active render, the GPU temporarily evicts the local language and embedding models. Background tasks (like title generation) briefly route to a cloud model and warm back up once the render finishes.