Best AI video model for draft clips: lowest per-second rows

The cheapest per-second row of all is grok-imagine-video-1.5 at $0.0125 a second, but it only animates an image (4-15 s, no text-to-video). For text-to-video drafts the cheapest is Gemini Omni Flash 1.1 at 360p, which bills $0.0375 a second. The next cheapest are wan-3.0, minimax-h3 and minimax-h3-max, all at 480p for $0.0625 a second. Those figures are each provider's list rate (fal, read from the Sume docs) multiplied by the documented 1.25 on every row. A 5-second draft is $0.1875 on Omni 360p and $0.3125 on a 480p row, before any rounding to cents.

What are the draft rates?

The Video Router doc publishes list rates for these rows. Sume bills list x 1.25. Seedance rows are priced per 1,000 video tokens, not per second, so they are left out of this table; read their pricing_skus from GET /v1/videos/models if you want to compare.

What does a cheap draft not tell me?

A draft on one family is not a preview of another family. If you draft on Omni and finish on Wan, the look can change, because each row is a different model. Draft on the row you will finish on, one resolution tier down: wan-3.0 480p before 1080p, minimax-h3-max 480p before 1080p. Omni 360p is the cheapest probe, but it only tells you what Omni will do.

Mind each row's floor. Omni will not run below 3 seconds, H3 starts at 5 seconds, and wan-3.0 is the only row that goes down to 2. If your draft is a 2-second hook, Wan is the only match.

How do I keep a draft loop from running up a bill?

Pin the model id, send duration and resolution explicitly, and cap the batch size in your own code. Submits reserve funds first; a balance below the reserve answers 402 insufficient_credits, so a runaway loop stops by itself when the wallet does. Do not rely on that as a budget control.