promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_

The cost math — when AI video is viable and when it bankrupts you — step 1 of 6

The unit economics — cost per second across the 2026 lineup

Before anything else, the bare prices. This is API price per second of generated video, before retakes are factored in.

The table

Model$/sec60-sec spend (raw)
Veo 3.1 Lite (720p+audio)$0.05$3
Runway gen4_turbo I2V$0.05$3
Omni 1.1 Flash (720p, token-billed)~$0.10~$6
Veo 3.1 Fast (720p+audio)$0.10$6
Sora 2 (720p)$0.10$6
Runway gen4.5$0.12$7.20
Kling 3.0 (standard)$0.084$5.04
Kling 3.0 (Pro, video input)$0.168$10.08
Sora 2 Pro (720p)$0.30$18
Veo 3.1 Standard (with audio)$0.40$24
Sora 2 Pro (1024p)$0.50$30
Sora 2 Pro (1080p)$0.70$42

Sora 2 / Videos API is removed 2026-09-24. Those rows are the raw rate through that date. A Sora job is 16-20 seconds; a minute is three extends. The $30 at 1024p is 60 × $0.50, not one call. Moderation failures still get billed.

The four tiers

The lineup separates cleanly into four price tiers:

  1. $0.05/sec — the cheap tier. Veo 3.1 Lite (720p+audio), Runway gen4_turbo I2V. A 60-second piece at raw price runs $3. This is where high-volume work (performance marketing, social, B-roll) has to live. The "$3-5/finished-minute including retakes" headline only holds if you stay on Lite and the retake rate is about 1.

  2. ~$0.10/sec — the default-quality tier. Omni 1.1 Flash (~$0.10/sec token-billed, ~$6/min), Veo 3.1 Fast at $0.10/sec, Sora 2 base at $0.10/sec through 2026-09-24. Roughly $6 for a 60-second piece at raw price.

  3. $0.12-0.17/sec — the value-flagship tier. Runway gen4.5 at $0.12/sec, Kling 3.0 Pro at $0.168/sec.

  4. $0.40/sec — premium Veo. Veo 3.1 Standard. $24 per 60-second piece. Same native-audio pass as Lite and Fast — speech, ambience, music in one shot — at the top of the Veo quality ladder. Used for talking-head / narrated content where the Veo look and the lip-sync matter. Omni also generates an audio track by default (~$0.10/sec); Standard is not the only model that ships sound. Sora 2 Pro at 1024p is $0.50/sec / $30 per raw minute, same caveat on the 2026-09-24 removal.

The 8x spread is the strategic question

Lite at $0.05 vs Standard at $0.40 is 8x. That spread is not "the same product at different prices." It's different products entirely. They serve different jobs.

If your project is "a 30-second product launch trailer for a Fortune 500 client," Lite is wrong. The frames will not hold up in a Super Bowl spot.

If your project is "60 short variations for a Facebook ad performance test," Veo 3.1 Standard is wrong. You'll burn $1,440 on the test set (60 clips × 6s × $0.40 × 2.5x retake) and learn the same thing $180 of Lite output would have told you.

Picking the wrong tier — in either direction — is the single most expensive mistake in 2026 video work. The next read covers the retake multiplier that makes all of these numbers worse.

A 1K still from gemini-3.1-flash-image is ~$0.067. An 8-second Lite clip is $0.40. Video seconds dominate. Stack the still for control.

The Higgsfield wrinkle

Higgsfield doesn't price per second. It prices in credits, and the credits buy time on the underlying base models (Veo 3.1, Omni, Kling 3.0, Sora 2 through 2026-09-24). Read the current plan page before you treat a subscription as cheaper than the raw API.

The credits are usage-capped and the camera-control feature is the actual product. Don't subscribe just for arbitrage — subscribe because you want the deterministic shot vocabulary.

What this gives you

The cost-per-second table is the floor of the planning math. Step 2 turns this floor into an actual budget by adding the retake multiplier — the "generate 10, pick 1" reality that makes every raw price 2-10x more expensive than the table suggests.