AI Video & Image-to-Video API Pricing Compared (2026)

October 31, 2026 · Updated November 6, 2026 · Prices checked November 2026 · 10 min read

Per-second AI video and image-to-video API rates run from $0.02 to $0.43 and up, depending on the model, resolution, whether audio is included, and whether the model adds a start-frame fee. There's no single number because several major vendors don't publish one: their own pricing pages are either rendered in JavaScript with nothing to read in the page source, or billed by the token instead of the second.

Why there's no single answer

Checking vendor pricing pages directly for this article turned up a real problem: Kling AI's consumer pricing page has no pricing table in its page source at all — it loads entirely client-side — and Google prices Gemini Omni Flash by output tokens, not seconds. Neither gives you a per-second number to quote without converting. On top of that, the rate itself moves with resolution (360p to 4K), with whether audio is generated, and with whether you supply a start-frame image or reference video — several models charge extra for reference inputs.

The most complete rate card we could find comes from Runway's own developer API, which runs its own models alongside many others — Veo, Kling, Seedance, Wan, Grok Imagine, Gemini Omni Flash and MiniMax — through one account. The tables below use those published credit rates, converted at Runway's stated $0.01 per credit. These are what Runway itself charges to call each model, not necessarily what you'd pay going directly to each vendor, which may be lower, higher, or simply not published anywhere in dollars.

Per-second rates, cheapest to priciest (checked November 2026)

A useful spot-check: Google's own Gemini API pricing page lists Gemini Omni Flash's video output at $17.50 per million tokens, at roughly 5,792 output tokens per second of 720p video — about $0.10/s, in line with Runway's listed gemini_omni_flash_1.1 rate of 10 credits ($0.10) for the same resolution. When a vendor's own number and a reseller's number agree like that, it's a good sign both are current.

Image-to-video API pricing: start-frame fees and input rules

If you are building an image-to-video pipeline — animating a product photo, a storyboard frame or an AI-generated still — the per-second rate is only part of the formula. Across the models on Runway's developer API (checked November 2026), image-to-video billing falls into three patterns:

• Image-to-video only, priced below text-to-video: Runway's gen4_turbo ($0.05/s) requires an input image and cannot run from a text prompt alone, making it 58% cheaper per second than gen4.5 ($0.12/s), which accepts either text or an image.
• Per-second rate plus a flat start-frame or reference-image fee: grok_imagine_1_5_lite, grok_imagine_1_5, gemini_omni_flash and gemini_omni_flash_1.1 each add 1 credit ($0.01) per start-frame or reference image on top of the per-second video charge. MiniMax's hailuo3 adds 2 credits ($0.02) per reference image or keyframe (including an image-to-video start frame).
• Start and end frames included in the per-second rate ($0 surcharge): h3_max (first and last frames), wan3 and wan3_prime, kling2.5_turbo_pro (optional last frame), kling3.0_standard / pro / 4k, klingO3, kling_4, and ByteDance's seedance2 and seedance2_5 do not add a per-image fee. On seedance2_5, reference images and reference audio are explicitly free; its +$0.10 to +$0.34/s input surcharge applies only when you pass a reference video.

If your pipeline generates the start frame first with a text-to-image API call before animating it, add the image model's per-image charge once per attempt. On the same Runway rate card (November 2026), generating a start frame costs $0.01 on muse_image, $0.02 on gen4_image_turbo or seedream5_flash, $0.04 on gemini_image3.1_flash_lite, seedream5_lite or 1K nano_banana_2.1, $0.05 on 720p gen4_image or gemini_2.5_flash, and $0.20 on 1K/2K gemini_image3_pro (Nano Banana Pro). Even with a $0.02–$0.05 generated first frame, a two-step text-to-image + gen4_turbo 5-second clip ($0.27–$0.30 total) costs half as much as generating a 5-second clip directly from text on gen4.5 ($0.60).

How video generation API pricing scales with usage

Unlike traditional cloud storage or bandwidth, self-serve video generation APIs rarely offer automatic volume-discount tiers on pay-as-you-go usage: Runway's developer portal charges a flat $0.01 per credit whether you generate 50 seconds or 50,000, and Google's Gemini API charges a flat $17.50 per million video output tokens on gemini-omni-1.1-flash. Instead, your effective cost per clip scales up or down across four levers:

• Resolution step-ups (2× to 10× per second): Moving from draft resolution to HD or 4K is the steepest multiplier. wan3 doubles at each step ($0.05/s at 480p → $0.10/s at 720p → $0.20/s at 1080p); gemini_omni_flash_1.1 rises nearly 9× from 360p ($0.034/s) to 4K ($0.30/s); and seedance2 rises more than 10× from 480p ($0.14/s) to 4K ($1.50/s). On seedance2_5, passing draft: true renders a 480p preview at $0.20/s so you only pay the $0.68/s 1080p rate on prompts you actually enhance (the preview and the 1080p enhance call are billed separately).
• Audio and delivery-format surcharges (+33% to +100%): Models that price audio separately charge substantially more when audio is enabled — veo3.1 doubles from $0.20/s to $0.40/s, veo3.1_fast rises 50% from $0.10/s to $0.15/s, and kling3.0_standard / kling3.0_pro add $0.04–$0.05/s (note that audio defaults to true on Runway's Kling 3.0 endpoints unless you explicitly pass audio: false). Requesting ProRes or PNG sequences on gen4.5 or aleph2 adds $0.05/s, while 10-bit and HDR profiles add $0.20/s ($0.40/s above 4 megapixels).
• Per-generation minimum floors on short clips: Several ByteDance and editing endpoints enforce a minimum charge per call regardless of duration: 28 credits ($0.28) on seedance2_mini, 44 credits ($0.44) on seedance2_fast, 56 credits ($0.56) on seedance2 and aleph2, and 80 credits ($0.80) on seedance2_5. A 4-second 480p clip on seedance2_5 would be 80 credits by duration (4 × 20), hitting the floor exactly, whereas a shorter clip at 480p still pays the full $0.80 minimum.
• Model-tier routing and enterprise commitments: At scale, teams cut unit costs by routing previews and background B-roll to lighter variants (seedance2_mini at $0.07/s vs seedance2 at $0.14/s at 480p; veo3.1_fast at $0.10–$0.15/s vs veo3.1 at $0.20–$0.40/s) or by moving to reserved capacity such as Google Cloud's Provisioned Throughput for GA Veo 3.1.

What changed at Google

Google shut down the Veo 3.1 Preview on the Gemini API on October 22, 2026 —veo-3.1-generate-preview, veo-3.1-fast-generate-preview and veo-3.1-lite-generate-preview are all gone, with no replacement preview endpoint. Google's migration guidance points to Gemini Omni Flash (gemini-omni-1.1-flash), now generally available on the Gemini API's paid tier, or to the GA models on Google's Gemini Enterprise Agent Platform, where Veo 3.1 and 3.1 Fast are generally available. That means moving the integration off the Gemini API. If you would rather keep one integration for several models, Runway's table above still lists Veo 3.1 at $0.40/s with audio, $0.20/s without.

Budgeting a month of 8-second clips

Using clips × seconds per clip × price per second at three points in the range above: the cheapest 720p rate ($0.03/s, Grok Imagine 1.5 Lite), a mid-market API rate ($0.12/s, Runway Gen-4.5), and a premium tier ($0.40/s, Veo 3.1 Standard with audio).

None of these totals include retries. If you expect to keep roughly 1 in 2 attempts, double the column you're budgeting against.

If you'd rather not integrate an API

Our Video Generator is a browser form, not an API: pick from five models — Seedance 2.0 (default), Kling 3.0, Veo 3.1, Grok Video and Happy Horse 1.0 — a length and a resolution, and submit a text prompt. It runs text-to-video only (there is no start-image upload on /video; still images are generated separately on Image Generator for 8 credits per image, and existing 2–15 second video clips can be continued on Video Extend at 10 credits per added second or restyled on Video Edit for a fixed 180 credits per 15-second output). Video generation charges a flat 10 credits per second whatever the model or resolution, with no account quota, SDK or key to manage. We don't offer a developer API today — that's on our roadmap, not something available yet.

On one-time packs, whose credits never expire, that's about $0.33/s at our cheapest pack rate (the $59 Power Pack for 1,800 credits) up to $0.45/s on the $9 Starter Pack (200 credits); the table uses each pack's per-credit price, and packs are bought whole. The monthly Pro plan ($49 for 2,000 credits) comes to $0.245/s, but plan credits reset each month. All of these sit in the upper half of the per-second range in this article, not the cheapest. The 20 free credits granted at sign-up don't cover even the shortest clip any of our five video models supports (Kling 3.0's 3-second minimum costs 30 credits), though they do cover two 8-credit images on Image Generator. What the flat rate buys instead of a lower number: one balance across five video models plus our image, subtitle, upscale and edit tools, no per-model or per-resolution price tiers to track, and no watermark on any output, free-credit or paid. See plans and packs or the live tool pricing for current rates. For Veo-specific detail beyond this article — retry budgeting, endpoint IDs, our own Veo option — see Veo 3 & 3.1 pricing; for consumer app plans alongside per-second rates, see AI video pricing and app plans; and for a quality-first comparison across models, see the best AI video generators of 2026.

AI video API pricing questions