AI Video Price Index: What 100 Clips Cost

A budget for 100 eight-second AI video clips ranges from $24 to $320 across the eight direct API options in this snapshot. Costs use a shared job definition; the model versions, capabilities and quality are not equivalent. Seedance is a token estimate with an assumed frame rate. Prices below are generation charges, not a promise that every output meets your brief.

Verified October 5, 2026. Magic Hour publishes this index and sells access to some models. We use official developer rate cards, not Magic Hour retail credits. No paid generations were run for this price snapshot.

100 Clips With the Same Output Settings

Defined job: 100 first-frame image-to-video outputs, requested duration eight seconds, landscape 1280 × 720 pixels, 24fps and native audio enabled. Use fixed duration rather than automatic duration. Google rows use the preview IDs documented in the downloadable dataset; access may depend on account and region.

Check current API access before choosing a row

Availability check, October 10, 2026: Google’s Gemini API deprecation schedule lists October 22, 2026 as the earliest shutdown date for veo-3.1-lite-generate-preview, veo-3.1-fast-generate-preview and veo-3.1-generate-preview, the three Google IDs in this dataset. Google recommends gemini-omni-1.1-flash. This notice concerns those Gemini API preview endpoints, not every hosted Veo product. Before building a new workflow, verify the replacement’s current access, settings and rate, then rerun the matched calculation. The October 5 table, chart and downloadable CSV remain a historical snapshot; their rates and capabilities must not be transferred to the replacement model.

Figure 1. Calculated listed generation charges for the defined job. Eight documented options, not eight paid runs. The hatched Seedance bar is a token-billing estimate. Sources and exact model IDs appear in the table and CSV.

These are standard listed direct API rates, with no enterprise, prepaid-package, offline, promotional or batch discount applied. Settings were checked against LTX I2V documentation, Google Veo specifications and BytePlus task parameters. This is a selected reference set, not an exhaustive market ranking or a claim that the least expensive option is best.

Why Token-Billed Video Needs an Estimate

For the seven per-second rows, the arithmetic is 100 outputs × 8 seconds × listed USD rate. Seedance 1.5 Pro uses standard online inference with audio enabled and draft mode disabled. Its published estimate is output seconds × width × height × frame rate ÷ 1,024.

For the defined job, 8 × 1,280 × 720 × 24 ÷ 1,024 gives 172,800 video tokens per output. At $2.40 per million, 100 outputs calculate to $41.472, displayed as approximately $41.47. The 24fps value is assumed for this token estimate, not a guaranteed Seedance request parameter or a verified returned output. BytePlus bills actual returned usage.completion_tokens; actual duration and token usage can differ. This row is not an observed invoice.

What Does the 1080p Upgrade Add?

Keeping 100 outputs, eight seconds and audio unchanged, the seven per-second rate cards imply a 1080p premium ranging from 0% to 100%. That is a resolution-price comparison within each listed model, not a quality score.

MiniMax: A Separate Six-Second, 768p Cohort

The MiniMax legacy rate card defines these prices per six-second 768p generation. Native audio is not documented in the cited legacy I2V specification. Duration and resolution differ; native audio equivalence is not established. These rows are not in the primary chart or its $24–$320 range. Newer H3 models are not covered in this legacy cohort.

Download the Data and Reproduce the Arithmetic

Download the dated CSV or open the fixed October 5 dataset and methodology. The 18 rows cover eight primary options, seven 1080p configurations and three separate legacy MiniMax options. Each row includes exact model ID, billing unit, settings, formula, source links and an unrounded calculated cost.

Recompute per-second rows as outputs × duration_seconds × rate_usd; token rows as outputs × duration_seconds × width × height × fps ÷ 1,024 × rate_usd ÷ 1,000,000; generation rows as outputs × rate_usd. Round only the final displayed amount. The CSV and original chart are reusable with Magic Hour attribution under CC BY 4.0.

Methodology and Limits

This is desk research and arithmetic from public rate cards. The eight-option coverage set is not a random sample, user study or controlled model experiment. Prices are USD with no currency conversion. We checked documented settings, not account-level availability, uptime, throughput, actual bills or acceptance rates.

Exclude input-image creation, taxes, storage, delivery, editing, upscaling and extra attempts. A technically successful output can still be unsuitable and billed. Google lists billing only for successful video generation; other providers have their own rules. Do not assume one refund policy applies to every row, or apply the general Gemini batch discount to Veo without a Veo-specific rate.

The October 5 snapshot establishes a repeatable format, not a historical price trend. Future editions should retain dated downloads and compare only matched model IDs, modes, dimensions, audio settings and billing tiers. Intended editorial cadence is monthly and after a verified rate change; the page does not update automatically.

For a Hailuo-specific purchase, see Hailuo 2.3 pricing. For spoken audio, the ElevenLabs pricing guide explains credits and voice-cloning access separately from video-generation charges.

For detailed plan and credit breakdowns, review Pika pricing, Runway pricing, Runway Gen-4.5 pricing, Seedance 2.0 pricing, HeyGen pricing, VEED pricing, and the Sora migration guide. Each page separates subscription invoices, credits, API billing and accepted-output costs.

For the other major consumer platforms, see Kling pricing, Luma pricing, PixVerse pricing and Invideo pricing. Compare the actual invoice and settings rather than treating different providers’ credits as equivalent. The Vidu AI review includes current Q3 direct-API rates and a cost-per-accepted-clip example.

A defined job: fifty accepted clips

Suppose the deliverable is fifty accepted six-second 768p clips and you budget three attempts per accepted clip. That means 150 generations. The published Hailuo 2.3 Fast rate gives a hypothetical $28.50 generation bill; Hailuo 2.3 gives $42.

The three-attempt allowance is an assumption, not a measured success rate. At the listed $0.19 Hailuo 2.3 Fast generation rate, the same fifty-clip job would cost $9.50 at one attempt per accepted clip, $19 at two, or $28.50 at three. These scenarios isolate generation charges; they do not establish equivalent quality or a future acceptance rate.

A worksheet for your own video budget

If no output meets the brief, report zero accepted outputs rather than a cost of $0 per usable clip. The commercial image-to-video dataset records charged credits and technical completion; acceptance costs remain unknown until the outputs are scored against a buyer's requirements.

How should you price a multi-model platform?

Use the platform’s quoted cost for the selected model and settings. Its credit unit, discounts, and included workflows may differ from the model developer’s direct API. Magic Hour’s API documentation and pricing page are the starting references for its own service.

For subscription purchasing, see the AI video generator pricing guide. Keep the full invoice and annual commitment visible when allocating subscription spend to outputs.

What costs are easy to miss?

• Additional attempts that do not produce an accepted clip.
• Audio, higher resolution, or separate editing and enhancement operations.
• Application storage and delivery outside generation charges.
• Unused subscription capacity and credits that expire.
• Team seats, taxes, and renewal-price changes.

Why is Sora absent from the active comparison?

OpenAI lists September 24, 2026 as the Sora API discontinuation date, after ending the consumer app and web experience in April. See OpenAI’s notice. Remaining historical rates do not justify recommending a new long-term integration.

How should this index be cited?

Cite Magic Hour AI Video Price Index, verified October 5, 2026, and the official provider rate source. Include model, output settings, billing unit and estimate caveat. Use the fixed dated CSV for reproducibility. Do not cite these calculated charges as a quality ranking, observed invoice or cost per accepted output.

Reuse this pricing chart with its context

Reuse the original first-party pricing chart in an article, presentation or procurement note, with this context: “Magic Hour AI Video Price Index, October 5, 2026: calculated listed API cost for 100 eight-second 720p first-frame image-to-video outputs with native audio.” Link to this index and its methodology so readers can verify the rates, settings and exclusions.

Keep the date and defined job visible when embedding the image. The models do not have identical capabilities, token estimates can differ from actual charges, and the chart is not a quality test or cost-per-accepted-output benchmark. Treat it as the dated snapshot shown here rather than a claim about every current plan or generation.

Continue learning

Continue learning: use the complete AI video workflow to test shortlisted models, then diagnose weak outputs with the prompt troubleshooting guide.