AI Video Pricing Report: Second-Half 2026 Guide

AI video pricing changed enough in the first half of 2026 that many older comparisons are now wrong. HeyGen moved to a broader credit system, Google introduced three Veo 3.1 price tiers, ByteDance published per-second API pricing for its Seedance 2.0 family, Runway changed the seconds included with its plans, Adobe added temporary unlimited-generation offers, and OpenAI discontinued the standalone Sora web and app experiences. The churn continued right into this edition: in the last week of July, ByteDance launched Seedance 2.5 and MiniMax launched its H3 flagship within roughly the same week.

This report converts the public price and usage rules of major AI video brands into a comparable cost per second of output. It was first reviewed on July 11, 2026 and updated for August on August 3, 2026. It uses official pricing or help pages — with screenshots of each source captured on the review dates — and will be updated monthly through the second half of 2026.

Quick answer: For focused lip-sync output, FreeLipSync Starter has a theoretical floor of $0.0014 per second when all 20 monthly Pro Videos use the full three-minute allowance. At a more practical average of one minute per video, the cost is $0.0042 per second. Among the large general-purpose platforms with transparent usage math, HeyGen Avatar IV/V starts around $0.016 per second, Pika 720p around $0.046 per second, Google Veo 3.1 Lite at $0.05 per second, ByteDance Seedance 2.0 Mini at $0.08 per second at 720p (and $0.04 at 480p), and Adobe Firefly around $0.10 per second on its standard credit allowance. The late-July flagships enter above that budget floor: MiniMax H3 at $0.08 per second (768p) and $0.13 (2K), and ByteDance Seedance 2.5 at about $0.10 (480p) and $0.23 (720p).

This is a pricing report, not a claim that every model produces equivalent work. A focused lip-sync tool, an avatar platform, and a cinematic text-to-video model solve different problems. Compare cost only after choosing the workflow you actually need.

---

What's New in the August 2026 Update

The last week of July brought two flagship launches, so this edition adds both models with their official launch-week API prices:

• ByteDance Seedance 2.5 (launched July 31). Two things matter. Capability: a native 30-second single-shot mode, up to 50 multimodal reference inputs, up to 4K 10-bit output in ByteDance's own apps, and native audio with phoneme-level lip sync. Price: the published API rates cover only 480p and 720p and land at roughly 1.5x Seedance 2.0 — about $0.10 per second at 480p and $0.23 per second at 720p.
• MiniMax H3 (released in late July). Two things matter here as well. Capability: a new-generation multimodal flagship that generates 4-15 second clips at 768p or 2K with audio generated alongside the video, driven by mixed reference inputs (up to 9 images, 3 video clips, and 3 audio clips). Price: pay-as-you-go only at $0.08 per second (768p) and $0.13 per second (2K) — the 2K rate undercuts Seedance 2.5's 720p rate, and MiniMax's video packages do not support H3 yet, so there is no bulk-discount path.

Every other platform was re-verified on August 3: FreeLipSync, HeyGen, Google Veo 3.1, Seedance 2.0, Runway, Adobe Firefly, and Pika prices are all unchanged since July 11, and the cheapest API anchors remain Seedance 2.0 Mini ($0.04 per second at 480p) and Veo 3.1 Lite ($0.05 at 720p). A dedicated head-to-head of the two launches is in Seedance 2.5 vs MiniMax H3: Launch Pricing Compared.

August 2026 Price Snapshot

All prices are USD before tax. Annual-plan rows use the monthly equivalent displayed by the provider. Subscription calculations assume the full included allowance is used before expiry. Promotional prices are excluded unless the provider clearly states that the offer is still active on the review date. Seedance 2.5 is billed by token; its rows use the provider's official five-second 16:9 reference examples, the same basis as the Seedance 2.0 rows.

---

How We Chose the Platforms

The report is intentionally smaller than a directory-style "50 best tools" article. A platform enters the main table only when it meets these rules:

1. It is a large, recognizable AI or creative-software brand.
2. It offers an active video, avatar, or lip-sync workflow relevant to creators.
3. Its official public pages expose enough pricing and usage data to reproduce the math.
4. Its role is meaningfully different from the smaller specialist lip-sync products that compete directly for the same narrow query.

HeyGen is included because it is the most recognizable avatar-platform comparison and now publishes detailed credit consumption. ByteDance's Seedance family entered in July because it has transparent per-second API pricing on both its Chinese (Volcengine) and international (BytePlus) platforms, and because Runway and HeyGen both resell Seedance access inside their own plans; the August edition adds the newly launched Seedance 2.5 on the same basis. MiniMax enters this edition for the same reason: its Hailuo line is one of the most recognized AI-video brands, and the new H3 flagship publishes direct per-second prices on the official MiniMax platform. Smaller direct lip-sync competitors are excluded. There are no affiliate links in this report.

Methodology: Three Numbers, Not One

Monthly plan prices are easy to compare and often misleading. We calculate three different values where the product rules allow it.

1. Published cost per second

For an API with direct per-second billing:

Cost per second = provider's published per-second rate

2. Maximum-efficiency subscription cost

For plans with credits or counted videos:

Cost per second = monthly plan price / maximum seconds available from the included allowance

This is a theoretical floor. It assumes every credit is used and every counted output reaches the duration used in the calculation.

3. Practical reference cost

For FreeLipSync, we also show a 60-second average-output scenario. This avoids presenting the three-minute Starter limit as if every customer always generates a full three-minute video.

We do not assign a cost-per-second number when the provider does not publish enough information to reproduce it.

---

FreeLipSync: Focused Lip Sync at a Low Entry Price

FreeLipSync Starter costs $4.99 per month and includes 20 subscription Pro Videos every 30 days. A Starter account supports up to three minutes of audio per generation, but the effective price depends on the average length of those 20 videos.

The $0.0014 per second figure is therefore a maximum-efficiency floor, not a universal customer average. For a creator producing one-minute videos, $0.0042 per second is the more useful planning number.

FreeLipSync is also a narrower product than the cinematic generators below. It starts from an existing face photo or face video and synchronizes it to text, uploaded audio, recorded audio, or a cloned voice. It does not replace a scene generator when the job requires a new environment, complex camera motion, or a fully synthetic sequence.

The Free plan remains useful for evaluation: no account is required to generate, outputs have no watermark, and the free input limit is 20 seconds of audio or 133 text code points. Downloading requires sign-in, and free downloads use the low-resolution derivative unless an original-resolution unlock is available.

HeyGen: New Credit Pricing Changes the Avatar Math

HeyGen's current Creator plan costs $29 per month and includes 600 credits. Its official credit guide prices Avatar IV and Avatar V at 20 credits per minute.

600 credits / 20 credits per minute = 30 minutes
$29 / 1,800 seconds = $0.0161 per second

That equals about $0.97 per minute of Avatar IV/V output when the full balance is used. Custom Expressive Motion consumes 40 credits per minute, doubling the effective cost for that workflow.

HeyGen also prices other jobs differently. Full video translation with lip sync uses 5 credits per minute, while Avatar III uses 3 credits per minute. A single "HeyGen cost per second" is therefore incomplete unless the article names the exact feature being measured.

This is one of the biggest changes from older 2026 comparisons: the current Creator plan has 600 credits, not the older 200-credit premium pool used in many archived calculations.

Google Veo 3.1: The Clearest API Price Ladder

Google publishes Veo 3.1 prices directly per generated second, so no credit conversion is necessary.

A one-minute 720p clip therefore costs $3 with Lite, $6 with Fast, or $24 with Standard. These prices include generated audio. Google states that a failed generation caused by an audio-processing issue is not charged.

For second-half planning, the migration detail matters: Google marked Veo 2 and the earlier Veo 3 API model IDs for shutdown on June 30, 2026. New cost models should use Veo 3.1 rather than carrying older Veo pricing forward.

ByteDance Seedance 2.0: Aggressive Per-Second API Pricing

ByteDance publishes direct per-second prices for the Seedance 2.0 family on its international BytePlus ModelArk platform, using a five-second 16:9 text-to-video or image-to-video generation as the reference:

The same models are sold in China through Volcengine at closely matching CNY rates (Seedance 2.0 at 720p is ¥0.99 per second, about $0.14, with the 5-second reference video priced at ¥4.97):

Three details matter for budgeting. First, billing is token-based under the hood (the per-second figures are official estimates for standard 16:9 output), and only successful generations are charged. Second, Seedance 2.0 Mini at 480p is currently the lowest published API price in this report at $0.04 per second — below Veo 3.1 Lite's $0.05 — while Veo Lite remains cheaper at 720p ($0.05 vs $0.08). Third, Seedance is also resold inside other subscriptions: Runway's Pro plan lists Seedance 2.0 access, and HeyGen charges 60 credits per 720p Seedance video as an asset-generation feature, so heavy Seedance users should compare the native API price against bundled access they may already pay for.

ByteDance Seedance 2.5: The New Flagship Costs About 1.5x Seedance 2.0

ByteDance officially launched Seedance 2.5 on July 31, 2026 and published API pricing for it on both of its model platforms. The launch features are a native 30-second single-shot mode, support for up to 50 multimodal reference inputs, up to 4K 10-bit output, region-level partial editing, and — closest to this site's topic — native audio generation with phoneme-level lip sync.

The API price list tells a narrower story than the launch demos. On BytePlus ModelArk, the dreamina-seedance-2-5 token rates are published only for 480p and 720p outputs: $10.70 per million tokens without video input, and $6.40 with video input. BytePlus's own reference examples (16:9, five-second output) convert those token rates into per-second figures:

The same model is listed in China on Volcengine at closely matching CNY rates (¥70 and ¥42 per million tokens; the five-second 720p reference video is ¥7.56, about ¥1.51 per second):

Three budgeting notes. First, Seedance 2.5 costs roughly 1.5x Seedance 2.0 per second at the same resolution ($0.231 vs $0.15 at 720p), and the whole 2.0 family remains on sale at unchanged prices — 2.5 adds a premium tier rather than replacing the budget ones. Second, the marquee 4K and 30-second output modes were app-side features (Jimeng and Dreamina) at review time: the API price list stops at 720p, and the published per-second examples assume five-second outputs, so budget longer clips with the token formula rather than a flat rate. Third, the familiar Seedance billing rules carry over — billing is token-based, only successful generations are charged, and video-input jobs have minimum-token floors.

MiniMax H3: 2K Output Priced Below Seedance 2.5's 720p

MiniMax — the company behind the Hailuo video line — released MiniMax H3 in the last days of July and now banners it across its platform as a new-generation open general-purpose multimodal video model. The official API docs list 768p and 2K output, 4-15 second durations, and mixed reference inputs of up to 9 images, 3 video clips, and 3 audio clips (12 files total), with audio generated alongside the video.

H3 pricing is pay-as-you-go with direct per-second rates on the official price page:

The comparison that matters this month: H3's 2K output at $0.13 per second is about 44% cheaper than Seedance 2.5's 720p output at $0.231 per second — the MiniMax flagship undercuts ByteDance's at a higher published resolution, and its 768p tier ($0.08) also beats Seedance 2.5's 480p tier ($0.103). ByteDance's counterargument is breadth: the cheaper Seedance 2.0 family below it, a far larger reference ceiling (50 inputs vs 12 files), and app-side 4K and 30-second modes that H3 does not match. The launch pricing head-to-head walks through the per-minute math for both models.

Two subscription notes. MiniMax's video packages state plainly that "MiniMax H3 is not supported yet," so there is no bulk-discount path — the per-second price is the price, while the older Hailuo 2.3 and Hailuo 02 models continue to be sold through point-based packages. And the $0.05-per-second regeneration lane makes a draft-first workflow arithmetic-neutral for keepers (768p at $0.08 plus regeneration at $0.05 equals the direct 2K rate of $0.13), which means drafting at 768p saves money in exact proportion to your reject rate: three 10-second drafts plus one 2K upgrade cost $2.90 versus $3.90 for three direct 2K takes.

Runway: Plan Price Is Low, Model Choice Controls the Real Cost

Runway's official page gives both the plan allowance and the approximate seconds produced by specific models. On annual billing, Standard is displayed at $12 per month and Pro at $28 per month.

The useful lesson is not that one Runway plan is "expensive." It is that the model can change the effective output cost by roughly 2x inside the same subscription. Teams should budget against the final-delivery model, not the cheapest draft model.

Runway remains a broader creative platform with scene generation, editing, and access to multiple third-party models. That breadth is part of the subscription value and is not captured by cost per second alone.

Adobe Firefly: Roughly $0.10 per Second Before Promotions

Adobe's current US pricing page lists Firefly Standard at $9.99 per month with 2,000 credits and up to 20 five-second videos. Firefly Pro is $19.99 per month with 4,000 credits and up to 40 five-second videos.

Standard: $9.99 / (20 x 5 seconds) = $0.0999 per second
Pro: $19.99 / (40 x 5 seconds) = $0.1000 per second

Adobe is also running a first-year unlimited-generation promotion on selected models and resolutions for eligible plans (listed as ending August 26 at review time). The exact model list varies by plan and Adobe says it may change. Because temporary unlimited access cannot be converted into a stable long-term unit price, the main table uses the standard published credit allowance instead.

Adobe's strongest purchasing argument is workflow integration and access to several model families in one interface, not a uniquely low base cost per second.

Pika: Resolution Doubles the Unit Cost

Pika's Standard plan is displayed at $8 per month on annual billing with 700 monthly credits. For Pika 2.5 text-to-video and image-to-video, a five-second generation costs 20 credits at 720p or 40 credits at 1080p.

720p: 700 credits / 4 credits per second = 175 seconds
$8 / 175 seconds = $0.0457 per second

1080p: 700 credits / 8 credits per second = 87.5 seconds
$8 / 87.5 seconds = $0.0914 per second

The calculation shows why a pricing report must state resolution. Saying "Pika costs $8" hides a 2x difference in effective output cost between 720p and 1080p.

Pika also sells editing effects with fixed per-generation credit prices, so the text-to-video rate should not be applied to every Pika feature.

OpenAI Sora: Not an Active H2 Buying Option

OpenAI states that the standalone Sora web and app experiences were discontinued on April 26, 2026, and that the Sora API will follow on September 24, 2026. That makes the older "Sora is included in ChatGPT Plus or Pro" calculations unsuitable for a second-half 2026 purchasing guide.

We keep Sora in the report as a market-change note because it remains a major brand and older pricing articles still rank for Sora comparisons. It does not receive an active price or cost-per-second estimate in the July table.

---

What to Buy for Each Workflow

Focused talking photos and face-video lip sync

Use a focused lip-sync product when you already have the face asset and need to replace speech, animate a portrait, make a photo sing, or test localized versions. FreeLipSync has the lowest entry price in this report for that specific workflow.

Managed avatar production and translation

HeyGen is the clearest fit when the requirement includes a large stock-avatar library, brand controls, translation, collaboration, and an established avatar-production workflow. Calculate costs against the exact credit rate for Avatar IV/V, translation, or Video Agent rather than the headline plan price.

Cinematic text-to-video and new scene generation

Google Veo, Runway, Adobe Firefly, and Pika belong in the shortlist when the job starts from a prompt or reference image and needs new scenes, motion, or visual transformation. Choose between them using quality tests on your own prompts, then apply the unit-cost math from this report.

API budgeting

Google Veo 3.1, ByteDance Seedance, and MiniMax H3 are the easiest platforms here to forecast because all three publish per-second prices. The budget anchors are unchanged: Veo 3.1 Lite for the cheapest 720p output ($0.05/s) and Seedance 2.0 Mini for the cheapest 480p drafts ($0.04/s). The late-July flagships sit above them as premium finishing tiers — MiniMax H3 at $0.08-$0.13/s and Seedance 2.5 at roughly $0.10-$0.23/s — and H3's $0.05/s regeneration lane rewards drafting at 768p before committing to 2K. Subscription platforms can be cheaper when every included credit is used, but they create waste when monthly balances expire or the wrong model consumes more credits than expected.

Second-Half 2026 Buying Checklist

Before subscribing, record these six values for your actual workflow:

1. Average finished-video length.
2. Required resolution.
3. Expected number of final videos per month.
4. Typical number of failed drafts or retries per final video.
5. Whether unused credits roll over.
6. Whether the plan includes commercial rights, watermark removal, and the model you tested.

Then calculate both best-case and expected cost:

Best-case cost = plan price / maximum published output

Expected cost = plan price / expected usable output after retries and unused credits

The second number is usually the one that belongs in a real production budget.

August 2026 Conclusions

1. Focused tools and cinematic generators should not be treated as interchangeable. FreeLipSync is inexpensive because it solves lip sync against an existing face asset, while Veo, Seedance, H3, and Runway generate broader visual content.
2. The late-July flagships are premium tiers, not new price floors. Seedance 2.5 costs roughly 1.5x Seedance 2.0 per second, and H3 has no subscription or package path at all. The cheapest API anchors are unchanged: Seedance 2.0 Mini at $0.04 per second (480p) and Veo 3.1 Lite at $0.05 (720p).
3. MiniMax H3 wins the flagship price-per-resolution comparison at launch. Its 2K rate ($0.13 per second) is about 44% below Seedance 2.5's 720p rate ($0.231), and its 768p tier ($0.08) also undercuts Seedance 2.5's 480p tier ($0.103). Seedance 2.5 answers with a 50-input reference ceiling and app-side 4K and 30-second modes.
4. Native audio and lip sync are becoming flagship defaults, and that changes lip-sync budgeting. Seedance 2.5 ships phoneme-level lip sync and H3 generates audio with every clip. For replacing speech on an existing face video, though, a focused tool still costs roughly 20x-55x less per second than the new flagship rates.
5. Resolution and model choice can double or quadruple effective cost. Pika and Runway demonstrated this in July; the new flagships repeat the lesson (H3's 2K costs 1.6x its 768p, and Seedance 2.5's 720p costs 2.2x its 480p).
6. Promotional and discontinued products need separate treatment. Adobe's first-year unlimited offer was still listed as ending August 26 at review time, and OpenAI's Sora API remains scheduled to shut down on September 24.

For creators who already have a portrait or face video, try FreeLipSync without signing up before paying for a broader video platform. For original-resolution output and longer inputs, compare the current Free, Starter, and Pro plans.

---

Official Sources and Review Log

Pricing was first reviewed on July 11, 2026 and re-verified for this August edition on August 3, 2026 from these official sources. Every source page was opened on a review date and the screenshots embedded above were captured directly from it.

• FreeLipSync product parameters and pricing
• HeyGen pricing and HeyGen credit usage guide
• Google Gemini API pricing for Veo
• BytePlus ModelArk pricing for Seedance and Volcengine model pricing for Seedance (Chinese)
• MiniMax pay-as-you-go pricing for H3 and MiniMax video packages pricing
• Runway pricing
• Adobe Firefly plans and Adobe promotion terms
• Pika pricing and model credit table
• OpenAI Sora discontinuation notice

Monthly update log

• August 3, 2026: Added ByteDance Seedance 2.5 (launched July 31) and MiniMax H3 (released late July) with official per-second pricing from BytePlus, Volcengine, and the MiniMax platform, including new source screenshots for both models. Re-verified every July platform — FreeLipSync, HeyGen, Veo 3.1, Seedance 2.0, Runway, Adobe Firefly, and Pika are all unchanged. Adobe's first-year unlimited promotion still lists an August 26 end date, and the Sora API discontinuation remains scheduled for September 24.
• July 11, 2026: Published the H2 baseline. Updated HeyGen to its 600-credit Creator plan, used Veo 3.1 pricing after the June 30 model migration, added ByteDance Seedance 2.0 per-second API pricing (BytePlus USD and Volcengine CNY), recorded current Runway/Adobe/Pika allowances, and removed Sora from active purchasing comparisons (web/app ended April 26; API ends September 24). All prices re-verified against the official pages on this date, with source screenshots embedded per section.