AI video API pricing comparison

The short answer

Prices checked on 2026-09-22. One row per lotgen media key. No winner column.

The xAI resolution split is still on the pricing page. The worked cell is 5 seconds at 720p, $0.70. 480p and 1080p are named beside it.

Anthropic is on the same invitation form, after ElevenLabs and before xAI. It is copy, text only, and it is not a thirteenth video row.

What the keyword is asking

ai video api pricing comparison is a list-price question. Someone is lining up API rates before they pick a key: what the bill is measured in, what one stated clip costs at list, and what the docs actually cap.

It is not “which model looks better.” A same-prompt quality comparison is a different job, and this page does not do it. There is no footage here.

It is also not the usable-shot question. How much AI video generation costs owns the 3-attempt math. This page is the API grid: billing unit, one worked list-price example, official limits.

One list-price row per key

The order is the one the invitation form uses: Google, BytePlus ARK, fal, evolink, Kling, OpenAI, MiniMax, Black Forest Labs, Alibaba, ElevenLabs, xAI, Runway. The key is BytePlus ARK, not the parent company.

fal’s price for someone else’s model is fal’s price. evolink’s price for someone else’s model is evolink’s price. Those dollars stay on the fal row and the evolink row. They are not copied into Kling, Google, or OpenAI.

Google. Source: Gemini API pricing, read 2026-09-22. Veo 3.1 Fast with audio is $0.10 per second at 720p, $0.12 at 1080p, and $0.30 at 4k. The worked cell uses Fast at 720p for 8 seconds because the Veo guide sells 8 seconds at every resolution, and 4 or 6 seconds only where 1080p and 4k are not required. Standard with audio is $0.40 per second at 720p and 1080p, and $0.60 at 4k. Lite with audio is $0.05 at 720p and $0.08 at 1080p. 4k is not supported on Lite. The same Gemini key also prices Gemini Omni Flash (gemini-omni-1.1-flash) at $17.50 per million video output tokens. The page says that is about $0.10 per second at 720p, at 5,792 tokens per second. Omni is a different model. The worked cell stays Veo. Standard, Fast, and Lite in full live on Veo pricing explained.

BytePlus ARK. Source: ModelArk pricing, doc 1544106, read 2026-09-22. The video table header is online inference in USD per million tokens. Seedance 2.5 (dreamina-seedance-2-5-260628): 480p and 720p, input without video 10.70, input with video 6.40; 1080p, input without video 11.7, input with video 7.0. Offline inference for that row says not supported yet. The page’s token formula is (input video duration + output video duration) times output width times output height times output frame rate, divided by 1024. A clip dollar needs those dimensions, so there is no honest flat per-second figure to print here. Seedance 2.0 fast (dreamina-seedance-2-0-fast-260128) prints an original 5.6 USD per million tokens at 480p and 720p, input without video, with a time-limited 25% off. The page dates that window from 14:00 UTC+8 on 7 August 2026 through 14:00 UTC+8 on 7 October 2026, for enterprise users, and says 720p then starts at about $0.09 per second. That sentence is the page’s approximate, not a second list price, and it ends when the window ends.

fal. Source: fal Kling Video v3 Standard, read 2026-09-22. Audio off $0.084 per second, audio on $0.126, voice control $0.154. Five times $0.084 is $0.42. Five times $0.154 is $0.77, which is the example fal prints. These are fal’s rates, not Kling’s.

evolink. Source: evolink Kling 3.0, read 2026-09-22. 720p sound off $0.080 per second (about 5.4 credits). 720p sound on $0.120. 1080p sound off $0.106, sound on $0.159. 4K $0.398 either way. Duration on that page is 3 to 15 seconds. Five times $0.080 is $0.40. This is evolink’s rate.

Kling. Source: Kling developer pricing, read 2026-09-22. Video API, 1 Unit = $0.14 list. Kling 3.0 per second: 720p, 0.6 units without native audio ($0.084) and 0.9 with native audio and no voice control ($0.126); 1080p, 0.8 and 1.2 units ($0.112 and $0.168); 4K, 3.0 units ($0.42) with or without audio. The credit-pack story, and the longer unit table, live on Kling pricing explained. One cell is enough here. Kling versus Veo dollars live on Kling vs Veo cost.

OpenAI. The lotgen OpenAI key pays for GPT Image stills and text drafts. It is not a video row. OpenAI’s deprecations page, read 2026-09-22, still lists shutdown 2026-09-24 for the Videos API, sora-2, sora-2-pro, and the dated snapshots sora-2-2025-10-06, sora-2-2025-12-08, and sora-2-pro-2025-10-06. The recommended-replacement column is empty (---). The date has not passed. The API is not gone. No successor is named. That is why this key’s cell stays blank. The published Sora dollars are in the next section, and they are not this key.

MiniMax. Source: MiniMax pay-as-you-go pricing, read 2026-09-22. Hailuo 2.3 is $0.28 for a 768P, 6-second video, $0.56 for 768P at 10 seconds, and $0.49 for 1080P at 6 seconds. The worked cell is the $0.28 clip. A video package page bills points instead (pricing-video, read the same day): $1,000 buys 3,760 video points. That package is not the list used in the table.

Black Forest Labs. Source: BFL API pricing and the FLUX 3 overview, read 2026-09-22. 1 credit = $0.01. FLUX 3 text-to-video and image-to-video, full render: hd $0.17 per second, fhd $0.29, qhd $0.40, uhd $0.80. Draft text-to-video at hd is $0.06 per second. The worked cell is hd at 8 seconds, $1.36. The overview says up to 20 seconds, up to 4K (3840 by 2176 for 16:9), 24 fps, synchronized audio, aspects 21:9, 2:1, 16:9, 4:3, 1:1, 3:4, and 9:16. hd, fhd, qhd, and uhd are BFL’s names, kept as printed rather than translated to 720p.

Alibaba. Source: Model Studio model pricing, international scope, read 2026-09-22. wan2.7-t2v and the dated international rows beside it (wan2.7-t2v-2026-06-12, wan2.7-t2v-2026-04-25) are $0.10 per second at 720P and $0.15 per second at 1080P. Free quota on that table is 50 seconds, valid 90 days from activation, model release, or application approval, whichever is later. The free quota is not the list. The text-to-video API reference says wan2.7-t2v duration is an integer from 2 to 15, default 5. Five times $0.10 is $0.50. The same pricing page has other deployment scopes. This cell is the international USD row only.

ElevenLabs. Source: ElevenLabs pricing, read 2026-09-22. The plan answer on that page: Free $0 (10,000 credits); Starter $6 (30,000 credits); Creator $22 (121,000 credits, $11 for the first month); Pro $99 (600,000 credits). The FAQ says Flash and Turbo models, English and multilingual, cost between 0.5 and 1 credit per character on API usage. Starter lists Image & Video as a feature, and the compare table has an Image & Video row. Nothing on that page gives a per-second video API price, so this key stays a voice key in the table.

xAI. Sources: pricing, Grok Imagine Video 1.5, and video generation, read 2026-09-22. The pricing page and the 1.5 model page print grok-imagine-video-1.5 at 480p $0.08, 720p $0.14, and 1080p $0.25 per second. The model-page headline “$0.08 per second” is the 480p floor. The worked cell is 720p. Five seconds at 720p is $0.70. Five seconds at 480p is $0.40. Five seconds at 1080p is $1.25. Duration allowed range is 1 to 15 seconds. Aspects listed: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3. Generated videos include an audio track by default (generate_audio can be set false). The output price does not add a separate audio surcharge. Text-to-video and image-to-video can be 1080p on 1.5. Reference-to-video is capped at 720p. 480p is the generation guide’s default resolution. An older model, grok-imagine-video, prints 480p at $0.05 per second and 720p at $0.07 per second on the same pricing table. The worked cell is 1.5 at 720p, not that older model.

Runway. Source: Runway Dev pricing, read 2026-09-22. Credits are $0.01 each. gen4.5 is 12 credits per second, so $0.12 per second. The row does not print a resolution or a duration, so no clip dollar is worked here. The same guide prices other model ids in credits. Those are Runway’s rates for Runway’s other models.

Sora and the routes that are not keys

Sora is not a thirteenth key and it is not the OpenAI row above.

evolink’s catalog is the route this product prices. evolink Sora 2, read 2026-09-22, prints $0.085 per second (about 5.76 credits) and three clip charges: $0.339 at 4 seconds (23.04 credits), $0.678 at 8 seconds (46.08 credits), $1.017 at 12 seconds (69.12 credits). The page accepts duration 4, 8, 12, and aspect 1280x720 or 720x1280. The model parameter it names is sora-2-preview. It describes synchronized dialogue and sound. Those dollars are evolink’s. They are a footnote on the evolink row, not a new line in the table.

fal’s catalog does list Sora, as of the 2026-09-22 read. fal Sora 2 text to video prints $0.10 per second for Sora 2, and for Pro $0.30 per second at 720p, $0.50 per second for legacy 1080p (1792x1024 or 1024x1792), and $0.70 per second for true 1080p (1920x1080 or 1080x1920). The same page says the endpoint will be shut down on 24 September 2026. That is fal’s rate for fal’s endpoint. It is not Kling’s row, not the OpenAI key, and not a thirteenth renderer. The same page warns that fal-ai/sora-2/ and fal-ai/sora/v1/text-to-video shut down on that date.

OpenAI’s own pricing page still prints a Sora table, read the same day at API pricing: sora-2 at $0.10 per second for 720p (720x1280 or 1280x720); sora-2-pro at $0.30 for 720p, $0.50 for 1024p (1024x1792 or 1792x1024), and $0.70 for 1080p (1080x1920 or 1920x1080). Those are OpenAI’s published list prices. They are not the lotgen OpenAI key. The deprecations block, quoted above, still removes the Videos API on 2026-09-24, and the replacement column is still empty. The date has not passed, so this page still prints these dollars. After that date, re-read both pages before repeating any of them. If the API is gone, the OpenAI key is not a published video rate.

Luma is a model, not a key, so it has no row.

What the official docs actually limit

Each limit below links the page it was read from on 2026-09-22. None of them is a quality score.

• Google Veo 3.1 (guide): aspects 16:9 (default) and 9:16. Durations 4, 6, 8 seconds. 1080p and 4k, extension, and reference images must be 8 seconds. Resolution 720p, 1080p, or 4k on Standard and Fast. Lite stops at 1080p. Frame rate 24 fps. Audio always on. One video per request. Extension is a different route: 720p, in steps, up to 148 seconds total on the guide’s wording.
• BytePlus ARK (pricing): cost moves with token rate and with token consumption. Higher resolution costs more. Video input is a different rate from no video input. Fast’s 25% off and mini’s 60% off are dated enterprise windows, not the 2.5 list.
• fal Kling v3 Standard (model page): the price changes with audio off, audio on, and voice control. A 5-second example is on the page.
• evolink Kling 3.0 (page): 3 to 15 seconds, 720p, 1080p, or 4K, sound off or on. 4K does not add a sound surcharge.
• evolink Sora 2 (page): only 4, 8, 12 seconds, and only 1280x720 or 720x1280.
• Kling’s own pricing page prints resolution and audio rows. It does not print a duration cap in the payload read on 2026-09-22.
• MiniMax Hailuo 2.3 (PAYG): the printed clips are 6 or 10 seconds at 768P, and 6 seconds at 1080P.
• FLUX 3 (overview): up to 20 seconds, 24 fps, the aspect list above, synchronized audio. Draft is a cheaper hd pass, not a second model family.
• Wan 2.7 (API reference): duration 2 to 15 seconds, default 5. Resolution changes the international per-second price. Optional audio_url.
• xAI (generation): duration 1 to 15 seconds. The aspect list above. 480p, 720p, and 1080p. 480p is the default. 1080p on 1.5 is for text-to-video and image-to-video. Reference-to-video is capped at 720p. Audio on by default. Output price is 480p $0.08, 720p $0.14, and 1080p $0.25 per second. Video edits keep the source length, capped at 8.7 seconds. That cap is the edit route, not text-to-video.
• Runway gen4.5: the pricing guide prints 12 credits per second and does not print length or resolution on that row.

What this page will not decide

Same-prompt takes, keep rate, and which clip looks better need a real run. There is no footage on this page, and no quote for a take that has not been rendered.

Usable-shot math, the 3-attempt model, lives on how much AI video generation costs. One pair’s dollars, Kling against Veo, live on Kling vs Veo cost. A calculator for those rates lives at the AI video cost calculator.

There is no best model here. The pairwise pages, Runway vs Kling and Sora vs Veo, each take one pair at one duration. A Kling-versus-Veo footage comparison does not exist yet.

How one workspace holds many keys is the product page, one API across your keys.

Request access

lotgen is where those keys can be quoted before you commit. The seat is separate from the lab bill. Labs still bill the clip.

Request access for the developer path. If you drive generation from an agent, the setup notes are on AI video MCP.

Request an invitationSeats and pricingThe developer path: quote on your own keys before the agent spends.

Seats, from the pricing page: Solo $99. Team $90 a seat, minimum 3, $79 from 6 seats for the whole team. Two prices. No credits.