Image and Video Generation API Pricing in October 2026: What One Image and One Second of Video Cost Across 19 Providers

Last updated: October 18, 2026

Short answer: at the 1K size label, the lowest exact per-image price is $0.003 (Ideogram's P-Image Ideogram, Very Low quality) and the highest is $0.22, a 73x spread. At 2K it runs from $0.006 to $0.25. For video, the lowest per-second price at 720p is $0.025 (Alibaba Cloud's wan2.6-i2v-flash, silent) and the highest is $1.50 (Luma Ray2 on Amazon Bedrock), a 60x spread. Turning audio on doubles the Alibaba price. A large part of the market cannot go in either ranking: GPT Image 2.5 and the Gemini image models bill per token, Runway bills in credits, Kling in Units, and Vertex AI prices Veo "per count" without defining the count.

Every number here comes from two datasets we keep: the image generation pricing table (12 providers, 74 models, 211 price rows) and the video generation pricing table (14 providers, 64 models, 179 price rows). That is 19 providers between them. Each row carries a verbatim quote from the provider's own page, and every quote was string-matched against the saved page on October 18, 2026: 336 image quotes and 438 video quotes, all passing. Prices stay in the provider's own unit. Where we did arithmetic on them, we say so.

Images: price per image#

The cheapest at 1K and at 2K#

Like-for-like means the same size label, so it compares no batch rates, no "from" prices, no approximate figures and no retiring or retired models. A size label is the provider's own: Google's 1K row says "(1024x1024)" and Black Forest Labs says "about 1 megapixel", so 1K at one provider is not guaranteed to be 1K at another.

Three things in that table matter more than the headline:

• Quality tiers are inside the spread. The 1K floor is a Very Low tier; the same model's Medium tier is $0.01. Ideogram 4.5 at 1K runs from $0.03 (Generate Low) to $0.10 (Generate High) for plain generation, and the $0.22 top of the group is a with-source-images row. Drop the source-image and edit rows and the plain 1K spread is $0.003 to $0.10, about 33x.
• "Current" is mostly unstated. Ideogram, Black Forest Labs and Recraft do not publish a lifecycle for these models, so our data records them as not stated. Among models whose provider marks them current, the cheapest 1K price is $0.0336 for Gemini Nano Banana 2.1 and Gemini 3.1 Flash Lite Image, then $0.04 for Grok Imagine Image 2.0 at Low quality. At 2K it is Nano Banana 2.1 at $0.0504, then Grok Imagine Image 2.0 Low at $0.06.
• 1024x1024 has one model in it, and it is legacy. GPT Image 2 is the only current row labelled 1024x1024, and OpenAI lists it as previous generation with no shutdown date. Its per-image figures come from OpenAI's guide cost table and exclude text and image input tokens. The other 1024x1024 rows belong to models that retire by December 1.

Gemini 3 Pro Image (Nano Banana Pro) does not appear in either group because Google prices it as one "1K to 2K" row: $0.134 per image, $0.24 at 4K. Provider pages by name: OpenAI, Gemini API, xAI, Black Forest Labs, Ideogram and Recraft.

Token-priced image models#

Six models that are not retiring publish only token rates for image output: GPT Image 2.5 Sunburst and Flare on OpenAI, the same pair plus GPT-Image-2 on Azure OpenAI, and Amazon Nova 2 Omni (Preview). Google's image models publish both a token rate and a per-image price.

Why you cannot line these up against per-image prices, or against each other:

1. The number of tokens in one image depends on size and quality. GPT Image 2.5 has five quality options (low, medium, high, xhigh, max), and OpenAI publishes no per-image table for it.
2. One token rate covers a wide range of image prices. GPT Image 2 has the same $30 per 1M image output tokens as GPT Image 2.5, and OpenAI's table for it prices one 1024x1024 image from $0.006 (low) to $0.211 (high). That is a 35x range at a single token rate. It also tells you nothing about how many tokens 2.5 spends on the same image.
3. A token at one provider is not a token at another. OpenAI and Google both charge $30 per 1M image output tokens on their cheapest current models. Only Google converts that to a per-image price.
4. Even one provider's pages disagree on the count. Google's pricing page says a 4K Nano Banana 2.1 image uses 3,780 tokens; its image generation guide lists 2,520. Our arithmetic: 3,780 tokens at $30 per 1M is $0.1134, which matches the $0.113 per-image price, while 2,520 tokens would be $0.0756.

For GPT Image 2.5 specifics, see our Sunburst and Flare guide, and for why a per-token price is a weak guide to cost per task, see why price per 1M tokens misleads.

Video: price per second#

The cheapest at 480p, 720p and 1080p#

Same rules: per-second output prices of active models at one stated resolution, with no provider estimates, credits, Units, token prices, disputed rows or list prices carrying an unstated discount.

All three floors are Alibaba Cloud Model Studio rows in its International deployment scope; other regions on the same page list different prices. Both are image-to-video models. Next up at 480p are Grok Imagine Video 1.5 Lite and wan2.2-t2v-plus at $0.02; at 720p, Grok Imagine Video 1.5 Lite at $0.03 (a row xAI's markdown page disagrees with, see below).

Audio is where like-for-like gets slippery. Many per-second rows do not say whether audio is included. Where a provider prices both, here is what audio adds:

xAI's per-second rows do not split audio at all; its video overview lists native audio for Grok Imagine Video 1.5 and 1.5 Lite.

Credits and Units: Runway and Kling#

Runway bills in credits per second, at $0.01 a credit, with no resolution in the price row. Kling bills in Units per second and prints the dollar figure next to each, at a $0.14 list price per Unit, sold in prepaid packages.

Neither feeds the cheapest ranking. Runway's rows have no resolution, and Runway adds 5 credits per second for ProRes or PNG sequences and 20 for 10-bit and HDR (40 above 4 megapixels). Kling's dollar figure holds only at the list price, and Kling says its plans "currently don't support Kling 4.0", so no Kling 4.0 price is shown.

Provider estimates: BytePlus Seedance and Gemini Omni Flash#

Two token-priced families publish their own per-second examples. These are estimates for one clip shape, not prices, and we never rank them against per-second prices.

• BytePlus ModelArk Seedance bills per output token (Seedance 2.5 is 10.70 USD per 1M tokens at 480p and 720p without video input). Its estimates, for a 16:9 5-second clip with no video input: Seedance 2.5 at 0.103 per second at 480p and 0.569 at 1080p; Seedance 2.0 at 0.07 at 480p and 0.78 at 4K; 2.0 Fast at 0.06 and 2.0 Mini at 0.04, both at 480p.
• Gemini Omni Flash costs $17.50 per 1M video output tokens. Google gives "approximately $0.10 per second" at 720p, calculated from 5,792 tokens per second. The Vertex AI pricing page lists 1,931, 5,792, 8,688 and 17,376 tokens per second at 360p, 720p, 1080p and 4K. At $17.50 per 1M that works out to about $0.034, $0.101, $0.152 and $0.304 per second. That is our arithmetic, not a Google figure.

Vertex AI's Veo rows are the other unit trap. Veo 3.1 lists $0.40 with audio and $0.20 without at 720p and 1080p, but every Veo row reads "/ 1 count" and the table does not define the count. The Gemini API's Veo 3.1 preview prices the same $0.40 with-audio figure per second. We still keep them apart, because a matching number is not a stated unit. See the Vertex AI video page, and the Gemini Omni 1.1 Flash guide for what the model does.

Retiring soon#

Retiring and retired models never feed a "cheapest" figure above. Dates come from each provider's deprecation or lifecycle page, cross-checked against our deprecations tracker.

All four Gemini API previews name gemini-omni-1.1-flash as the replacement (Gemini deprecations). That swap is also a unit change: from a per-second price to a token price with a per-second estimate. Further out, gemini-2.5-flash-image retires March 15, 2027 on both the Gemini API and Vertex AI, and Azure's gpt-image-1-mini on April 7, 2027.

Retired, but the price is still on the page:

• Sora 2 on Azure OpenAI retired October 15, 2026, and the Azure pricing page still lists $0.10 per second (tracker). On OpenAI's own API, Sora 2 and Sora 2 Pro retired September 24 along with the Videos API, (deprecations), and no video model appears on the current pricing page.
• Amazon Nova Reel (v1:0 and v1:1) reached end of life September 30 and is still listed at $0.08 per second (tracker).
• Amazon Nova Canvas reached end of life September 30 and is still listed at $0.04 to $0.08 per image on the Nova pricing page. Bedrock's lifecycle page says requests to the model ID fail after end of life (tracker).
• Imagen 4 on Vertex AI (Fast $0.02, standard $0.04, Ultra $0.06 per image) has a June 30, 2026 discontinuation date in its legacy doc and is still on the pricing page. On the Gemini API, Imagen shut down August 17.
• Veo 3 and Veo 3 Fast on Vertex AI passed their June 30 retirement date and are still listed and priced.

For every other provider's dates, see the AI model shutdown calendar.

The docs disagree#

These are places where a provider's own pages give two answers. Each side is a verbatim quote in our data.

1. xAI: markdown pricing page vs rendered pricing page. The rendered page at docs.x.ai/developers/pricing prices each Grok Imagine video model by resolution. The markdown version of the same page lists one flat rate per model, equal to the 480p tier:

   Model	Rendered: 480p / 720p / 1080p	Markdown
   grok-imagine-video-1.5	$0.08 / $0.14 / $0.25 per sec	$0.080 / sec
   grok-imagine-video-1.5-lite	$0.02 / $0.03 / $0.14 per sec	$0.020 / sec
   grok-imagine-video	$0.05 / $0.07 per sec	$0.050 / sec

   The same split applies to images: the rendered page prices Grok Imagine Image 2.0 output from $0.04 to $0.08 by resolution and quality, while the markdown page lists a flat $0.04. If an agent prices your 720p job from the markdown file, it reads $0.08 instead of $0.14 for Grok Imagine Video 1.5, about 43% low. The two versions of the Grok Imagine Video 1.5 model page also differ on inputs and regions: the rendered page lists audio input and one region, the markdown page lists text and image input and three regions.

2. Replicate Wan 3.0: billing table vs readme. The model page bills $0.025, $0.05 and $0.10 per second at 480p, 720p and 1080p. The readme on the same page says $0.05, $0.10 and $0.20, and a banner says "Wan 3.0 is 30% off this week". Half price is not 30% off, so these rows sit out of every cheapest figure.

3. Vertex AI Veo 3.1 Fast: 4K on the pricing page, not on the model card. The pricing page lists 4K at $0.30 with audio and $0.25 without. The model card lists 720p and 1080p.

4. Luma Ray 3.2: a 360p draft tier on the pricing page only. The pricing page lists 360p draft at $0.06 per 5-second video. The model page lists 540p, 720p and 1080p.

5. Gemini API Veo 3: still Stable after shutdown. The Veo guide still marks Veo 3 Stable, while the deprecations page shut veo-3.0-generate-001 down on June 30, 2026.

6. Gemini Nano Banana 2.1: two token counts for a 4K image, 3,780 on the pricing page and 2,520 in the image guide (worked through above).

7. Gemini 3.1 Flash Image: replaced or stable? The deprecations page names Nano Banana 2.1 as the recommended replacement, at half the price per image output token, while the models page still lists 3.1 Flash Image as stable.

8. Alibaba Wan 3.0 Video: list price, unstated discount. The row reads "List price $0.1/second (Limited-time 30% off)" at 720p and never states the discounted price, so it is left out of the rankings.

Worked example: 1,000 images and 100 clips a month#

The workload: 1,000 product images at the 1K label, plus 100 eight-second clips at 720p. That is 1,000 images and 100 x 8 = 800 seconds of video a month. Every price below is per image or per second, at those exact size and resolution labels.

Assumptions, so you can swap in your own:

• A1. One generation per deliverable. No retries or rejected outputs.
• A2. Text-to-image at the provider's 1K label. No source images, edits or input-image fees. 1K is each provider's own label, and the tiers differ between mixes (Ideogram's Very Low quality, Gemini's Standard rate, xAI's Low quality), so the mixes are not equal in output quality.
• A3. Standard rates. No batch, Flex or promotional pricing. Prompt text tokens are not counted (Nano Banana 2.1 lists $1.50 per 1M input tokens).
• A4. Billed per second of output: 800 seconds. Alibaba's rows are its International deployment scope. Our data records 1 to 15 seconds for the xAI video models, and no maximum duration for wan2.6-i2v-flash, so check that it accepts 8-second clips.
• A5. Image-to-video needs a start frame. wan2.6-i2v-flash is image-to-video, so each clip can start from one of the 1,000 images. Our data lists no input fee for it. Grok Imagine Video 1.5 lists $0.01 per input image, which would add $1.00 for 100 clips.
• A6. xAI at the rendered pricing page's tiers, with the markdown figure shown alongside.

Mix 3 is 6.6x Mix 1 for the same count of deliverables. On xAI's markdown page, Mix 3 would read $40.00 + 800 x $0.08 = $104.00, which shows how much the docs disagreement is worth here: $48 a month. Swapping Mix 3's video to Grok Imagine Video 1.5 Lite at 720p (800 x $0.03 = $24.00) brings it to $64.00.

What could not go in this table without converting units: GPT Image 2.5 and Nova 2 Omni (tokens only), Runway (credits, no resolution), Kling (Units at a list price), Vertex Veo ("per count"), and Seedance and Gemini Omni Flash (tokens, with per-second figures that are estimates). For those, run a small batch and read the bill. If batch discounts matter to you, the cost levers directory covers how batch works across text APIs, and if the clips need narration, speech API pricing is in its own table.

FAQ#

What is the cheapest image generation API per image?#

At the 1K label, P-Image Ideogram at Very Low quality, $0.003 per image. At 2K, the same model at $0.006. Among models whose provider marks them current, Gemini Nano Banana 2.1 and Gemini 3.1 Flash Lite Image at $0.0336 per 1K image. Batch rates, starting prices and retiring models are left out.

What is the cheapest video generation API per second?#

Alibaba Cloud's wan2.2-i2v-flash at 480p ($0.015 per second) and wan2.6-i2v-flash at 720p ($0.025) and 1080p ($0.0375), all silent, image-to-video and in the International deployment scope. With audio, wan2.6-i2v-flash is $0.05 at 720p and $0.075 at 1080p.

How much does GPT Image 2.5 cost per image?#

OpenAI does not publish a per-image price for GPT Image 2.5 Sunburst or Flare, only token rates: $30 per 1M image output tokens and $8 per 1M image input tokens. The cost of one image depends on how many tokens its size and quality use, so it cannot be compared directly with per-image prices.

Which image and video models retire soon?#

The Gemini API Veo 3.1 previews and gemini-omni-flash-preview on October 22, 2026, gpt-image-1 on October 23, xAI's grok-imagine-image-quality on November 2, and OpenAI's gpt-image-1.5, gpt-image-1-mini and chatgpt-image-latest on December 1. The deprecations tracker has the full list.

Continue Reading#

• GPT Image 2.5 Sunburst and Flare: How to Use OpenAI's New Models - the token-priced models from the image section, with quality tiers and cost estimation.
• Gemini Omni 1.1 Flash Goes GA - the replacement for the Veo 3.1 previews retiring October 22.
• Grok Imagine Image 2.0 Ships - the xAI image model in Mix 3, and what its quality tiers change.
• AI Model Shutdown Dates 2026-2027: Calendar and Migration Plan - every provider's dates by month, beyond image and video.
• How to Cut Your LLM API Bill in 2026 - the same show-the-arithmetic approach applied to caching and batch on text models.

Sources#

Every figure above is a verbatim quote recorded in our image and video generation datasets, each read from these provider pages on October 18, 2026:

• OpenAI: pricing, image generation guide, deprecations
• Google Gemini API: pricing, image generation, Veo, deprecations
• Google Cloud: Vertex AI image pricing, Agent Platform pricing, Veo 3.1 model card
• Microsoft: Azure OpenAI pricing, model retirement schedule
• AWS: Bedrock pricing, Nova pricing, Bedrock model lifecycle
• xAI: pricing (rendered), pricing (markdown), video overview, Imagine Image Quality migration
• Ideogram API pricing, Black Forest Labs pricing, Recraft API pricing
• Alibaba Cloud Model Studio pricing, BytePlus ModelArk pricing, Kling API pricing, Runway API pricing
• Luma: pricing, model page; Replicate: alibaba/wan-3

Prices verified October 18, 2026.