Image and video pricing

Sociaro gateway · all prices in USD · 2026-11-04

Video

Priced per second of video: the length you ask for on Kling, PixVerse and pruna/p-video, the length that comes back on the others. A 5-second 720p clip costs five times the 720p figure. A dash means the model does not offer that resolution.

alibaba/wan-2-7-r2v and alibaba/wan-2-7-videoedit (and alibaba/wan-3-video-prime, like alibaba/wan-3-video) are also billed for the seconds of video you send plus the seconds returned. The wan 2.7 and flash models have no 480P tier. alibaba/wan-2-6-i2v-flash is cheaper without sound, and the price follows what the generator reports for your clip.

alibaba/happyhorse-video-edit is the exception to the per-second rule: it is billed for the seconds of the video you send plus the seconds it returns, and the duration you request is ignored, because the length follows your file.

On kling/kling-2.6-*, native audio and a first + last frame are available at 1080P only, so a 720P request for either is refused. kling/kling-3.0-turbo-* always includes audio. kling/kling-3.0-omni is priced for text, frames and reference images; reference videos are not available on it yet.

Kling 4.0 (kling/kling-4.0-t2v, kling/kling-4.0-i2v, kling/kling-4.0-omni) opened at Kling on 2026-10-23. Kling's pricing page still labels its 4.0 prices a preview, so these may change when Kling publishes final pricing. Its default resolution is 1080P, and settings.duration is required. With input videos (feature_video, base_video) on kling/kling-4.0-omni, the billed seconds are Kling's: twice the output when the input videos together are no longer than it, else input plus output. A 10-second 1080P edit of a 12.5-second video bills 22.5 seconds: $3.17925. A custom voice made on /v1/kling/voices is $0.007, charged once when it is ready; an element made on /v1/kling/elements is free, and naming either in kling/kling-4.0-omni adds nothing to the clip's price.

Kling 4.0 Flash (kling/kling-4.0-flash-t2v, kling/kling-4.0-flash-i2v, kling/kling-4.0-flash-omni) is 720P only, 3 to 20 seconds, and settings.duration is required. With input videos on kling/kling-4.0-flash-omni, the billed seconds are counted as on 4.0: a 10-second edit of a 12.5-second video bills 22.5 seconds: $1.413. Naming an element in it adds nothing to the clip's price.

Seedance is metered in tokens rather than seconds, so the real charge tracks the frames actually rendered — expect it within about one percent of the figure above. A Seedance request that carries a source clip is metered at a lower rate than one generated from a prompt, and those with video input also have a minimum charge that scales with the output length you request. A still image used as a first frame or a reference is not video input.

On the routes that prepare your files before generating (/v1/generations/refs, and /v1/modelark when a task sends a file), and for a final rendered from a draft that sent one, a Seedance job is priced separately from the figures above. The cost of every job is shown in the console.

PixVerse

Per second of video requested, by quality; the second figure applies when generate_audio_switch is true.

PixVerse Video Expand is billed per second of the video you send, rounded up.

MiniMax H3

Served on the Jobs API (/v2/jobs) only. Priced per second of output at the resolution the model reports it rendered, plus each second of reference video at the rate in the fourth column, plus each picture (first frame, last frame, reference images) past the free allowance. Reference audio is free. A 10-second 768P clip from minimax/h3 with one picture is $0.96. minimax/h3-regeneration renders one of your own 768P minimax/h3 videos again at 2K: its reference video and pictures are the original job's, billed again at the rates in its row, separately from the original job.

Pruna models billed on the finished video

Priced per second of the video that comes back, read from the finished file and billed in whole seconds, rounded down; the charge is known when the job completes. pruna/p-video-2-pro-reference also charges for what you send: each reference image, and each second of reference video and audio, counted up to the output length plus half a second and at most 15 seconds of each in all. pruna/p-video-2-pro-avatar charges for the output only: the portrait, the audio, the voice and the language do not change the price. Until 17:00 UTC on 5 November 2026 it is 50% off the prices below (launch price): $0.01 per second at 480P and $0.02 at 768P, for jobs that complete by then.

Pruna models (pruna/*): your use of them and of their outputs is subject to Pruna's Terms of Service, including its Acceptable Use Policy (section 10), and to the licence of the model each one is built on (pruna/p-video: LTX-2; pruna/p-video-2-pro-reference and pruna/p-video-2-pro-avatar: MiniMax H3). You must follow them, and you may not use the outputs to train or improve a competing AI model; for the MiniMax H3 models the restriction is wider: no other AI model at all. If you make these models or their outputs available to your own users, you must bind them to restrictions at least as protective and tell them these restrictions apply.

Pruna video upscaling

Priced per second of the video that comes back, read from the finished file and billed in whole seconds, rounded up; the charge is known when the job completes. The rate is set by the target you ask for (1080P when left out) and by the frame rate of the finished video, the source's or the fps you set: above 30 fps it is double. turbo does not change the rate; the source's length sets the seconds returned, and so the total. At the regular price a 10-second result at 1080P and 24 fps is $0.20 ($0.10 at the launch price). Until 17:00 UTC on 12 November 2026 it is 50% off the prices below (launch price), for jobs that complete by then.

Pruna video replace

Priced per second of the video that comes back, read from the finished file and billed in whole seconds, rounded up; the charge is known when the job completes. The rate is set by the resolution you ask for (768P when left out); the pictures, instruction_prompt and the audio settings do not change it. A 10-second result at 768P is $1.00 at the regular price ($0.50 at the launch price). Until 17:00 UTC on 17 November 2026 it is 50% off the prices below (launch price), for jobs that complete by then.

Music and sound effects

Priced per second of the audio generated for your video, rounded up to a whole second. A shorter track is billed at the minimum. Each music variant (variants_num) is charged at the first track's billed length.

Images

Priced per image produced. Where a model lists sizes, the size you get decides the rate; where it says any size, the provider charges the same regardless of resolution or aspect ratio.

bytedance/seedream-5-pro is charged per image by the size it comes back at; leaving out size gives 2K. With layer_decomposition, each output image (the base and every layer) is charged at its own size, and leaving out size gives auto, the input image's own size.

Kling images (kling/kling-v3, kling/kling-v3-omni) are charged for the images delivered, never more than you asked for: n, or series_amount for a series (auto counts as at most 9).

Editing — what the pictures you send cost

Three of the models above also charge for the pictures you send them to edit. You pay the per-image rate above for what comes out, plus this much for each picture you put in:

So a single edit on alibaba/qwen-image-3 — one picture in, one out — comes to $0.03 + $0.003 = $0.033. Up to three pictures may be sent in one request. Every other image model on this page either takes no input picture, or charges only for what it produces. For models that are not on this page, ask your account manager.

Ideogram 4.5 — priced by quality

Priced per image the generator produces, by the quality you ask for, times num_images. On ideogram/ideogram-4-5 the rate also depends on whether you send input images (images); very_low needs them. Leaving quality out gives high without input images and medium with them (medium on Precise Edit). The size of the image and context_window do not change the price. Every image the generator produces is billed, including an image it withholds after its own safety check; a withheld image is not delivered, so a job can be billed for more images than it returns.

Ideogram 4.0, 3.0, tools and vectors — priced by speed, size or quality

Priced per image (or SVG) the generator produces, times num_images where the model has it. On the 4.0 and 3.0 models the rate is set by rendering_speed (default when left out); on the transparent models also by output_resolution (4.0, 1k when left out) or upscale_factor (3.0, x1 when left out); on ideogram/ideogram-vector by quality (medium when left out) and whether you send input images. Remix, inpaint, reframe and replace background bill the model's own rate. Sizes, aspect ratios, styles and style references do not change the price. As on 4.5, an image the generator withholds after its own safety check is still billed. The two vector models return SVG files.

What is not charged

A job that fails is not charged, and a job whose result we cannot deliver is not charged either, except an Ideogram job whose images were all withheld by the generator's safety check: it ends failed and is billed for the images produced. Listing models, polling a job and downloading a finished result are all free.