Features

What the PixAI API can do: text-to-image with anime models and LoRAs, reference images, inpainting and outpainting, image-to-video and reference video, plus how tasks, credits and limits work.

The PixAI API runs the same anime models and community LoRAs as pixai.art, so your application can generate what pixai.art users generate. This page lists what the API can do, how it works and what it does not cover yet, with links to the reference for each feature.

What you can generate

FeatureEndpointWhat it does
Text to imagePOST /v2/image/createGenerates 1 or 4 images from a prompt with a PixAI model, optional LoRAs, style presets and color palettes.
Reference imagesPOST /v2/image/create with contextImagesUses up to 3 images as references, for example to keep a character consistent or take a pose from another image.
Inpaint, outpaint and redrawPOST /v2/gen-task/inpaintRedraws a masked area, extends an image beyond its edges, or redraws a whole image.
Image to videoPOST /v2/frame-video/createAnimates an image into a 5–15 second clip with the Tsubaki video model, optionally ending on a second image and adding sound.
Reference videoPOST /v2/reference-video/createGenerates a video guided by reference images and, depending on the model, reference videos and audio.
Models and LoRAsGET /v2/generation-model/searchSearches the PixAI Model Market for base models and LoRAs to generate with.

Images

  • Models: PixAI's own models, such as Tsubaki.3, Tsubaki.2, Haruka v2 and Hoshino v2, and other base models on the Model Market. See Models.
  • Sizes: 11 aspect ratios from 1:3 to 3:1, in two resolution tiers (1k and 1.5k). For other sizes, such as Tsubaki.3's larger outputs, send an exact width and height within the model's size ranges. See Supported resolutions.
  • Quality and speed: mode chooses an inference profile: lite, standard, pro or ultra. Not every mode is available on every model or plan.
  • LoRAs: apply up to 3 LoRAs without a membership and up to 15 with one, each with its own weight. A LoRA must be made for the same base model as the generation model.
  • Prompting: prompts of up to 10,000 characters on Tsubaki.2 and Tsubaki.3 (4,096 on other models), a negative prompt, a fixed seed for repeatable results, and an optional prompt helper that expands short prompts.
  • Styles and colors: named style presets, and color palettes on Tsubaki.3.

Image editing

  • Inpaint: redraw the area a mask marks white.
  • Outpaint: place the image on a larger canvas and fill in the new area.
  • Redraw: redraw the whole image without a mask; strength sets how far the result may move from the original.

See the Inpainting and Outpainting example for the mask format.

Video

EndpointModelsReferencesLength
Image to videoTsubaki (tbkv1.0.1, tbkv1.0)First frame, optional last frame5–15 s
Reference videoTsubaki1–6 images, up to 3 audio clips5–15 s
Reference videoSeedance 2, Seedance 2 FastUp to 9 images, 3 videos and 3 audio clips4–15 s
Reference videoSeedance 2.5Up to 30 images, 10 videos and 10 audio clips4–30 s
Reference videoVidu1–7 images3–16 s

Both video endpoints can generate a soundtrack, with speech in English, Japanese, Korean or Chinese, or no speech. professional mode adds an upscale on supported models. Image to video needs a PixAI membership.

How it works

  • REST over HTTPS at https://api.pixai.art, authenticated with an API key sent as Authorization: Bearer <key>. See How can I use PixAI API?
  • Asynchronous tasks: every generation endpoint returns a task. Poll GET /v2/task/{id} until it is completed, failed or cancelled, or receive status updates by webhook.
  • Media IDs: inputs and outputs are media IDs. Upload your own images and audio with the upload endpoints, or reuse an image you generated. GET /v2/media/batch turns media IDs into download URLs.
  • Download your results: generated media is not kept forever, so store the files you need soon after a task finishes.
  • Private by default: generations made through the API do not appear in your public generation history on pixai.art.
  • Machine-readable docs: an OpenAPI spec, llms.txt, a Markdown version of every page and a docs MCP server for coding agents.

To try it end to end, follow Your First API Call.

How credits are charged

The API spends the credits of the PixAI account that owns the API key.

  • When: credits are charged when a task is created. A task that fails is refunded.
  • How much: the cost depends on what you ask for.
    • Images: the model, the output size, the mode and the batch size.
    • Inpaint, outpaint and redraw: the image size and strength.
    • Video: the model, the duration, professional mode and whether a soundtrack is generated.
  • Where to see it: each task's paidQuota field shows the credits it was charged. Run a request once to see what it costs before you run it at scale.

Some options need a PixAI membership, such as image to video, more than 3 LoRAs and some sizes and modes. A request for an option your plan does not include is refused with an error.

Limits

  • Up to 10 tasks per account can wait in the queue at once. Tasks that are running do not count, and the cap can be raised per account.
  • Poll a task no more than once every 1.5 seconds, and back off when you retry.

See Limits for details.

Not available in the API yet

PixAI Reference Pro, PixAI Edit and PixAI Edit Lite are available on pixai.art but not through the API yet. They are under active development.

Ownership and commercial use

You own the copyright to the images you generate through the API, and PixAI claims none over them. Approved users may use PixAI's official models, such as Tsubaki and Haruka, for commercial work, but may not resell or redistribute the model files. Generated content must not infringe third-party rights. See Copyright and related rights.

Next steps

  • Quick Start: get an API key and make your first request.
  • Examples: complete scripts for image to video, consistent characters, inpainting and LoRAs.
  • API v2 reference: every parameter of every endpoint.