Generate images and video

Image and video jobs use the same submit / result flow as text. The difference is the input object, which is the model's own, and the result type, MediaResult.

Submit an image jobCopy Anchor Link

Pass the model's input object. The keys it names decide how the job is metered (see Units):

const handle = await client.submit({
  model: "black-forest-labs/flux-2-dev:fp8",
  input: { prompt: "a lighthouse in fog, oil painting", width: 1024, height: 768, seed: 42 },
  sla: "batch",
  rateOut: 20n,
  provider: 7,
});

Output is priced per pixel: width × height × num_images (one image when num_images is absent). Use client.models.paramsSchema(model) to see which params a model accepts.

Submit a video jobCopy Anchor Link

A request that names duration or duration_secs is a video job, priced in pixel-seconds:

const handle = await client.submit({
  model: videoModel,
  input: { prompt: "a slow pan across the valley", resolution: "720p", aspect_ratio: "16:9", duration: 5 },
  sla: "batch",
  rateOut,
  provider,
});

• resolution is one of 480p, 720p, 1080p, 4k; aspect_ratio one of 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, auto or adaptive. Explicit width / height win over a tier.
• duration is a whole number of seconds (a number, a digits-only string, or "auto", priced as 15 s). Absent, it is priced as 5 s.

Pass reference assetsCopy Anchor Link

References (a start frame, an end frame, a source clip, style images) travel as bytes inside the sealed payload, never as URLs. Each one states its media type and true dimensions:

// In Node: Buffer.from(bytes).toString("base64")
const toB64 = (bytes: Uint8Array): string => {
  let binary = "";
  for (const byte of bytes) binary += String.fromCharCode(byte);
  return btoa(binary);
};

const handle = await client.submit({
  model: videoModel,
  input: {
    prompt: "the lighthouse beam sweeps through the fog",
    image:     { b64: toB64(first), media_type: "image/png", width: 1280, height: 720 },
    end_image: { b64: toB64(last),  media_type: "image/png", width: 1280, height: 720 }, // optional
    resolution: "720p",
    duration: 5,
  },
  sla: "batch",
  rateIn,
  rateOut,
  provider,
});

• A clip (video, or an element of reference_videos) must also state duration_secs, rounded up to whole seconds.
• The input side is priced from the dimensions you declare, without decoding the assets. Providers check them against the real bytes, so declare them truthfully or the job may fail.
• Limits are checked before anything is signed; see Reference assets.

Read the framesCopy Anchor Link

import { MediaResult } from "@vorq-ai/client-sdk";

const result = await handle.result();
if (result instanceof MediaResult) {
  const [first] = result.bytes();                // decoded frames, no network call
  console.log(result.frames[0]?.content_type, first.byteLength, "bytes", "seed", result.seed);
}

An image job returns one frame per image; a video job returns one frame holding the clip. Each frame carries b64, content_type and its dimensions. bytes() decodes them strictly and raises ResultIntegrityError on a frame that is not valid base64.

In a browser, turn a frame into something displayable with URL.createObjectURL(new Blob([bytes], { type: frame.content_type })).