Multimodal

Image upscaling

Same POST /v1/images endpoint as generation — pass input_references (exactly one image) to an upscaler model and get back a higher-resolution version of that same image. Unlike image editing, no prompt is required — these models take an image and (usually) a scale_factor, nothing else.

import requests

resp = requests.post(
    "https://videorouter.sh/api/v1/images",
    headers={"Authorization": "Bearer llmr_sk_live_..."},
    json={
        "model": "fal/clarity-upscaler/fal",
        "scale_factor": 2,
        "input_references": [
            {"type": "image_url", "image_url": {"url": "https://example.com/photo.jpg"}}
        ],
    },
).json()
print(resp["data"][0]["url"], resp["usage"]["cost"])

const res = await fetch("https://videorouter.sh/api/v1/images", {
  method: "POST",
  headers: { Authorization: "Bearer llmr_sk_live_...", "Content-Type": "application/json" },
  body: JSON.stringify({
    model: "fal/clarity-upscaler/fal",
    scale_factor: 2,
    input_references: [
      { type: "image_url", image_url: { url: "https://example.com/photo.jpg" } },
    ],
  }),
});
const data = await res.json();

curl https://videorouter.sh/api/v1/images \
  -H "Authorization: Bearer llmr_sk_live_..." \
  -H "content-type: application/json" \
  -d '{
    "model": "fal/clarity-upscaler/fal",
    "scale_factor": 2,
    "input_references": [
      {"type": "image_url", "image_url": {"url": "https://example.com/photo.jpg"}}
    ]
  }'
# -> {"data": [{"url": "..."}], "usage": {"cost": 0.126, ...}}

scale_factor (a number, typically 1–4) defaults to 2 when omitted. Real price depends on the OUTPUT size, which this platform computes itself from your input image's real dimensions × scale_factor — never guessed, never billed on a size we haven't confirmed (a handful of flat-priced rows below ignore scale_factor entirely and bill the same regardless of size, noted per-row). That cost already includes our 2% platform fee — see Pricing & billing.

Which models

11 models across 2 providers — fal.ai and Atlas Cloud.

Every model here also shows up under the Image Upscale filter on the models catalog.

Not yet supported

Mask-based/region-restricted upscaling isn't supported — every model here upscales the whole image. Every Topaz row is fixed to its family's flagship sub-model (e.g. Wonder 3.5 for the "generative" row) — picking a specific Topaz sub-model isn't exposed as a request parameter. stream: true is rejected outright, same as generation.