FLUX Video Upscale

The request

creativity lives under settings and takes 0 or 1. Set it explicitly when you want the precise mode, since the default is the creative 1.

import { createClient } from '@runware/sdk'

const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()

const [result] = await client.run({
  model: 'bfl:flux@video-upscale',
  deliveryMethod: 'async',
  upscaleFactor: 2,
  settings: {
    creativity: 0
  },
  inputs: {
    video: 'https://vm.runware.ai/video/os/a14d18/ws/2/vi/9f0a1b2c-3d4e-5f60-a1b2-c3d4e5f6a708.mp4'
  }
})

import asyncio
import os

from runware import Runware


async def main():
    async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
        results = await client.run({
            "model": "bfl:flux@video-upscale",
            "deliveryMethod": "async",
            "upscaleFactor": 2,
            "settings": {
                "creativity": 0
            },
            "inputs": {
                "video": "https://vm.runware.ai/video/os/a14d18/ws/2/vi/9f0a1b2c-3d4e-5f60-a1b2-c3d4e5f6a708.mp4"
            }
        })


asyncio.run(main())

curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "upscale",
      "taskUUID": "c8d9e0f1-2a3b-4c4d-9e5f-6a7b8c9d0e1f",
      "model": "bfl:flux@video-upscale",
      "deliveryMethod": "async",
      "upscaleFactor": 2,
      "settings": {
        "creativity": 0
      },
      "inputs": {
        "video": "https://vm.runware.ai/video/os/a14d18/ws/2/vi/9f0a1b2c-3d4e-5f60-a1b2-c3d4e5f6a708.mp4"
      }
    }
  ]'

runware run bfl:flux@video-upscale \
  deliveryMethod=async \
  upscaleFactor=2 \
  settings.creativity=0 \
  inputs.video=https://vm.runware.ai/video/os/a14d18/ws/2/vi/9f0a1b2c-3d4e-5f60-a1b2-c3d4e5f6a708.mp4

{
  "taskType": "upscale",
  "taskUUID": "c8d9e0f1-2a3b-4c4d-9e5f-6a7b8c9d0e1f",
  "model": "bfl:flux@video-upscale",
  "deliveryMethod": "async",
  "upscaleFactor": 2,
  "settings": {
    "creativity": 0
  },
  "inputs": {
    "video": "https://vm.runware.ai/video/os/a14d18/ws/2/vi/9f0a1b2c-3d4e-5f60-a1b2-c3d4e5f6a708.mp4"
  }
}

[
  {
    "taskType": "upscale",
    "taskUUID": "c8d9e0f1-2a3b-4c4d-9e5f-6a7b8c9d0e1f",
    "videoUUID": "5b6c7d8e-9f0a-4b1c-8d2e-3f4a5b6c7d8e",
    "videoURL": "https://vm.runware.ai/video/os/a14d18/ws/2/vi/5b6c7d8e-9f0a-4b1c-8d2e-3f4a5b6c7d8e.mp4"
  }
]

Precise mode holds identity

Set creativity to 0 for faces, logos, products, and brand-sensitive footage. The precise pass sharpens what is there without redrawing it, so a face keeps its features and a product keeps its exact shape. The creative pass, tuned to invent, can quietly change a face while it rebuilds detail, which is the last thing you want on a recognisable person:

On the presenter, the precise side stays the same person from the source, only cleaner. The creative side sharpens aggressively but nudges the features as it invents skin and hair detail. When the source is the source of truth, a client's talking head, a labelled product, a logo, the precise mode is the safe choice.

Creative mode rebuilds texture

Set creativity to 1, the default, for the footage that has no identity to protect and plenty of texture to gain: scenery, fabrics, foliage, crowds, and clips from a generative model that came out soft. Here the pass rebuilds plausible detail across a dense festival crowd, where a faithful sharpen would only firm up the blur:

The creative pass invents believable faces and clothing across the crowd, and because none of it has to match a specific real person, the invention reads as richness rather than error. This is the mode the owl wanted too: when detail matters more than exactness, let it invent.

Tips

1. Default to precise for anything recognisable. Faces, logos, products, and brand footage should use creativity: 0, since the creative pass can alter what it rebuilds.

2. Default to creative for texture. Scenery, fabric, foliage, crowds, and soft generated clips gain the most from creativity: 1, where invented detail reads as richness.

3. Check faces at full size. After a creative upscale, review any face against the source before you ship it. If the identity has drifted, re-run in precise mode.

4. Let the content pick the mode, not the resolution. Both modes upscale by the same factor. Creativity only changes how much the model invents, so choose it from what is in the frame.

5. Map it as an integer. Treat creativity as a numeric field in your integration rather than a boolean, so it stays compatible as the value set grows.