FlashVSR v1.1 video upscaling

Sogni runs the open-weight FlashVSR v1.1 (Tiny Long, BF16) video super-resolution model on the Sogni Supernet as a standalone job. Give it one finished video, such as a MiniMax H3, LTX, or Wan render or a clip you upload, and it returns a sharper copy at 1080p or 1440p on the short edge. It is promptless: the output size is the only choice.

The result keeps the source's frame count, its exact frame rate (fractional rates such as 23.976 fps included), the full aspect ratio, and the original audio track. It never trims, crops, restyles, or interpolates. Each edge is enlarged up to two times, so 1080p needs a source at least 540 pixels on its short edge and 1440p needs at least 720.

#Workflow ID

#Product availability

• Sogni Web: Video → Upscale Video (Enhance), and Upscale video… on any finished video. Output 1080p or 1440p / 2K.
• Sogni API and SDKs: the exact model ID above; JavaScript SDK 5.37.1 and later export FLASHVSR_VIDEO_UPSCALE_MODEL_ID.
• Sogni Intelligence (hosted chat and durable workflows): the promptless upscale_video tool.

#Source requirements

• Up to 768 pixels on the short edge and about 1344×768 pixels overall (768×1344 in portrait); the output is at most 2560×1440.
• 1–362 frames (about 15 seconds) at a constant frame rate from 1 to 60 fps. The server enforces the length limit and refuses a longer source with a clear error.
• SDR, square pixels, any rotation already applied, 100 MB or less. HDR, variable-frame-rate, and rotated sources are refused.
• You do not send the source's frame count, frame rate, or size: the server probes the upload and uses its verified values. Any frames, fps, width, or height you do send must match the source.

#Pricing

An upscale is priced on the source's length and pixel area, in the same Spark units as every other model, and 1080p costs the same as 1440p because the model does the same work for both. The 1344×768, 158-frame reference clip (6.6 seconds) estimates at about 53 Spark on the Fast network; Sogni shows the exact estimate before the job runs. Unlimited subscribers use FlashVSR under fair use; without a subscription it runs on pay-as-you-go Premium Spark.

#Timing and worker hardware

In release qualification the reference clip upscaled to 2520×1440 in 212 seconds on an RTX 4090 and 101 seconds on an RTX 5090, both from a cold model load. On the Supernet the worker that takes the job sets the time. Workers need Comfy Worker 1.0.211 or newer, an NVIDIA GPU of the Ampere generation or newer with at least 23 GB of memory, and the 6.4 GB FlashVSR model pack, which eligible workers download automatically; see the Comfy Worker release notes.

#JavaScript SDK example

import { SogniClient, FLASHVSR_VIDEO_UPSCALE_MODEL_ID } from '@sogni-ai/sogni-client';
import fs from 'node:fs';

const sogni = await SogniClient.createInstance({ appId: 'my-app', network: 'fast' });
await sogni.account.login('username', 'password');

const project = await sogni.projects.create({
  type: 'video',
  network: 'fast',
  modelId: FLASHVSR_VIDEO_UPSCALE_MODEL_ID, // 'flashvsr_v1.1_tiny_long_bf16'
  positivePrompt: '',
  numberOfMedia: 1,
  referenceVideo: fs.readFileSync('./clip.mp4'),
  upscaleResolution: 1440 // or 1080: the output's short edge
});

const [upscaledUrl] = await project.waitForCompletion(); // MP4 with the original audio

To show a price first, call estimateVideoCost() with the output width/height (the source scaled so its short edge equals the target, both edges rounded to even pixels), the source's frames and fps, steps: 1, and sourceWidth/sourceHeight; the job itself is charged from the verified source.

#Notes

• FlashVSR sharpens detail and keeps the content: the same frames, frame rate, framing, and audio come back. It is not a generative re-render.
• Output is H.264 MP4 with the source's audio packets copied unchanged.
• For MiniMax H3 renders, FlashVSR is the recommended path to 1440p today.