1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
const axios = require('axios');
const api_key = "YOUR API-KEY";
const url = "https://api.segmind.com/v1/flashvsr";
const data = {
"video": "https://segmind-resources.s3.amazonaws.com/input/da981ec9-cb58-4a1d-b56f-49f4e0f38ac7-flash_720.mp4",
"mode": "tiny",
"scale": 2,
"color_fix": true,
"local_range": 12,
"seed": 42
};
(async function() {
try {
const response = await axios.post(url, data, { headers: { 'x-api-key': api_key } });
console.log(response.data);
} catch (error) {
console.error('Error:', error.response.data);
}
})();Attributes
Input video to upscale. Supports URL or video file; max duration: 60 secs.
Choose quality or speed. 'full' is high quality; 'tiny' is faster.
Allowed values:
Set upscaling level. 4 for max detail, 2 for balance.
Allowed values:
Enable color correction for video consistency. Recommended for natural visuals.
Set frame attention range. Higher for temporal accuracy.
min : 5,
max : 15
Set seed for reproducibility. Use -1 for random.
min : -1,
max : 99999999
To keep track of your credit usage, you can inspect the response headers of each API call. The x-remaining-credits property will indicate the number of remaining credits in your account. Ensure you monitor this value to avoid any disruptions in your API usage.
FlashVSR: Real-Time Video Super-Resolution Model
Edited by Segmind Team on October 24, 2025.
What is FlashVSR?
FlashVSR is an advanced AI model for real-time video quality enhancement. It utilizes sophisticated diffusion approaches to seamlessly boost video resolution even when it is being streamed live. FlashVSR with its integrated one-step diffusion and locality-constrained sparse attention that enables it to handle high-resolution videos at 17 frames per second on a single GPU, making it eons ahead of other models. The model is capable of rendering high-quality results due to its robust training, based on the vast VSR-120K dataset of 120,000 videos and 180,000 images, making it ideal for various platforms that need high-resolution videos in real-time.
Key Features of FlashVSR
- It can execute real-time processing at 17 FPS on a single GPU
- It is up to 12x faster than traditional diffusion-based VSR models
- It has flexible upscaling options (1x to 4x resolution)
- It is integrated with an intelligent color correction system
- It includes an adjustable frame attention range for temporal consistency
- It comes with two operational modes: full quality and tiny (speed-optimized)
- It is trained on a diverse VSR-120K dataset for superior results
Best Use Cases
- Content enhancement for live streaming videos
- Online video platform quality improvement
- Professional video post-production
- Security footage enhancement
- Gaming content upscaling
- Legacy video restoration
- Real-time broadcast upscaling
Prompt Tips and Output Quality
For optimal results, follow these prompt tips:
- Choose 'full' mode for the highest quality output, 'tiny' for faster processing
- Set scale factor based on your needs: 4x for maximum detail and 2x for balanced performance
- Enable color_fix for realistic results, especially in varied lighting conditions
- Adjust local_range (5-15) based on video motion; higher values for complex movement
- Use consistent seed values for reproducible results across multiple runs
FAQs
How does FlashVSR compare to other VSR models? FlashVSR offers up to 12x faster processing while maintaining high quality, utilizing its integrated diffusion framework and optimized attention mechanism.
What video formats and lengths are supported? The model supports standard video formats via URL or file upload, with a maximum duration of 60 seconds per processing session.
Should I use full or tiny mode? Use 'full' mode when you want high-quality results and processing time isn't critical. Choose 'tiny' mode for real-time applications and speedy processing.
How does the local_range parameter affect output? Local_range determines how many adjacent frames influence each frame's enhancement. Higher values (up to 15) provide better temporal consistency but result in increased processing time.
Can I process high-resolution videos in real-time? Yes, FlashVSR can process high-resolution videos in real-time, achieving approximately 17 FPS on a single GPU, making it suitable for live applications.
Popular Models
