awesome-genmedia CCSR Upscaler
Upscale and enhance images using the CCSR (Content Consistent Super-Resolution) model. Supports configurable scale factors, color correction methods, tile-based processing for large images, and fine-grained control over the diffusion process.
Quick Start
Requires an each::labs API key. Get one at eachlabs.ai.
curl -X POST https://api.eachlabs.ai/v1/prediction \
-H "Content-Type: application/json" \
-H "X-API-Key: $EACHLABS_API_KEY" \
-d '{
"model": "ccsr",
"version": "0.0.1",
"input": {
"image_url": "https://example.com/low-res-photo.jpg",
"scale": 2,
"steps": 50
}
}'
Parameters
| Parameter | Type | Default | Description |
|---|---|---|---|
| color_fix_type | string | adain | Color correction method. Options: none, wavelet, adain |
| image_url | string | - | URL of the image to upscale |
| scale | number | 2 | Upscale factor |
| seed | integer | - | Random seed for reproducible results |
| steps | integer | 50 | Number of inference steps. Higher values produce better quality but take longer |
| t_max | number | 0.6667 | Maximum timestep for the diffusion process |
| t_min | number | 0.3333 | Minimum timestep for the diffusion process |
| tile_diffusion | string | none | Tile diffusion mode. Options: none, mix, gaussian |
| tile_diffusion_size | integer | 1024 | Size of tiles for tile diffusion |
| tile_diffusion_stride | integer | - | Stride for tile diffusion |
| tile_vae | boolean | false | Enable tile-based VAE processing for large images |
| tile_vae_decoder_size | integer | 226 | Size of tiles for VAE decoder |
| tile_vae_encoder_size | integer | 1024 | Size of tiles for VAE encoder |
Examples
Basic 2x upscale:
curl -X POST https://api.eachlabs.ai/v1/prediction \
-H "Content-Type: application/json" \
-H "X-API-Key: $EACHLABS_API_KEY" \
-d '{
"model": "ccsr",
"version": "0.0.1",
"input": {
"image_url": "https://example.com/small-portrait.jpg",
"scale": 2,
"steps": 50,
"color_fix_type": "adain"
}
}'
High-quality upscale with tiling for large images:
curl -X POST https://api.eachlabs.ai/v1/prediction \
-H "Content-Type: application/json" \
-H "X-API-Key: $EACHLABS_API_KEY" \
-d '{
"model": "ccsr",
"version": "0.0.1",
"input": {
"image_url": "https://example.com/landscape.jpg",
"scale": 4,
"steps": 60,
"tile_diffusion": "gaussian",
"tile_diffusion_size": 1024,
"tile_vae": true,
"color_fix_type": "wavelet"
}
}'
Related Models
- Eachlabs Image Upscaler | Pro | v1 - Simple one-click upscaling
- Qwen Image - AI image generation
- Cartoonify V2 - Convert images to cartoon style