Kandinsky LabKandiSuperRes
KandiSuperRes Flash
The Flash package uses a distilled 2× super-resolution model and the existing Kandinsky Flash refiner.
Refine after increasing the resolution
The upscaling stage uses four pixel-diffusion steps. The existing Kandinsky Flash refiner then uses one latent-diffusion step; it is not a second newly trained Flash base model.Sber: KandiSuperRes Flash training, August 2024
Inspect details added by the model
Practical guidance. An upscaler can synthesize texture and alter small features. Treat the output as a generated reconstruction when exact source detail matters.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | ai-forever/KandiSuperRes/KandiSuperRes_flash_x2.pt |
| Weight file | KandiSuperRes_flash_x2.ptai-forever/KandiSuperRes — model card and files |
| Source-code license | Apache 2.0; the weight card has no separate license label.KandiSuperRes: source-code licenseai-forever/KandiSuperRes — model card and files |
Related models
- Kandinsky 3 Flash
Kandinsky Lab · Kandinsky 3
A distilled four-step Kandinsky 3 image model also used as the 3.1 release's refiner.
Launch date not confirmedDownloadable