DeblurDiff — Deblur AI Model
DeblurDiff reconstructs detail a blur destroyed rather than merely sharpening edges, and it is the best of the deblur models here. It is also the slowest of them — roughly 7GB of VRAM, and not for images past 2K.
What it does
DeblurDiff reconstructs the detail a blur destroyed rather than merely sharpening the edges that survived it, which is what makes it the best of the deblur models here.
It is a diffusion model, co-trained with Stable Diffusion so that it has a sense of what a sharp image looks like, and it predicts the blur separately per region so an unevenly smeared frame is handled. It is also the slowest of them: one to three minutes.
When to use it
- Motion blur in a real photograph, not a synthetic test case
- Blur that varies across the frame instead of being uniform
- Texture and fine detail rebuilt, not just edges pushed harder
- Quality first, for when the faster deblur models fall short
- A photograph that justifies a one to three minute wait
Details
- CategoryDeblur
- Year2025
- LicenseMIT
Compliance & Provenance
- ProviderOpen-source · Specialized
- EU AI Act RiskMinimal Risk
- Art. 50 TransparencyNot applicable
Inputs & Outputs
- ImageInputimage
Blurred source image
- ImageOutputimage
Deblurred output image
Parameters
- seedintegerdefault -1range -1–2147483647
Random seed (-1 = random)
Random seed for reproducibility. -1 (default) randomizes; set a fixed integer to reproduce a previous deblur result.
Limitations
- Around 7GB of VRAM, and not intended for images past 2K
Tags
Alternatives in Image Restoration
- Image Colorization
Auto-colorize B&W photos (DISCO). 10-30s.
- Image Denoiser
Denoise photos (SwinIR). 5-15s.
- JPEG Quality Restoration
Remove JPEG compression artifacts (SwinIR). 5-15s.
- Motion Blur Removal (MSSNet)
Balanced motion-blur removal (MSSNet). 5-15s.