That night photo was almost perfect. Except your hands decided to shake at exactly the wrong moment.
Meet ComfyUI-PhotoDeblur, a free, open-source node pack by porfeng that helps restore photos affected by camera shake and motion blur.
And here's the interesting part: it works at your photo's original resolution. No mandatory downscaling just to fit the image into VRAM.
Instead of reaching for another generative upscaler, you get two specialized image-restoration models inside ComfyUI:
HI-Diff: A diffusion-based deblurring model that gradually reconstructs sharper details.
MISCFilter: A direct, single-pass restoration model. No diffusion sampling steps required.
Think handheld night photos, smeared neon signs, soft storefront edges, blurry architectural details, and motion trails around lights.
What makes the nodes useful?
Tiled processing: Split large photos into manageable regions and blend them back together. The suggested starting point is 768-pixel tiles with 96 pixels of overlap.
Flexible blending: Linear, cosine, Gaussian, and average options help reduce visible seams.
8-way TTA: Average rotated and mirrored predictions to improve stability, at roughly eight times the model computation.
Match Input Tone: Reduce broad color shifts so the restored photo stays closer to the original.
Official checkpoints: Download and verify RealBlur_J, RealBlur_R, or GoPro weights directly through a ComfyUI node.
Want to try it?
Install comfyui-photodeblur through ComfyUI Manager or clone the repository into custom_nodes. Load one of the included example workflows.
Start with RealBlur_J, tile 768, overlap 96, TTA off. Run the same photo through both models and compare the results at 100% zoom.
The developer even reports using an 8 GB GPU for the tiled examples. MISCFilter does require NVIDIA CUDA and CuPy, so check the installation notes.
One important reality check: deblurring isn't time travel.
No model can guarantee recovery of details that were never captured. Severe defocus, blown highlights, and tiny heavily blurred faces remain difficult. And because HI-Diff is a learned restoration model, you should still inspect fine details for artifacts.
But if you've ever lost an otherwise beautiful photo to shaky hands, this is exactly the kind of practical ComfyUI tool worth bookmarking.
GitHub, installation and sample workflows:
https://github.com/porfeng/ComfyUI-PhotoDeblur
#ComfyUI #PhotoDeblur #HIDiff #MISCFilter #PhotoRestoration #MotionBlur #OpenSourceAI
YouTube / ABVX
