How to fix blurry photos from a wedding processional shot at 1/30 second

A processional at 1/30 second is motion, and motion has a direction. Fix the blurry photo on one frame before you touch the rest of the set.

The couple asked how to fix blurry photos from the walk up the aisle, and the file is ceremony-0042.CR2. The shutter was 1/30 of a second. I know because the EXIF says so, and because the picture says so in a way EXIF cannot. The dress is sharp. The groom is not. She paused on that step. He was still arriving. I restore wedding files when the album is already a month late, and this is the frame the mother asked for, not the sharp portrait they made on the steps afterward.

TL;DR. ceremony-0042 was exposed at 1/30, f/2.8, 50 mm, ISO 1600, and the raw is 5184 by 3456. The lace on the bodice is crisp. The boutonniere is ghosted about 14 pixels to the right, the hands sit in three clusters across 40 pixels, and the face is smeared by about 6. A global sharpen outlined the ghost flower and crunched the beads. A full-frame unblur tidied his smile into a generic one and doubled lace edges. I masked a mild face repair onto the original dress, left the hands soft, and shipped ceremony-0042-ship.jpg for an 8 by 10. The mother recognized him. I did not draw fingers.

Two sharpness levels in one shutter

I opened the raw in a plain converter and exported ceremony-0042-base.tif at the native size, no sharpening, no noise reduction beyond the converter's default which I then turned off. I want the smear as the sensor recorded it. The uncle had already made ceremony-0042.jpg from a phone app, 2048 pixels on the long side, with a heavy clarity pass. I set that JPEG aside. It is a clue to what the family has been looking at, and it is a bad master. The beads have white rims. The groom's ear has a double contour. We do not repair from the double contour.

On the base TIFF I picked three measurements and wrote them in the notes before I tried to be clever.

The boutonniere. A white flower on his left lapel. The flower's bright core is about 18 pixels wide, and a second, dimmer core sits 14 pixels to the right. That offset is the motion during the exposure, or most of it. Fourteen pixels at 50 mm and this distance is a small move of the chest, a sway, not a sprint. Small enough that I hoped the face might be recoverable. Large enough that a sharpen will draw both flowers.

The hands. His right hand is at her waist. I see three clusters of fingers, spanning about 40 pixels, each cluster semi-transparent. Her hand on his arm is sharper, because her arm was still. The two hands overlap. Any repair that "picks a hand" has to pick from a stack of three and from her fingers in the same pixels. I am not picking.

The lace. A bead on the bodice is about 3 pixels, round, with a dark thread hole. The edge of the bead does not have a twin. She was still. This is the control region. If a process puts a halo on that bead, the process is too strong for the frame, even if the groom's eyes get better.

The face sits between the flower and the hands. The collar shows a short smear, about 6 pixels, less than the chest. The head did not travel as far as the lapel, or it traveled more slowly. The eyes are soft ovals. The smile is smeared along the direction of the walk, which is roughly horizontal in the frame, a little downward because he was mid-step. Direction matters. This is not a focus miss. A focus miss would soften the dress too, or it would soften a plane and leave the other plane alone in a way that follows distance, not time. She is in the same plane. She is sharp. He moved.

How to fix blurry photos starts with that sentence. Name the blur. If you cannot name it, you will throw a sharpen at a motion trail and call the trail detail.

What a global sharpen did to a still dress

I built the failure on purpose. ceremony-0042-sharp-global.jpg is an unsharp mask on the whole base TIFF, amount 150, radius 1.2, threshold 0. I chose those numbers because they are the kind of numbers a slider gets dragged to when someone is angry at a soft face.

The lace became jewelry from a bad catalog. Every bead grew a bright ring. The thread holes filled with white. The dress, which was the best part of the exposure, now looks processed standing next to any other frame from the day. I compared it to ceremony-0011, a portrait at 1/200 where the lace is merely lace. The global mask on 0042 is crisper than 0011 and worse. Crisp is not the family resemblance of the album.

The boutonniere became two flowers. The ghost core, 14 pixels over, picked up its own edge. At the 8 by 10 size you can see a pale second bloom. The mother saw it immediately. She asked if he had worn two. He had not. I closed the file. A sharpen does not know which of the two positions was the flower at the middle of the exposure. It honors every edge you give it, including the edges of the mistake.

The eyes got a little more contrast and stayed smeared. There is no high-frequency eye to amplify. The mask boosted the boundary of the smear. His eyelid became a darker streak. That is the usual betrayal, and it is why I do not start motion repair with the sharpen tool. If you want the general order for a soft picture that is not this walk, I put a longer version in how to sharpen a blurry picture. This aisle has a mask in it, so it gets its own notes.

A full-frame unblur was a new photograph

Next I ran the whole frame through the unblur image path and saved ceremony-0042-unblur-all.jpg. I used the fix blurry photo tool rather than a raw-plugin deconvolve because the family had already been sold "AI unblur" by the uncle's app, and I wanted to see the same class of tool on the real pixels.

The dress got worse in a new way. The beads stayed round and became plastic, too equal, the thread pattern regularized. A few beads that were partly in shadow, genuinely dim, came back as bright as the lit ones. The model has seen a lot of wedding lace and it completed the set. She did not wear a completed set. The lower bodice was in his shadow. Those beads should be dim.

His smile changed. The smeared open mouth, which was a real mid-laugh with an uneven corner, came back symmetrical. I have a reference, ceremony-0036, a sharp frame twenty seconds later at 1/200, where the laugh is crooked on the camera-left side. The unblur smile matches neither the smear nor 0036. It matches a well-exposed groom in general. The mother said the word "stock." She was right, and she does not work in this job.

The background went nervous. Out-of-focus lights in the aisle were smooth discs in the base TIFF. The full-frame pass gave them rings, little onion edges. Bokeh is not motion blur, but a model asked to undo blur will sometimes decide that softness is the enemy everywhere. I do not want onion lights in a church.

The hands became one hand. The three clusters collapsed into a five-finger shape that is plausible and not his. The grip is wrong relative to 0036, where his fingers are spread because he was nervous and she has teased him about it since. A neat grip edits the nerves out. I will not do that for a sharper album.

I kept the full-frame pass in _rejected. It is the file I will show the next couple who ask for a one-click aisle. One click made a different wedding.

The mask, and how small the repair stayed

The working file is a layered TIFF, ceremony-0042-work.tif. The bottom layer is the base conversion. The dress, her arm, the aisle, the lights, all of that stays on the bottom. I painted a mask for his face, ear, neck, and the collar, feathered about 12 pixels so the jaw does not have a seam. I did not include the boutonniere. A 14 pixel ghost is a second object. Repairing it would mean choosing a moment inside the exposure, and I do not have a reason to prefer the left core over the right except that the left one is brighter. Brighter is not the same as "where the flower was at the midpoint." I left both cores on the original layer. At 8 by 10 the ghost is faint if I do not sharpen it. Faint I can live with. Two hard flowers I cannot.

On the face I ran a short deblur, the same tool, and kept the result only inside the mask. The output of that region is ceremony-0042-face-only.tif. The eyes tightened. You can see a catchlight that was a smear of window and is now a small rectangle, which matches the church windows in the sharp frames. The crooked corner of the mouth survived, softened, not replaced. I checked it against 0036 at the same zoom. Not identical. A mouth moves. Similar enough that the mother stopped saying stock.

The ear was the dangerous edge of the mask. Motion had trailed the earlobe. The repair pulled it back and wanted to invent a bright rim. I reduced the opacity of the repaired layer to 70 percent over the ear only, a nested mask, so some trail remains. A half-true ear beats a cast ear. At the print size the difference is small. At 100 percent it is the difference between a lobe I trust and a lobe I drew.

I did not run the tool twice. A second pass on an already repaired face starts inventing pores that the 1/30 smear never contained. One pass, 70 percent on the ear, 100 percent on the eyes and mouth, stop. The file is ceremony-0042-ship.jpg, flattened from that stack, 5184 by 3456, sRGB, JPEG quality 11 for the lab, plus the layered TIFF in the archive so I can undo the mouth if anyone objects.

Hands, left in the smear

I tried a separate hand mask because I felt I should, and I am writing it down so I do not try it again on a triple exposure of fingers. The tool returned a clean hand with a thumbnail highlight the source clusters do not support. I put it at 30 percent opacity and it still looked pasted, a sharp hand on a soft sleeve. The sleeve is smeared. A sharp hand on it breaks the exposure. I deleted the hand layer.

What I did instead is local and dumb. I darkened the hand region by about a third of a stop and lifted the surrounding dress a hair, so the eye goes to the lace and the face. The triple fingers are still there if you hunt. They are not the loudest thing in the print. This is not a fix for the hands. It is a way to stop the failed part from shouting. I told the couple that sentence. They would rather have the walk than a drawn hand, and the sharp portrait on the steps already shows his fingers.

The 8 by 10, and the enlargement I refused

They print at 8 by 10. The frame is vertical. The picture is horizontal, 5184 by 3456, so the lab will crop or the printer will letterbox. I made the crop myself, ceremony-0042-8x10.jpg, 3000 by 2400, which is 300 pixels per inch at 10 by 8. I cropped from the top of her hair to just below the hands, and I kept both of them. A tighter crop on his face would have hidden the hands and also hidden the walk. The walk is the picture.

3000 pixels on the long side of the crop, from a 5184 original, means I scaled down. There is no upscale in this job. People pair blur repair with enlargement because both feel like "make it better." Scaling this smear up would have given the next tool more soft pixels to explain. I had more pixels than the print needs. I threw pixels away on purpose.

At the 8 by 10 proof the lace matches ceremony-0011 well enough that the spreads will not jump. His eyes read as eyes. The ghost boutonniere is a slight pale echo, and the mother, who noticed two flowers in the bad sharpen, did not mention the echo. I pointed it out anyway. She said she would rather know. We left it.

I refused a 16 by 20 of the same repair. At that size the 6 pixel face smear, even after the masked pass, becomes a softness you can name from the sofa, and the hands become a topic. The steps portrait can go large. The processional stays at 8 by 10. I wrote that on the order sheet so an enthusiastic parent does not reorder big.

Noise, which was along for the ride

ISO 1600 on this body is not a crisis, and it is not nothing. The black of his suit shows a fine chroma speckle, magenta and green, and the global sharpen had made it into colored sand. The masked repair boosted a little of it on the cheek. I ran a light color-noise pass on the face layer only, strength low enough that the catchlight rectangle stayed rectangular. Luminance I left. The cheek should not become the plastic of the rejected lace.

If the speckle had been the main complaint I would have treated this as a grain job and left the motion alone. It was not the main complaint. The mother never said noise. She said his face. I mention the speckle because a sharpen-first workflow would have made noise the second crisis, and then someone would have denoised the lace flat. Order saved me a detour. Name the motion. Mask. Then, if the repair area got sandy, calm the color speckle and stop.

What I told the uncle with the app

He meant well. He had run four frames through the phone and printed a draft book. I sat with him for twenty minutes and we looked at 0042 at 100 percent on the base TIFF and on his JPEG. I asked him to point at a bead. On the base TIFF he could point at a dim bead and a lit bead. On his JPEG every bead was lit and rimmed. He saw it. People can see it when the comparison is a single bead and not a slogan.

I gave him a short list, not a sermon.

• Keep the raw. ceremony-0042.CR2 is the master. The 2048 pixel JPEG is a sketch.
• Do not run the whole aisle through one unblur. The dress and the groom did not share a velocity.
• If you sharpen, check a region that was already sharp. If that region grows a halo, undo.
• Read the shutter. 1/30 with a walking person is motion. More megapixels would have smeared in more detail. They would not have frozen the step.
• Pro the motion frame smaller than the still frames from the same day.
• Do not rebuild hands that occupy three places.

The questions that usually follow, about whether a blur is the kind a tool can undo, are in can you unblur a photo. Clearing a whole set of soft files that are soft for mixed reasons is how to clear blurry photos. I am not going to pretend this aisle is those essays. This aisle is one shutter, two velocities, and a mask.

The versions, including the ones that lied

File	What it is	Kept?
ceremony-0042.CR2	Raw, 1/30, 5184 by 3456	Master
ceremony-0042-base.tif	Flat conversion, sharpening off	Bottom layer
ceremony-0042.jpg	Uncle's 2048 px app export	Reference for the damage, not a master
ceremony-0042-sharp-global.jpg	Unsharp 150, radius 1.2, whole frame	Rejected. Two flowers, rimmed beads
ceremony-0042-unblur-all.jpg	Full-frame unblur	Rejected. Stock smile, plastic lace, one false hand
ceremony-0042-face-only.tif	Repair inside the face mask	Used at partial opacity on the ear
ceremony-0042-ship.jpg	Flattened layered file	Archive master of the repair
ceremony-0042-8x10.jpg	3000 by 2400 crop	The print

The picture above is not the wedding. It is a generated sample, a teal panel and a round head and a small green shape, blurred so every edge bleeds. I show it when someone wants a single definition of blur. The sample is blur everywhere, one kind, even and optical. The aisle is not that. Half the aisle is sharp. If your file looks like the sample, you may be fighting focus or a dirty lens or a resize. If your file looks like ceremony-0042, with one person crisp and one person trailed, you are fighting time, and the sample will not teach you the mask.

What I would change on the next aisle

I would have liked 1/250. The rest of the day, when the photographer remembered, is 1/200 and 1/250 and the faces are simply faces. Frame 0042 is the one where the mode dial and the church light met. I cannot reshoot it. I can refuse to let a model reshoot it inside the file.

The mother has the 8 by 10. She wrote to say his mouth looks like the laugh she knows. That is the bar. Not beads made of glass. Not a grip he did not use. A mouth that belongs to the man who was walking, repaired only inside the few pixels the smear left me, and a dress left in the stillness she actually gave the shutter.