Totally blurred – Can AI rescue something as bad as this ?

One of these ladies asked me if I could make this photo better?
"It’s lovely but a bit blurry, you can’t see any details, especially when you zoom in a bit", she said.

I looked and it is very low res, probably the focus was missed and the phone did the best it could to remove the noise from a situation that was probably quite dark.
Then someone slapped a “PARTY” sticker on it and sent it by Whatsapp. Terrible :-)
I hasten to say that it was not taken by me.

I said I can’t do anything at all, but know a few experts that might be able to do something?

I haven’t tried AI but have seen some remarkable results.
If anyone wants to have a go and see if any details can be recovered or probably better to say “generated“, then I’d be interested to see what you can do.

Thanks in advance
Trevor

Focus Magic.

Two things spring to mind:

RLT deconvolution - good for trying to improve "focus"
Wavelet decomposition - good for editing at specific detail levels i.e. 1px, 2px, 4px, 8px, etc.

Don't know of any particular AI that would do it.

I like the bottom left - wavelets at 1px and 2px plus UnsharpMask in the GIMP

Bottom right is wavelets plus deconvolution in RawTherapee

Is the posted original really only 640 x 640 pixels?

Anyway ... here's a face-recovered, slightly sharpened (I'm not sure that sharpening things other than the faces would be a benefit), slightly color adjusted, enlarged version produced by Topaz Photo AI. I would also crop it.

Used Topaz Photo AI on autopilot

Interesting how Autopilot also emphasized the otherwise hidden face behind the text bubble.

Thanks for showing the results of those programs and settings, appreciate that.
None of them make any sort of improvement, the results could even be regarded as worse than the admittedly very poor input I provided.
The results are full of visible digital pixel noise.

Many thanks! Yes the photo I was sent really only had that lousy 640x640 resolution. The original from the phone would have had more, but I don't have that. The Whatsapp reduces the photo size considerably :-(
This Topaz AI in the way you've used it has has made quite an impressive improvement on the faces.
You can now see more details eg. :
- improvement of the teeth, so they now look like individual teeth and not just a whiteish block
- the eyes have a definite pupil and you can see the whites of the eyes, wheras the original have just a dark blob.
- the lips are better defined
- the skin looks smoother, not so blotchy like in the original
and it hasn't introduced any obvious visible pixel artefacts which is really good.

However, what does detract from the overall result is that although the faces are definitely well improved, other objects (as you mentioned) are not improved.
So the hands, the skin around the neckline, the flowers and the glasses all still look "terrible" and the photo still has the overall look of being quite blurred

Many thanks too for this!
Of the three replies I would class this as the best overall result so far, because not only are the faces improved but also all the other details are sharper and have more details and contrast too. The saturation, warmth and brightness have also all been tweaked for the whole image. So the overall impression looks very good considering the terrible input that you started with.
However,
In the version from Topaz AI from sybersitizen where he concentrated on "just the faces", "his" resulting faces are definitely better than from this autopilot version, with even less blur and even more details.

I suppose the best result would be a combination of those two Topaz AI results :-)

Thanks for showing us what Topaz AI can do.

That is quite amazing, I guess that person was originally "removed" for some reason with the bubble :-)
I never noticed before, but the hidden face is also just about visible in the original

Photo AI offers a lot of flexibility. It can reduce noise, sharpen, crop, upsample, and adjust color and lighting. It can also confine some actions only to masks that it autogenerates or that you generate. My example could have (and should have) had more areas sharpened, which would be easy to do by modifying the mask. I might still prefer to leave some areas like the back wall unsharpened, though.

The face recovery feature is one of its highlights, and it's adjustable. The upsampling that I applied can also help.

The app now also has the ability to 'remove' things and replace them with AI generated content that's intended to look natural. This sometimes works wonders and other times just makes a mess. I tried it on the extended arm of the hidden person, but what was generated didn't look any better.

The various parameters could all be fine-tuned in a way to improve that photo more than either of us has done. Finally, adding the use of traditional manual retouching techniques would take it even further. It just depends on how much time the user spends.

This image has been also processed with AUTO tone in LR classic before sending to Topaz.

I wonder why no one thought to straighten the image somewhat? No follow-up image taken after dozens of bottles being consumed?

That's not one of the capabilities of the AI software. But try it. This approach introduces distortion and loses information from the sides.

When comparing the image on the left, it feels like [to me] as though the photo was taken on a boat...lol
Depends on what your priorities are, I presume...