Tutorials Shop

Supporting special needs music education since 2008

New!
Try our new Social Feeds, look out for the navigation icon: . You can now set default social feed view in your account preferences.

Reply to thread

Message: <blockquote data-quote="MikeMorrell" data-source="post: 630566" data-attributes="member: 4674">@Targa,[USER=5948]@Wonko[/USER] good and very valid questions! I got a few clues from the (voluntary) 'restorer', but I have yet to experiment with them myself on the same photos. I strongly suspect that he used some kind of AI 'generative fill' based on the 2 small blurry photos.What I do know is that the 'restoration requester' was absolutely over the moon with the detailed quality of the 'AI-restored/generated' photos! So however AI-generated the 'restores' were, they obviously fitted really well with either a) her personal memories of her ancestors but more probably b) how she 'imagined' they might look in more detail. So in either (or both) cases, she was delighted by the AI-generated 'restores'. So a 'photo restore' is probably not the right term in this case. Rather a high quality 'imagined detail'  which fits well within the parameters (such as face shapes, hair colors, age, etc.) of the 2 blurry photographs.FWIW, there's been an ongoing discussion between photo-restorers for years. One (valid) viewpoint that I wholeheartedly support - and the current guideline - is to be be completely honest with 'requesters' that some small, poor quality scanned photo's simply cannot be 'restored'. And to immediately ask 'requesters' to provide larger (with a higher scanned pixel density) for photo restoration. An alterative viewpoint is that that 'requesters' receive (copied) family photos via many sources. Such as via a long chain of social media, perhaps based on an old newspaper print, etc. So the alternative viewpoint is - given all the guidelines for 'requesters' to work as best we can with what we're given.But AI has changed the balance of these viewpoints somewhat. Firstly in the sense that 'AI generative upscale' is these days much better at increasing the (pixel) size of a photo by filling in 'potential/probable' details that simply don't exist in the smaller low-quality images. The second is in 'imagining' what the finer details might be in the 'upscaled image'.So yes, I completely agree that 'AI can't restore information that just isn't there'. And (therefore) it's doubtful that AI produces anything accurate.But some 'restorers' seem to be able to use AI tools to produce 'restored photo images' that so closely resemble the original (sometimes small, blurry, grainy) scanned photos that they are recognized by - and fully believed to be 'restores' of original photos by requesters.I should add that that 'photo restoration' has benefited from the application of some AI-enabled tools in (for example) PhotoShop. But that good 'photo restoration' still largely depends on the the 'restorer's eye' and his/her skill in applying a combination of both AI-enabled and 'traditional' (manual) restoration skills to achieve the 'best photo restoration'.So recent AI-enabled tools are just 'additional tools' in the photo-restorer's 'toolshed'. Useful at times, but not in themselves a 'one-click solution (or recipe) ' to photo restoration.</blockquote>