วีดีโอ

We're lucky to have that choice right now. Think back to the early 1900s when cars first appeared. No paved roads, no traffic laws, clunky vehicles, and inexperienced drivers. It made perfect sense to stick with your trusty horse then! But once roads improved, cars got better, and drivers gained experience, well... riding a horse wasn't the smartest option anymore. AI might be at that 'early car' stage now, but things can change fast. It's wise to keep an open mind.

Idd there is no such a model in the provided link so I removed it from the description. I did not use this model anyway in this restoration and in general I do not use it for any restorations. I sometimes experiment though and I use the ones from InvokeAI from Hugging face, both the plus or the smaller one. Searching Google will result those models at the top of the list.

What I do in the video, is changing values in the text editor, then save, then I am double clicking on the kml file in the file explorer to refresh the info in the Google Earth. The refresh is not happening automatically.

@@nikolakakism thanks, smart!

Thank you, I appreciate that!

You can find it on the following page www.panotours.gr/using-ai-to-colorize-black-and-white-photos-with-emphasis-on-the-ddcolor-models/. In the 7z files you can find another recolor process that uses Controlnet and SDXL checkpoints which may be much slower but can output better results.

Спасибо

There won't be a next time for this type of content anyway. Thank you though!

Thank you! I am not using ComfyUI to run AniPortrait even though I see that already there is a custom node to use it from there github.com/chaojie/ComfyUI-AniPortrait/tree/main. I also see that there are a couple of videos dedicated to the installation so you may have some luck following these.

I did not know of its existence until now :) I am looking at the 'brand new' AniPortrait, which unfortunately is great but requires a beefy PC and I may take a look at VideoRetalking as well and post a video update.

Ευχαριστώ για τα καλά λόγια! Θεωρητικά μπορείς να εκπαιδεύσεις ένα μοντέλο τεχνητής νοημοσύνης να παράγει οποιοδήποτε μουσικό όργανο, αλλά θέλει πέρα από καλό και ποικίλο φωτογραφικό υλικό και αρκετό χρόνο και πολύ καλό υπολογιστή. Πάλι η απεικόνισή του δεν θα πετυχαίνει πάντα και 100%. Στα προηγούμενα δύο βίντεο αυτό που 'αντιλαμβανόταν' το μοντέλο ως μουσικό όργανο ήταν η κιθάρα, για αυτό και στο βίντεο του Δελιά το σχήμα του μπουζουκιού είναι περίεργο. Σε αυτό τουλάχιστον , επειδή έχω κάνει την εκπαίδευσή του για μπουζούκι, τουλάχιστον κρατάει το σχήμα καλύτερα.

Thank you! Cheers!

Glad you enjoy it!

Glad it was helpful!

Thank you much appreciated!

Να είσαι καλά! Πάντως δεν είναι τόσο οι ικανότητές μου όσο τα εργαλεία και οι μέθοδοι που εξελίσσονται με ραγδαίο ρυθμό το τελευταίο διάστημα. Μια απόδειξη είναι το EMO το οποίο αν δεις υπάρχει η πιθανότητα σε μικρό διάστημα να μη χρειάζεται και καμιά ικανότητα για να δεις το Βαμβακάρη όχι μόνο να τραγουδάει, αλλά τραγουδάει και με συναίσθημα! humanaigc.github.io/emote-portrait-alive/

Eυχαριστώ για το lλινκ,το σάιτ είναι φοβερό!

Thank you, much appreciated.

There are good and better tools for one click photo restoration, but there is no perfect one for the job. It is like expecting Chat GPT will produce you a novel, or whatever with one prompt without interfering. Sure, it will produce one, but is it going to be the one that you want? Now, Magnific AI which for some reason does not give a demo upscale to test it, is a fishy service. It is expensive, it does not give you the ability to make refunds and the available information is limited to say the least. The creativity sliders you mention are another fancy way to call the denoising strength. If you keep the denoising strength to 0 you get the same image that you started with. To restore a bad-quality image you will need to introduce data though, so I am sure that the creativity slider value of 0 means a very low denoising strength. By introducing data, inevitably you alter the image and if the image is of a known person their characteristics will be altered as well. Sure it will be a person, probably with sharp details and great eyelashes and such, but is it going to be the person that was really in your source photo? One easy way for you to test it is to take one photo of yours and scale it to 220 X 300 pixels, introduce some noise to it, to resemble my source photo, and upscale it to see if the output is really you! The reason that I am using many steps of generations is an attempt to control the output more and introduce this new data that will recover the bad parts of the image. It does not matter to me if it is going to be SD1.5, SDXL, or Stable Diffusion 3 as long as I have better data than the source to work with. For these restorations I predominantly use SD1.5 based checkpoints and not SDXL to speed up the generations and sure when they release Stable Diffusion 3 with the same license as they have with their previous models I will test it and hopefully use it as well.

@@Sczw I do have a licensed Topaz Photo AI and I sometimes use it since it does perform a bit better as an upscaler than Photoshop's Super Zoom and with the detection of faces for the face restoration process. However if you want to compare a one click 4Xupscale from Topaz Photo AI of the photo that I did in my video, I posted it in my community tab th-cam.com/channels/gg75iAFYuUeAafHkkabyjg.htmlcommunity

@@Sczw have you not tried Magnific AI? It’s the best upscaler right now. It’s works totally differently to Topaz. You have to lower the creativity slider so it doesn’t change it. You can do literally infinite upscale in theory

@@nikolakakism it isn’t fishy. It’s just expensive because there’s no alternative right now. Plenty of people that use it you can look up on TH-cam or X. You can do 8x upscale in one click.

Thank you! Cheers!

Thank you for your edit. The Stoker sounds better and changed it in the description. I didn't know the word of course so I went for the descriptive phrase :D.

Pleased to be of service - kali sinechia!

Thank you my friend. I will make some more videos about photo restorations and reimagination soon.

Να είσαι καλά φίλε μου. Ήχο δεν κάνω, πέρα από τα βασικά που μου δίνει δυνατότητα το Davinci Resolve. Βελτιώνει την ποιότητα αλλά δεν κάνει θαύματα. Σε αυτό το video η πολύ δουλειά ήταν στην βελτίωση της ποιότητας των φωτογραφιών, γιατί ήταν πολύ χαμηλής ανάλυσης και κακές.

Indeed, the whole photo in the state shown in the video needs more attention. I already have addressed some issues like the legs that you mention, the floor that it is like they are floating on a grey background etc. In general, photos that are in bad shape, it is better to generate more ai images with low denoising strength to improve it step by step in a better and more controlled way.

@@nikolakakism thank you for your advice ,yes exactly, I had few old photos with many missing of bodies and agree with you

Αυτή τη στιγμή δεν είναι δυνατόν να παραχθεί με τεχνικές ΑΙ παραπάνω από 2-4 δευτερόλεπτα απόσπασμα video. Επίσης σε περίπτωση που υπάρχει ένα ή ακόμα χειρότερα, πολλά πρόσωπα το αποτέλεσμα είναι κακό. Η σκηνή προς το τέλος του video που 'ζωντανεύει' η φωτογραφία της τετράδας μπορεί να παρήγαγα και 200 τέτοια αποσπάσματα για να βρω κάποια ανεκτά και αυτά απομονώνοντας το κάθε άτομο και μετά βάζοντάς τους πάλι μαζί. Κοντολογίς είναι σχεδόν αδύνατο με τους τρόπους που χρησιμοποιώ. Αυτό θα μπορούσε να γίνει αναπαριστώντας και καταγράφοντας την εκτέλεση με πραγματικά πρόσωπα και έπειτα με τη χρήση των σκηνών αυτών να οδηγηθεί η παραγωγή του video με face swap ή και με παραγωγή εκ νέου από AI, αλλά αυτό δεν θα είναι απλή παραγωγή αλλά υπερπαραγωγή.

Thank you, I appreciate it a lot!

Να είσαι καλά και καλή χρονιά να έχουμε. Δεν την είχα ανεβάσει, αλλά το έκανα τώρα στο community tab του καναλιού. Επειδή η ανάλυσή του μειώνεται αυτόματα την έβαλα και στην ιστοσελίδα μου. Έβαλα το σύνδεσμο στην περιγραφή του video.