- 17
- 3 573
Fabio's Film Restoration Corner
Peru
Registrace 25. 10. 2019
Film Restoration and Digitization training, tricks, experiments and more related to the digital intermediate.
Color Restoration Demo: Photoshop & Nuke Copycat Feature
In this machine learning/AI demonstration, we're not examining a black and white image but attempting to restore the colors of a faded dye scan or magenta-tinted film. The subject is the trailer from "Ben," the sequel to "Willard," which exhibits the classic magenta tint. Although the scan appears to have complete information across all channels, a one-light adjustment in Resolve is ineffective, as is using the RGB mixer to transfer information from one channel to the damaged ones. Complementing this with a splitter combiner node to enhance the red channel's contrast yields a more satisfying result, but it's not perfect yet.
Next, we export the first, middle, and last frame from each shot and employ Photoshop's new neural filter to colorize these frames, with some manual tweaking to refine detection. Then, we face two options: the traditional "Match Grade" node, which creates a LUT to mimic the reference or target-with unsatisfactory results-or the "Copycat" method. For Copycat, we use the three colorized Photoshop frames as ground truths for our Resolve output, spending about 10-20 minutes on training, and finally achieving the desired outcome.
A good reference, such as a telecine or a photograph, would have simplified the colorization process, but such references are not always available. It's important to note that the trailer has not been restored or digitally cleaned, so flicker and dust may affect the final result if you attempt a similar process on your own.
Next, we export the first, middle, and last frame from each shot and employ Photoshop's new neural filter to colorize these frames, with some manual tweaking to refine detection. Then, we face two options: the traditional "Match Grade" node, which creates a LUT to mimic the reference or target-with unsatisfactory results-or the "Copycat" method. For Copycat, we use the three colorized Photoshop frames as ground truths for our Resolve output, spending about 10-20 minutes on training, and finally achieving the desired outcome.
A good reference, such as a telecine or a photograph, would have simplified the colorization process, but such references are not always available. It's important to note that the trailer has not been restored or digitally cleaned, so flicker and dust may affect the final result if you attempt a similar process on your own.
zhlédnutí: 157
Video
Film Restoration Experiment: Machine Learning in Action
zhlédnutí 790Před 28 dny
In this second machine learning test, I aim to tackle a more intriguing challenge. As evident, this 16mm film suffers from advanced hydrolysis and vinegar syndrome, rendering the initial restoration efforts inadequate. While concluding this subject, a telecine transfer of the film, recorded around the time of the original movie's release, became accessible. This telecine has its own set of issu...
Channel Alchemy: A Demo on Restoring Lost Color in Film Preservation
zhlédnutí 79Před 28 dny
While scanning footage for a restoration project at my workplace, we discovered that one of the reels, specifically the last one we scanned, had partially lost its blue channel. The signal wasn't completely gone, but it was significantly weakened. You might wonder why not use the channel mixer in Nucoda/Resolve to "restore" (or more accurately, replace) the missing information? I attempted that...
Preserving Film Heritage: The Art of Correcting White Balance and Highlights
zhlédnutí 54Před měsícem
It may seem trivial and perhaps obvious to many, but I've observed numerous scans from around the globe, including those from esteemed sources, that suffer from incorrect white balance and, occasionally, overexposed highlights. Such issues compromise every step of film restoration, especially automated tasks like de-flickering and dust removal, not to mention color correction and grading. The c...
Blue Channel Magic: An Experimental Recovery Demo
zhlédnutí 44Před měsícem
This reel presents a unique challenge, as is evident in the video below. Due to the unchecked progression of vinegar syndrome and the conditions in which this film was stored, some frames in the final two minutes of the reel have been partially lost. Additionally, a persistent flare throughout the reel further complicates the issue. I attempted the same approach as before, separating the channe...
Restoration Experiments: Crafting Synthetic Dirt Maps for Film Restoration
zhlédnutí 93Před měsícem
High-end scanners feature an innovative function: an infrared bulb that, along with RGB sensors, creates a fourth channel containing special information. This information forms a map of the "dirt" on the film's surface, which is stored in the alpha channel and then used as a basis for cleaning, either through in-hardware processing or a more comprehensive process later on. I considered whether ...
DVCPro Revival: Crafting Pristine Visuals from Burned-In Subtitles
zhlédnutí 39Před měsícem
I needed to create a new master for film preservation and distribution. The film was originally shot on DVCAM and completed in 480i in 2004. To craft the new master, I was provided with various versions of the film in different formats and quality levels. These included a DVD in Spanish without subtitles, a DVD in Spanish with English subtitles, a Blu-ray in Spanish, a Blu-ray with English subt...
Scanner Light Calibration: Quick Film Digitization Demo
zhlédnutí 48Před měsícem
Even with proper calibration of scanner lights and color channel balancing to capture the maximum information, films with faded dyes can still emerge. No amount of RGB light tweaking seems to yield an optimal restoration output. Typically, color correction and restoration are distinct phases in most facilities. However, subjects like these require a "One light" correction before any cleanup can...
Experimental Color Recovery: Machine Learning Restores Magenta tinted 35mm Film.
zhlédnutí 82Před měsícem
This is my inaugural project utilizing machine learning for film restoration workflows. As previously mentioned, the film heritage of Peru largely consists of positive prints, which have suffered various issues due to inadequate storage, including a magenta tint from fading dyes. Unfortunately, this color loss is irreversible, rendering standard color correction methods ineffective. The restora...
Film Restoration Quick Tips: Channel-Separated Grain Reduction and Restoration
zhlédnutí 244Před měsícem
During a client visit, I conceived a method to effectively utilize DVO Velvet, which can be quite destructive if left unchecked. I pondered, if Velvet is akin to a nuclear option, why not employ it to obliterate the grain, not of the entire image, but of an individual RGB channel? The final outcome is quite satisfying and extends beyond mere grain reduction.
Restoration Demo: Reviving Colors of a Classic 35mm Trailer
zhlédnutí 79Před měsícem
Drawing on the machine learning and AI training knowledge I've gained over the past few weeks, I plan to restore the original color of this 35mm film trailer. Earlier this year, I demonstrated how to scan footage with a magenta tint or color degradation using this trailer. Traditional color grading methods, such as masks and qualifiers, prove inadequate when one or more channels are damaged. Th...
16mm Jungle Adventure: Restoring Film Footage
zhlédnutí 50Před měsícem
I aim to create a video showcasing the materials I typically handle. While I'd relish the opportunity to work in a large facility with advanced cameras and separation negatives, there's a distinct advantage to my current "exhibition print" approach. Tackling film restoration under challenging conditions is an educational experience, allowing me to experiment with new workflows and ideas. Althou...
Restoration Reel: Phoenix DVO Tackles the Toughest Film Challenges
zhlédnutí 147Před měsícem
I've compiled a video showcasing some of the most challenging projects I've participated in over recent years. The focus is to demonstrate the DVO tools in action and their ability to assist restoration operators in meeting their objectives. While I embrace modern technologies like AI and Machine Learning, I believe it's crucial to remember the importance of a skilled operator's role; ultimatel...
Film Restoration Quick Tips: AI colorization with Photoshop Neural Filters and Nuke's Copycat
zhlédnutí 295Před 2 měsíci
In this episode, we demonstrate a colorization workflow for black and white and faded pictures using three different software packages. You can find the scan of the Village of The Damned Trailer here: archive.org/details/village-of-the-damned-1960_202108 #film #video #media #filmscan #digitization #archive #filmheritage #audiovisualheritage #filmrestoration #filmarchive #workflow #Filmpreservat...
Film Restoration Quick Tips: Recovering lost features with Composite mode and Matte Paint.
zhlédnutí 646Před 2 měsíci
In this episode, we create a custom workflow for recovering lost details on a rendered file. You can find the scan of the Point Blank Trailer here: archive.org/details/point-blank-trailer And this is my project file: drive.google.com/file/d/1XD1Zk2TAvC19jOZKIosoLtDgr3aPc8k3/view?usp=sharing #film #video #media #filmscan #digitization #archive #filmheritage #audiovisualheritage #filmrestoration ...
Film Restoration Quick Tips: Using Generations for a better workflow
zhlédnutí 496Před 4 měsíci
Film Restoration Quick Tips: Using Generations for a better workflow
Film Restoration Quick Tips: Stabilization using Phoenix and DVO Frame Lock
zhlédnutí 257Před 5 měsíci
Film Restoration Quick Tips: Stabilization using Phoenix and DVO Frame Lock
you have a very good explanation of your process
Thank you :)
Dont give away your secrets man! ;)
My primary role is that of a teacher, and I cannot escape it.
@@fabiocolor then if that’s the case could you tell me the best solution for removing splice bumps at edit points in Pheonix? Wrap the bump with cut points and apply Steady 2? I’m looking for the best solution. I originally applied Steady 2 to the entire film but the client isn’t pleased since I stabilized the picture too. They just want the spice bumps removed. Thanks
To accurately remove splices, reel footage must be scene-detected. In the Steady DVO, there's a specific setting for handling significant splicing interruptions. An alternative approach is to utilize the recently introduced DVO frame lock. If you wish, I can review your footage and provide some recommendations. Please feel free to reach out to me at fabio.bedoya@filmworkz.com.
good work
Thank you! Cheers!
impressive
It’s really cool how old film could always look beautiful but just didn’t have the tech to, shows how far ahead and behind they were. But seeing old music videos be remastered and looking incredible, yet different to digitally recorded videos, is a really cool thing.
Movies have way more flexibility and visual range than we usually think. Even old prints can look stunning if taken care of.
Nice!
Very interesting, but im assuming it only works on clips that have an alternate cleaner source? Also it seems like the edges become very messy. Still a really cool proof of concept though!
Certainly, additional training time and a more comprehensive set of frames are necessary to achieve production readiness.
This gonna blow soon.
Excellent video, mate!
Ola Fabio, qual GPU voce esta utilizando?
I went from a 3070 to a 4070, and now a A4000 mostly for the VRAM copycat uses a lot of resources
@@fabiocolor A quadro p5000 ou p6000, funcionaria bem com copycat?
@@marcossouza9868 Yes, it works, but it's significantly slower because Copycat utilizes RT cores for acceleration, both Apple Silicon and AMD GPUs are also slow. A good budget-friendly option would be an RTX 3060 with 12GB of VRAM.
Ola Fabio, trabalho interessante, eu trabalhei um tempo com restauração na empresa Cinecolor Digital do grupo Chile Filmes, se vc vier a São Paulo - Brasil, gostaria de conversar com vc e ver seu metodo de trabalho com o Phoenix de perto, valew
Sure write me an email to info@fabiocolor.com
Boa noite Fábio, a quanto tempo você trabalha com Phoenix?
From 2020, before that I was on Davinci Revival, so it was a big jump.
@@fabiocolor eu já trabalhei com Diamant, PFClean, o Phoenix é mais rápido que esses que eu utilizei?
@@marcossouza9868 Diamant has improved with the addition of machine learning in its latest versions. Although pfclean is no longer under development, some people continue to use it, but I never favored it. I prefer Phoenix because it allows me to handle the entire process from beginning to end, and if a tool is missing, I can create it using the available compositing tools. It also comes with Nucoda, offering a comprehensive color-restoration package.
Buenos tips!
Gracias luis
Excellent tutorial Fabio, I Love Your Channel, Hope you can teach us on Scratch removal on your future video.
Will do!, scratches are always a pain to deal with
Awesome to find content like this, I’m a digital restoration artist with 7+ years of experience, mostly on Diamant, but Phoenix is one of my favorite tools also, great tips! I found your tutorials trough LinkedIn 👌
Thank you. Hopefully, I can create more content like this in a timely manner. Many valuable tips are often only taught during personal training sessions, and most users never get to learn them. I intend to change that by making such information more accessible.
Hi Fabio, thanks for your lessons, what screen-tablet are you using?
I recently upgraded to a Huion RDS-220 tablet. The bigger screen helps a lot when doing manual cleanup. Additionally, I have an XP-Pen 15.6 Artist for when I need to travel.
thanks for these gifts Fabio!