[Paid Help] – $500 or higher for better video restoration workflows

Topaz Video General
1

I would like to hire someone to provide workflows for upscaling and restoring a classic Chinese TV series, “San Guo Yan Yi” (known in English as Romance of the Three Kingdoms).

src-2
src-21920×1080 917 KB

I currently have a baseline restoration result, but I believe it can be further improved. That’s why I’m inviting you to help tune a better restoration solution.

If you are interested, please try processing the test clip below (duration: approx. 140 seconds):
[src input url]

https://drive.google.com/file/d/1P_VMursrB1oPPbhOWjsHK2cEYvRShdl4/view?usp=sharing

Please note: the source video has been badly upscaled – this wasn’t done by me; it’s the quality of the source I received.

Baseline restoration demo:

[baseline demo]

https://drive.google.com/file/d/1UwYQ4T98xGV1QnLOP-ofNdmG89MCGbeT/view?usp=sharing

Comparison (baseline vs. source):

https://imgdiff.net/s/d8a07b1f6a25ce1b0dc9eb3666ff094c

https://imgdiff.net/s/b45499d0efed992efe1fe3f5565cffb0

When tuning your restoration solution, besides improving general background details, facial details, and subtitle/text quality, please pay special attention to dark edges and purple fringing (as shown in the examples). I hope these artifacts can be removed more cleanly.

Here are a few screenshots with these issues highlighted. Candle flames, for instance, often show a dark line on one side and a purple fringe on the other – extremely annoying.

企业微信截图_17799697362694
企业微信截图_177996973626941920×1080 866 KB

企业微信截图_17799698086766
企业微信截图_177996980867661920×1080 795 KB

You are free to use different separate applications or models in combination – whatever you have access to. If the restoration quality meets my expectations, I would be happy to pay for your workflow. Starting at $500 USD. If you believe your workflow is significantly better than the baseline, the price is negotiable.

Please feel free to show your demo on the test clip. Thank you!

2

Restoration (remastering) has nothing to do with Topaz Video.
What you are doing is creating a personal version of this video with generated faces that clearly look AI generated.
Restoration budget is usually over $100k and involves scanning of the original source at 4k resolution.

3

We have different understandings of “restoration.” What you’re referring to starts from the very beginning—the original film medium—and involves rescanning it for digitization. That process is indeed very expensive. But my current need is different: I want to take an already scanned digital version and use software algorithms to improve its quality.

4

Also, as I mentioned, I already have a baseline restored version. If you’re interested, please check the quality of my baseline version first.

5

What you are doing with Starlight models is not called restoration in English. You can check any dictionary if you don’t believe me.
Artemis antialiasing and Proteus 2 are the most conservative models, so you can try them if your goal is to reduce compression and oversharpening artefacts, so the video won’t have a lot of distortion or AI processed look. The first step is to downscale to the original resolution or a little bit higher. Gaia 2 may be of use as a single pass at 100% to sharpen the edges without a lot of distortion if the downscaled video looks too blurry. Proteus 4 with 100% original detail recovery may help as well.

6

Hello, the quality of the source material is crucial, and anything that has already been upscaled is always a poor starting point. If there’s a better source available somewhere, try to obtain it. As for upscaling, we’re happy to help with tips and advice free of charge — you can spend that money instead on getting the best possible source material. :wink:

7

I completely agree with you: the quality of the source material is crucial.

But for Romance of the Three Kingdoms (1994), this is the best quality source currently available in the industry. I work professionally in digital restoration, and I know exactly where this source comes from. Unless we go back to the original film medium and rescan it at a higher resolution, there is no better source than this.

What I need is a workflow tuned specifically for this video — one that can produce better restoration results than the baseline. I don’t need general advice, which may work or may not work. I don’t have enough time to check all the suggestions. This task requires time and computing power, and getting paid for that is only fair.

8

Honestly, I’m not a native English speaker. If you have a more accurate professional term to recommend, please let me know. Thanks a lot.

9

If there isn’t a better source, then that’s clear — that’s just how it is. However, you wrote that someone else had upscaled the source poorly, which is why I wrote this, assuming that an original source exists that hasn’t been poorly upscaled → click :wink:

10

His source is probably based on that Blu-ray. It seems the movie/series was shot on interlaced PAL SD video, not on film.

11

“Romance of the Three Kingdoms” has been adapted many times over the years. The version I’m restoring is the earliest one, filmed in 1994.

I’m 100% certain that the test source I’m providing is the highest-quality digital version of the 1994 adaptation available today. Searching the internet for a better source would be a waste of your time.

Just to clarify:
When I say “the source video has been badly upscaled,” I mean that the original resolution was most likely standard PAL (720×576), but the rights holder has stretched it to 1920×1080, resulting in noticeable aliasing and jagged edges.

12

The link you provided is for the 2024 remake of Romance of the Three Kingdoms, not the 1994 version I’m restoring.

13

NOT TRUE!

14

The DVD-Set would be a way better than this worse 1440x1080 upscale is this the 1994 one?

Well, that’s for you to decide, but I would never take on a project of this magnitude (84 episodes) if I already knew the final result would be mediocre because I didn’t have the proper source material. Badly already upscaled source is the worsest condition you can have.

I would take another look to see if you can get the DVDs from somewhere after all — used DVDs would be fine too. Otherwise, I’d start a different project for which you already have the source material. I noticed there are some poor-quality upscales of it on YouTube. Did you get your source from there?

15

Here is my remastering attempt! Secure File Sharing - Filemail

Sanguo 1
Sanguo 11920×1080 971 KB

Sanguo 2
Sanguo 21920×1080 1 MB

Sanguo 3
Sanguo 31920×1080 1.06 MB

I think it looks quite good, though the Starlight tiles are pretty visible in some shots when using this workflow…

16

For the Chromatic abberation (coloured edges), you could export it into Tiff or PNG and run it through Adobe Camera Raw (or other software that could solve this) to remove it.

But i can spot it in the source image in the post (colored edges).

I would apply some noise too to remove the AI Artifical look.

17

of course NOT!

Due to the client’s authorization restrictions, I cannot disclose the source of the source. But I’m 100% certain that the test source I’m providing is the highest-quality digital version of the 1994 adaptation available today.
As for the 28-DVD9 version you mentioned, its image quality is very poor. This blog post compares the quality of various versions circulating in the market:

If you’re interested in this project, please start with this source. Searching the internet for a better source would be a waste of time.

18

please show a demo.

19

Thank you very much for the demo video. But unfortunately, your restoration results do not meet the standard of the baseline I provided.

Specifically, while the image clarity is very high, the AI-generated artifacts are too obvious, making the footage look less natural than the baseline.

In addition, the dark edges and purple fringing that I’m particularly concerned about have not been effectively removed.

20

By the way, I think using only Topaz Startlight models will most likely not achieve better results than the baseline. So I suppose we need to use different separate applications or models in combination – whatever you have access to.

I think some pre-processing modules need to be considered, something like this: