I did not write to regular users like me. Troll does not pay 200+ EUR for the product he wants to use. Topaz has to address all these complaints and there are many of them.
I am software developer with 35 years of experience too.
I just want to see the damn skin fixed in the videos and not airbrushed.
I wrote about it long time ago. ZERO reaction from this company.
Anyway. Instead of Topaz replying with humilty, I am being attacked by a few users who don’t get updates as well. Interesting company forum and company.
This is not a hobby forum, but official one. Where are employees?!
Beg them for fixes, pay them 200, 300 USD and don’t get significant updates for months.
You pay a lot of money and get… a few updates a year. This is misunderstanding.
Should be updates every week or every 2 weeks, to every of these products.
end.
There was a time that the team did try and release a patch update every two to three weeks to address bugs and issues. At times this was achievable in the older app Video AI when it was less robust. However, at times this rapid schedule also meant less time for QA testing and a lot of things were missed.
Most recently, this was seen with the inclusion of Neuroserver a few months ago, with the rapid release of 1.3.0 - 1.6.0, and the call was made to shift and try to focus back on knocking out long-standing bugs and issues that have negatively impacted users for a while. Part of this slowdown has led to two versions of 1.7.0 beta releases for further testing, and they are still working on this.
The skin texture aspect that you referred to is still on the research team’s list to see what can be done with further training of the models. This is taking a lot of time, just like the improvements for faces did, going from Iris to Rhea and, most recently, Proteus Natural.
about bugs, please also keep an eye on program is “freezing when closing”, also happens with 1.6.1…too often. and i had issues when using “square pixels” and SLP, I had cases where selected upscale factor was not applied, fo example i selected 3x but it did soemething like 2x, but in the output rendering tab, the shown resolution was 3x.
If you still have the logs from the upscale attempts, can you send those in for review? Ill try and replicate later today with various source files that are square pixel formats.
Thanks for the answer. Please talk to the managers to bump up the whole-body-skin-restoration on the list. You have faces restored (very good stuff!) but the body skin is just bad. The result is - clear improved face installed into blurred skin without much of a detail.
I undertand there could be a slider “generative skin” to improve the skin this way if the original details are missing.
Gigapixel’s Wonder 3 is really a good thing. It stands out.
I hope to see updates coming to this model as it is a good working model.
The same with Topaz Video Precise model - this is better than anything before and the improvement is clear., compared to HQ and other modes
I hope to see next iterrations coming soon.
I would be more excited about Starlight if it was something that I could run on my hardware. I mean, yes I can run it, but even though I originally built my computer with the intent of running TVAI for days at a time to complete one video, I don’t think it’s worth it to take more than one week for one video. And I don’t have the money for a new GPU.
So yeah. I use Proteus, and am about to start work on doing a whole show with Proteus → Apollo → Chronos to see if if turns out how I want, but that’s all years old models at this point.
I’m curious…what would you like Proteus to do that it doesn’t do today. Keep in mind that Proteus is a general purpose model. Until the Starlight models appeared, it was the best upscaling model available. Sure, there are situations it can’t handle (severely out of focus or blurry faces for one) but it really is excellent at cleaning up noise, recovering a lot of detail, and sharpening the output. It is still my goto model for most tasks, and my favorite for polishing the Starlight output.
Have you tested the Proteus Natural variant that was released in version 1.2.0? That was the most recent update to Proteus and was meant to help with heavily compressed sources, reduce artifacts without overly processing the source.
I have tried Proteus and specificaly Proteus Naturals to polish up SLM, and I still find Iris MQ set to manual and only modifying FC to 50 that gives me the best results. Proteus is very similar but when I look close it seems to leave some jagged edges where Iris MQ does not, plus the timeframe between the two is much better with Iris MQ.
I don’t use Starlight Mini…never liked the soft output. My goto model has been Starlight Sharp. Unfortunately there are some bugs in neuroserver that stops Starlight Sharp, Fast and HQ, as well as Hyperion 2 to not run on either of my systems. I have been using Starlight Precise, followed by Proteus to denoise the output. I had tried using Nyx, but my content has a lot of starfields and Nyx kills all the stars unless there is some other object (like a nebula) on screen. Even then, it suppresses the stars more than a quarter screen away.
I have and it’s an improvement over regular Proteus.
It’s been a minute since January though, and since most of the conversation here has become fixated on Starlight it seems the other models have been become ‘less popular’.
So I guess I just want to say that I hope there are still new models in development for MQ+ sources.
And I believe that MQ+ sources are where Video AI can really shine because there is enough data to work with and possibly produce amazing outputs.
@kyle.topazlabs if it’s correct that Video 1.7.0 includes a better version of SLP2.5 with less visible texture change, is it planed to release that new version on the web-app at the same time? and is it marked as a better version like 2.6 or do we have to try it out for ourselves to see if the quality has changed?
