Tip how to dramatically speed up rendering time with PP (50-100x)
FP_b647bf2dae
Posts: 38
I am an impatient man. 10 min rendering seems like an eternity! But I found out how to get pretty decent, borderline amazing results, with just a few minutes of rendering.
First, both images were rendered in about 40 sec to 1min. You think I have some crazy machine, but no, I have HP Pavilion with NVIDIA GTX 1660 TI from best buy when they were on sale last year.
The images below are 3200x2400 with a lot of details, click on them to see I am not kidding.
Note: I am writing the guide, so come back a bit later.
Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
...waits to hear about trick...
Was there a folow up on the tip yet? lol
Redacted
Hold on your horses, of course, there is a guide coming, just give me a little time!
LOL - Why bother posting until you have the info?
It's called suspense.
I'm dying to know!
Bookmarks this thread
Sorry my daughter came and we went watching Archive 81, so yeah, in some way it's suspense. Anyway back on track.
Those images above were all rendered incomplete and in a small resolution (800x600 pixels, then stopped before 1 min). Then AI process was used to "finish" them.
This is the original rendered image from the top of the page in its entire 800px glory:
Note: I am redoing the half-assed bad-lit image with a character I posted yesterday at 3 a.m. It wouldn't look good even in a full render.
The next step is PP magic ingredient. Let's mess with the pixels using AI.
There are many AI models, since the TensorFlow and whatnot. I am not saying one is better than another, (I haven't tried them all) but the biggest difference is the speed and ease of use. You can go from the python command line to a full GUI in any OS. It is obvious what I'll choose, I am not torturing myself.
So we have a half-baked image and I am now throwing AI model on it the simplest way possible, using AI Photo & Art Enhancer with 3D HDR Art model. (Windows only and needs at least passable GPU, which we all probably have anyway), see mediachance.com
After about 30 sec I got this: 3200x2400
Notes and observations (and ideas from comments):
Architecture can survive a lot, so 800 resolution is fine, and you can easily stop it in the middle of rendering while still having a noise.
But figures are a bit different, a render noise on the face would transform into a blurness so the approach would be to have full render in a small resolution. Then D&D it to AIPAE, and try any of the supplied models. The 3D HDR Art is specially built for this type of trickery although it gives a slightly cartoonish look. The second and third models do work for 3D in various degrees of softness, so maybe a blend with 3D HDR can be explored.
The DAP models are for DAP paintings that enhance brushstrokes. You can also combine stuff.
2. The Face.
As predicted by the comments, AI upscaling faces is rather challenging. Unlike the above image that was rendered half-assed then somehow made into a full-fledged image, faces are finicky.
The first trial was to render a scene in smallish 900px resolution then try to upscale. In this case, the 3D HDR Art preset didn't work that well on the face, resulting in a cartoonish look (unless that's the goal :) ). After trying the first 3 presets (ignoring the DAP as they are for something else), I settled on the third: Photo, 3D Graphics, Soft. Although the second was also somehow acceptable as well.
The best way was to take the 3600x3600 result from the AI and then shrink it down to about 60% 2160x2160 and use some sharpening. Yeah, the image looks a bit like a barbie doll - it is likely due to the unappealing skin shader and simple flat light.
I also tried Gigapixel Ai and the results were worse. Gigapixel is obvioulsy trained on photos.
Interesting note, if I used nvidia or intel denoising on the 900x900 rendered image then tried to upsacale it in AI, the AI could not do anything the result was as blurry as the original - the denoising does something to the image, turning it unacceptable to the AI trained data. That's probably a good lesson, no post-processing before AI, rather after.
You can see the images by clicking on the last 2 thumbnails. Original and the best attempt with Photo, 3D graphics Soft and 60% sized down.
My next idea is to render it in large size, but then stop it soon and see how AI deals with that.
It is almost scary what you can get from AI, left is the original, right is the AI finished image. If that's not cheating, then I don't know what is. You would swear this has to be created the other way as a click bite, rendering a nice sharp image and then blurring it on left, but no, I am feeding it garbage and get a small miracle as the output. You can even see what it added inside the alcove - where is basically a mush in original rendering.
Maybe you just used poor examples but, for me, the girl looks all wrong. The skin and shirt shadows and highlights in particular. I'm not a stranger to this method as I used it previously before I upgraded my GPU to a RTX 3090. Back then I had a GTX 1070 and I used Gigapixel AI to do what you are doing here. Some images were suited to the AI, others were not. It is a bit like the denoisers - some people are happy with the results but I have not seen an example I would be happy with and I don't consider myself to be very demanding at all.
I downloaded the trial version and tried several renders and photos with many of the program options. So far, I didn't get any results that I would consider at all acceptable. Things were either over simplified with extreme noise reduction or ridiculously over sharpened. I didn't see any options for controlling the amount of the affect that was applied.
Technically this is a post work tip as it doesn't speed up any render time?
I use AI upscaler for photos and videos. They are great for objects for not great for human characters. Human brain is just too sensitive in picking up irregularies in images of another human.
I just use this product and it's been a lifesaver for the Viewport Iray especially.
https://www.daz3d.com/render-throttle-for-iray
It has a script to add the buttons for rendering the previews and final renders, in various speeds, to the toolbar. So it's very convenient.
I have not used the regular DAZ Iray Drawstyle preview window, which seems to always be bogging everything down, since I purchased this tool.
I use waifu2Xcaffe to upscale and Flowframes to interpolate videos both were free when I grabbed them
Thanks anyway, it's only a solution if you spend $28, and the program is too auto-tuned, no real customization whatsoever... I already have an upscaling AI-based program, I wish you could have used a freeware, and not a shareware solution.
Doesn't look any better than rendering for just a short time to get a super grainey image at the huge resolution and using a the nvidia or intel denoisers. I'm running a 1660ti as well so...
Might work with extremely detailed background sets... but looks horrible on that female figure in the picture above...
So, might be a great help when you do your renders in layers.
I mean... the Nvidia denoiser is using code to interpret the noise pixels into useful data, also.
AI doesn't care. It's not saying, "Oh, that's Bob's forehead", searching the internet for foreheads for people who look like Bob, and then making up a forehead pixel. It looks at the surrounding pixels with useful data, makes an educated guess about the noise pixel.
After playing with this some more, I found that the "Photo, 3d Graphics, Soft" tool gives much more natural looking results than "3D HDR Art". "3D HDR Art" is just way over the top, unless you are trying to cartoonize your render.
I am intending to buy Dynamic Autopainter one of these days but I like the painterly options, never looked at the HDR one I admit
I like Dynamic Auto Painter.
I think you are right, that is what the model aims for, it is after all called 3D HDR Art. I do like it, but I never try to go for photorealism, rather a bold 3d look. It would be interesting to render one in 3D HDR Art, second in Soft and layer them in photoshop. It could be an interesting way to adjust the sharpness.
No arguments there... I did half-assed it for sure.
FP speed is king. I still do OpenGL renders in Daz Studio because I'm in a hurry to get my stuff done and 3Delight and Iray are taking tooooooo long. I dont ever go for realism in a comic strip. If Iray is giving you headaches I would recomend Blenders Evee which is nice and fast. Or just do OpenGL and use post editing for your 3D renders.
I played with Blender a few months back and the interface improved a lot from what I remember.
In fact it was very enjoyable even if a bit confusing :)
I just looked at the HDR product by Mediachance the OP is using, it has a free trial too
I might try it later
it is cheaper than Topaz Gigapixel AI after all
I'm not that fond of Topaz Gigapixel AI. I'm pretty unhappy with Topaz as a company. They used to be good, but their AI apps seem to have a lot of bugs and everytime they update, they introduce new ones. They now charge yearly fees to update. I recommend looking at alternatives.
Subscriptions are evil, but there is a difference between subscription and renting. I understand maintenance subscriptions, but the ones where if you stop paying your software is dead - that's not a subscription, that's renting. Imagine if you have a subscription for National Geographic and then stop paying one day, and the next day a guy came and takes away all your old magazines.
I've also been trying to find decent alternatives but if I just go by comparison reviews, Topaz seems to win every time.