Render with ultra high resolution
mikek
Posts: 195
in The Commons
I would like to render with the size of 16000x16000 but daz limits the input to 10000. Is there a way to go higher? My first thought was to split it up and do four images so for each quadrant one and merge them later but that doesn't work while keeping the lens distortion like it would be with one camera. Was anyone able to do iray images with higher resolution than 10000?
Daz 3D is part of 
Connect
DAZ Productions, Inc.
224 S 200 W, Suite #250
Salt Lake City, UT 84101
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
Click on the cogwheel that appears on the "Pixel size (Global)" setting. Change the max from 10000 to something higher, or uncheck the "Use limits" checkbox.
Click the gear in upper corner of the render size parameter and turn off limits.
Click the gear icon next to Pixel Size, choose Parameter Settings..., uncheck Use Limits, hit Accept. Now you can enter your exceed the limit. Be aware that you will need tons of RAM to render an image that big. I just tried 12364x16000 and it took up 24GB of RAM for a single G3 character with hair (nothing else in the scene).
Iray is not designed for print resolution images. The higher the final resolution the more vram it needs for the frame buffer and working space. So I feel you will need a very good card to go that high. A better render engine for that would be Cycles that is tile based, so the needed vram is independent from the final resolution.
Nice so it was just a limit. Thats definitly easier than trying to use 4 cameras.
Ram I have luckely just enough for my scene as it sits at 27GB now. The gpu went on strike but thats something I expected.
Thanks for the fast help from everyone here now I just have to wait.
Tile based would be nice but if I switch the render engine I will have to redo the materials with the result of it probably not looking as nice. I will test for now a bit how far the cpu gets me and how many days it takes to finish. Its only this image I need with that resolution at the moment.
Just out of curiosity - what are you trying to render that needs such a resolution?
"Is taxing for the system to render large images" is not the same as "not designed for large images".
------------------------------
Tiling as such, moving th camera, would not work as the lens distortion would not match - what might work better is using the offsets in the camera to render different segments without moving the camera in space (though it's not something I've tried). You do need to bear in mind texture sizes - a 12,000 pixel image of a figure, even a full length portrait, woudl quite likely be a limited gain over a smaller image given that most textures for head and body are only 4,000 pixels square or thereabouts.
You could use the lens shift feature in the camera settings. Go into Iray viewport preview and use the Lens Shift X slider to see what it does. Essentially it moves the frame without changing perspective.
It's somewhat like those high resolution panorama images where I take a full scene and the user can then zoom in where he wants. The high resolution is so the zoom won't get to pixelated.
Concerning character textures thats no issue as I'm using a scene only with background and some items.
Ok moving the view around with lens shift works. Will have to test if the merge will be good. That would allow me to do it on the gpu.
As stated above, unlock the limits and set your size. During test renders, render smaller so its easy on your system. Make sure the aspect ration is locked.
When you are ready to render big, click on the Spot Render Tool. Open the Tool Tab. Check the option to "render to new window"
Now you can spot render different areas of the image without having to move a camera around. Iray will render small areas faster. I do not know if rendering this way will reduce memory consumption or be any differnt from rendering the entire scene in one shot.
As an alternate option, you can try this software which is designed for blowing up images for large print - https://www.alienskin.com/blowup/ - It is not freeware.
Greetings,
I wonder; if you set your resolution to 16,000x16,000, but then manually use the 'spot render' tool to render subsets of the frame, does it only use enough video memory for that subset? This would allow you to manually 'tile' render your scene by doing subsets that are small enough to fit in vram (and thus render fast) but then aggregate them in Photoshop, Pixelmator, or some other image aggregation tool.
-- Morgan
Edit: Ooooh, ninja'ed by @Mattymanx! (Love your character renders, by the way! :) ) Glad to know that I'm not the only one with that idea...now to write a script to automatically do that for me. :)
Thanks, I couldn't recall the name but that's what I meant
Thank you very much sir! :)
I think that is in there to mimic a tilt shift lens - https://en.wikipedia.org/wiki/Tilt–shift_photography
...this is why I need to build that 128 GB work system The only GPU card that could handle such a load is a 5,000$ Quadro P6000. Keep in mind that part of the 24 GB used is just having the Daz programme and scene itself.open. This is also why I really wish we had Iray Batch rendering.
...if your not looking at producing absolute photo real images (and Daz Iray still falls short in that department due to available materials and the fact it is a slightly hamstrung version compared to the full standalone engine), Carrara has a tiled render engine that produces excellent quality, and can use most Daz content (save for G8). You can also easily pose figures/props in it unlike Vue or Blender. I find the shader controls to be more intuitive and easier to work with for converting shaders compared to node based ones like Blender and Daz use. Also Carrara can suport up to 100 CPU cores/threads. (again the basis of that 128 GB work system I designed as it has dual older generation 12 core Xeons for a total of 48 threads).
...compositing individual small rendered segments sounds like it would be a total pain compared to multi pass rendering
The Spot Renders will all be the ttoal size, and will have alpha channels if saved to tiff or PNG, so they justs tack - no "compositing" as such required.
Yse, I would assume so - thats certainly the context in which I first encountered it (with respect to the modo renderer)
..so how does that save in resoruces if you are still rendering the entire scene? Spot renders also work off whatever is seen in the viewport and don't use the render settings tab so I don't see how one would get higher resolution sizes with them unless they had a very big display.
When you Spot Render to a new window, what you get is a small square in a big field the size of the total assigned render-size. The stuff not rendered is left transparent. YOu can then layer the spot renders in Photoshop and they line up perfectly.
(and spot renders do run faster than the whole render. I've not done 16k renders but I HAVE done 6k renders and had to rerender small bits as fixes and they completed much much faster.)
I've done this before, when I needed a print resolution end result. If you have Photoshop, (at least in CS6, can't vouch for CC,) there is a script the will stack files into one image, with everything exactly where it belongs in the final image.
If I end up doing a corrective spot render to add to the "stack," I open the file and add a temp layer and fill it with color. I then "group" the two layers, and drag the group into the stacked image. That allows me to align the group to the edges, putting the new layer right where I need it. I move spot render layer out of the group and then delete the group with the temp layer.
...don't have PS or CC. Also again how can you adjust quality and final image size settings if you are not using the render settings tab since the spot render tool only works off the viewport? To spot,render, you have outline the section of the scene you want to render in the main viewport so how do you keep that consistent each time since the spot render tool works like the "square/rectangle selection" tool in a 2D programme.
Well, whatever layer-handling software you do use will work the same way. The way you do this is you select the Spot Render tool and then you go to the Tool Settings pane (which you may have to open in Window -> Panes) and then you select 'New Window' in the Render To part of the Tool Settings. And then it DOES use the render settings tab size and other elements. And as long as you're able to use software that handles layer and layer transparency (and there's a lot out there these days?) it doesn't matter if you quite line up the Spot Render selections-- they can overlap and it won't matter once you stack them. Here, I can show you.
...ok, the size limit for uploads seems to have dropped a bit, but I've included the associated Spot Render, at least. You can see if you open it that it's a small square in a much larger transparent image. I load up my original render and then I drop that on a layer above the original and it matches up seamlessly. You can do the same thing to render something in roughly selected blocks and then stack them in layers in whatever you use for layers.
The spot render tool will always obey the render tab settings. I have never seen it do anything less. Since the output is saved, you just need to eyeball your previous render to see where you left off. If you have the aspect frame on in the view port then you will know where the edge of your image will be.
Set up the full scene in the viewport. Set the final resolution in the General render settings. Select the Spot Render tool. Open Tool Settings and select the render to new window option. Optionally, turn on the Rule of Thirds guide. Drag the Spot Render tool over the scene, making a mental note of where you'll need to overlap. (The Rule of Thirds guide can be real handy for that.) When the spot render finishes rendering, drag the Spot Render over another area of the scene, one side overlapping the area you just rendered. Repeat that until the entire image has been rendered.
Hope this helps.
ETA: Don't forget to save the scene before you start rendering. If something goes wrong and DS crashes, you can start over with all the same settings. (I bring this up because I am really, really bad about saving before rendering.)
Ok, I just did a quick test to see what the memory consumption would be between the different sizes.
Render test conditions - Naked G3F with hair and skin. Lighting was ONLY the default HDRI map. Background was not rendered. (Rotated the dome 90 degrees for better illumination)
Render 1: 1000 x 1000 - 4GB of system ram used by Daz Studio while rendering the scene.
Render 2: 10,000 x 10,000 - 12GB of system ram used to render the scene.
Spot render: 1000x900 - 6GB of system ram to render her face and some hair.
As expected, the large 10k render was very slow at rendering. Iterations were being counted one at a time. The post render was faster of course.
While doing the spot render method may seem tedious at first, its main bennefit is that it does not have to be done all at once so you dont have to have your computer tied up doing one giant render. You will have to do some tests to determine what is the best approximate size to break it down into for spot rendering though.
...still sounds sort of like having to paste a jigsaw puzzle together in post which to me seems more difficult and error prone than multi pass rendering.
If say I want the final image to be 18,000 x 12,000 like for a large gallery quality print, using the rule of thirds guide or splitting into 4 quarters would still mean fairly large segments need to be rendered, so I would have to break them up into even smaller sections with no guide to use. As I am old, my vision is....well...not so sharp anymore these days, so keying up the individual pieces and determining where I left off rendering would be a lot more challenging for my tired old eyes than simply combining full sized layers of the entire scene together or simply throwing a lot of memory at it. There are times technology needs to compensate for our human imperfections, hence the big memory dual CPU system I designed and hope to scrape up the money to build.
Oh, and I always save before running a render process, even a test render. My experience from back in the old days when I was still working on a 32 bit notebook was an excellent teacher.
It really doesn't have to be precise. But it's fine to do what works for you, too. It may be worth experimenting a little, though-- I get the impression we're not explaining how simple it is very well.
Ok the spot render tool works quite well. It's now running with around 15GB and doesn't drop from the gpu. Daz saving the image in full size makes the merge also very easy. Seeing how well this works I could probably even try 32k x 32k for the raw image. Thanks for this suggestion. Sharp zoom images here I come.