Mutinate's avatar

Mutinate

David Haden
165 Watchers178 Deviations
16.3K
Pageviews

Three free ways to use the free Intel OIDN denoiser, on a desktop PC with DAZ Studio or Poser 11:


1. Inside DAZ Studio, run on the renders from the stack - https://www.deviantart.com/mcasual/art/denoise-images-using-Intel-tm-Open-Image-Denoise-802556617


2. Standalone with batch processing, but uses an older version of OIDN - https://github.com/chr-9/oidn-gui


3. Standalone with drag-and-drop, uses latest OIDN version, but has no batch - https://github.com/DeclanRussell/IntelOIDenoiser/releases


Poser 12 / 13 and 'subscription Vue' users don't need these, as they already have OIDN built-in to the render process.


OIDN is the best, because Intel trained it specifically to clear the grain and fireflies from 3D renders. It means you can do fast and somewhat grainy renders, clear the grain with OIDN, and thus save a whole lot of render time.

Join the community to add your comment. Already a deviant? Log In

My suggested ten-step workflow for AI image generation, post-work, posting and culling:


1. Work out your finished and refined prompt, in some other session. Then, in a new session, use it to generate 1,000 images in batches of four (i.e. 250 clicks, which is free with Playground.AI where you get 1000 free images per day).


2. While doing so, locally save perhaps 140 of the images that visually 'work' (i.e. maybe 15%). Also save the very occasional images that have nice elements that might be easily cut-out and pasted to other pictures.


3. Of these suitable saved pictures, use your judgement to select perhaps 20 of the best for light Photoshop post-work. IrfanView can display huge thumbnails in a Windows folder, which is useful here. Make a new 'Sources' folder for the selection, and also other folders called 'Upscale', 'Discards', and 'Finals'.


4. Upscale your choices with Gigapixel AI (ideally not with its 'Art' model, which I find never works well) to 2x. Put the new up-scaled .PNGs in the 'Upscale' folder.


5. Do the required Photoshop post-work for an hour or so, working on the up-scaled versions, looking carefully for glitches (but also not stressing too much, and working through the set with some speed). Liquify, spot-heal, clone, then save and over-write.


6. Expect a half-dozen of this selection not to respond well to your speedy Photoshop post-work. Put these into the 'Discards' folder.


7. You will probably now have a set of perhaps 10 to 18 final pictures which work 'as pictures', and which don't immediately scream "AI!!" to 80% of viewers.


8. The final pictures are still in 2x .PNG file format, so now Photoshop's Image Processor batches them to .JPG files with a quality setting.


9. Upload these lightweight .JPGs to DeviantArt, then make a named Gallery folder and put them in. Ideally, place them under Creative Commons Attribution Non-Commercial.


10. Wait for enough "Likes". After a while, these will effectively get you a 'community vote' on which pictures in the set are the most pleasing. Go back and consider culling the set, removing pictures with no likes (after due consideration of their worth as pictures). The remaining pictures in the set then can be considered for later proper over-painting, to work them up so they have a real 'painted picture, not AI' look.

Join the community to add your comment. Already a deviant? Log In

Off the top of my head, in five minutes, some reasons to play around with what AI can do...


1. It's really free (e.g. if you use Dream by Wombo, Playground .ai), which is nice in these difficult times. Let someone else (venture investors) pay for the graphics cards and their huge electric bills.


2. It's new and fun, and getting better each week (e.g. the big improvement in generating hands, which arrived in March 2023). Let's see what it can (and more importantly, can't) do. Where are the gaps in which real 'hands-on' artists will still be able to work and (maybe) get paid? Starving artists need to know such things.


3. It lets me visualise a concept, fantasy race (e.g. "Imp Magicians of the Dewlands") or place (a Syd Mead spaceport), work out the looks. One day I might take those ideas and paint or draw the best, with some added twists and combinations undreamed of by the AI. Now is the time to make and stash such images - AI is a fast-moving target and will never be/work like this again.


4. Personally, I'm able to select just the best from my session of AI-generated pictures. I'm not someone who has to post 30, 50 or 100 nearly-identical pictures. Thus, I (hope) I'm not imposing too much on others when I post an AI set.


5. Speed. Relatively quickly generated, swiftly selected (maybe one-in-30 is good enough to even consider), and then relatively quickly worked over in Photoshop. Many people are short of time. One E-on Vue sci-fi 3D picture may take 12 hours to approach completion, and a lot of runtime-bashing and scene-wrangling. An AI set with a similar setting might take two hours, and result in a set of eight views.


6. AI can (with a little tweaking and prompt-crafting) gets me new work "as if, sort of" by dead artists. Such as Moebius and Syd Mead. Not perfect, but it satisfies the itch.


7. The lighting and depth-haze are always perfect, even if the construction and faces are not. Colour choices are not bad either.


8. AI sometimes introduces me to picture types I'd never have otherwise considered, and does so serendipitously. A high-end doll-art photography style used for fantasy pictures, for instance. Who knew?


9. The image sets that result are rather beautiful, if you're doing it right. Which still doesn't mean you're getting into real places where you might sell for fat cash - white-walled galleries, festivals, local annual and biennial art shows, regional museums etc. So it makes no difference there. They just despise anything 'digital', regardless of it being AI-generated or just the good old 'digital art' that's been around for 25 years now.


10. It's naughty. We need more naughtiness in the fields of creativity.

Join the community to add your comment. Already a deviant? Log In

Is there a more stable way to upload pictures to DeviantArt?


80% of the time the useless official upload page fatally crashes, immediately on trying to upload a picture. Drag-and-drop, or "find file"... the same crashing problem.


Just trying to upload standard .JPGs, 500kb or so. Not huge .PNG files. No such problem is encountered with any other website.

Join the community to add your comment. Already a deviant? Log In
The vital DeviantArt Filter Web browser addon has a wholly new way to filter a user, in version 6.1 (the current version). You now right-click on a search result, "Create Filters..."  The little 'X' in the corner of the result, that used the way to permablock, is now gone.  Adding permablock is now now slower and clunkier than before, but more granular.
Join the community to add your comment. Already a deviant? Log In
Featured

DeviantArt Filter changes the permablock method by Mutinate, journal

How to fix 'missing formulas' in Mandelbulb by Mutinate, journal

How to rescue a picture from the DeviantArt Stash: by Mutinate, journal

Installing brush sets in Krita 4.x by Mutinate, journal

From Zazzle to Redbubble by Mutinate, journal