AI DIGITAL OIL PAINTING – HOW TO DO THIS

NOTE: I keep updating this blog as new info becomes available so check out new resources in text. I have found a couple resources to share with you for getting what I consider some really interesting if not fabulous results by just following a few steps with Photoshop Beta. (BTW, I have not had any issues with the Beta version, but using PS2023 for most of my regular work.) Personally it does make me wonder somewhat why I am spending so much time doing digital painting when AI can be used to get possibly as good an end product. At this point, I am not sure that the painting aspect is that great, especially since you cannot use certain painter’s names in the AI prompt, but it certainly does have some great potential here.

The AI images shown were all generated from the right image below. The original image on left was taken at the Natural Bridge Historic Landmark in 2003 – used a 2-MG camera shot. The original image had four different Generative Fill layers added to expand the size of the original image – one for adding area below, one on top, and one on the right. The last one removed the big tree on the middle – this is a good tip – if you want something removed on your AI generated image, select it and run the Generative Fill again with no prompt and it will be removed. The images for Method 1 and Method 2 were all created using the right-hand photo already AI generated image!

METHOD 1: The top oil digital painting was created by following Marty at Blue Lightning’s steps in his video Photoshop AI Transform Photos into Oil Paintings. The technique involves going into the Channel’s Panel and adding an Alpha Channel for a partial opacity selection to use as the basis for the Generative Fill task. In Italics are the basic steps from the video. (Open Channels panel and click on the Foreground Color swatch changing H0/S0/B30 or 40 which is the Brightness – then fill the Alpha Channel with this gray color (ALT+Backspace) on thumbnail; next CTRL+click on thumbnail to create a brightness selection-ok if no pixels selected-they are there; click on the RGB Channel thumbnail and go back to the Layers Panel – since a selection has been created the Regenerative Fill prompt can now be used.) I really liked the results and it is pretty easy to do. The top image used the prompt “very spring oil painting” from the first 4 Generate iterations. Once an effect was selected by highlighting the icon in the Properties Panel, it needed a lot of clean and up. For post-processing info used in the top image, check out my short Same Natural Bridge with Different AI Oil Paint Results Tidbits Blog.

The AI generated image below was just a different icon from one of the Generate button choices for the top image. The Generate button was run two more times with the prompt “river and flowers oil painting.” See my AI Painted Image of Natural Bridge Tidbits Blog for an extensive post-processing discussion on my basic workflow. The results were quite different using the same process. I do not believe adding river and flowers made any difference on what was generated.

METHOD 2: Moving on to the next digital painting video that creates similar results but uses a different process to get a partial selection. Brian Mataish’s Photoshop Generative Fill AI Can Turn Photos into Paintings is also a really good video, but he is using the Quick Mask to make the partial opacity selection – basically it is doing the same thing as the Channel selection. (Here are the basic steps: Highlight the image layer, open Quick Mask mode or press Q, go to Edit -> Fill or SHIFT+Backspace and set to Color with the color set to H0/S0/B30 and OK – a partial mask has been created, exit Quick Mask mode by clicking Q again, and go to Generative Fill to begin process.) He goes into a bit more detail on this and it appears to me that both techniques give about the same results, but I decided to check it out myself. The same extended edge Natural Bridge image was used with the same first prompt – “very spring oil painting” and below is what I decided looked best. This image took 5 Generate buttons with the last iteration being the one I liked. That surprised me since I thought it would give as good a result as Method 1 or at least be as quick! But it finally gave a good result so it worked fine. (Dave Kelly’s videos Photoshop Beta (Photos to Art) Tips and Tricks (Generative Fill and Photoshop Beta (Photo to Art – Part 2) Workflow Change and a New Action to Go with It are also excellent so give them a watch – download his AI action set from Part 2 You Tube page which includes 14 actions using different brightness percents (double-click Quick Mask icon to make sure Selected Areas is set when using) – extremely useful! These include Brian’s Method 2 steps.)

Check out my Another AI Oil Painted Image of the Natural Bridge Tidbits Blog for post-info and resources for this image. There was this ugly blob on the right side of the image which was supposed to be a tree. Had to create a stamped layer and used the Lasso Tool to select the tree area – generated it with no prompt and got what you see in this image which looks so much better. So if you do not like the results, you can always regenerate just a selected portion of the original generated image to remove bad areas. Ah yes, I did add the swan for a focal point.

Another image was generated using the “flowers and river oil painting” prompt and Brian’s Quick Mask technique. Still had problems to find an instance I liked, so one was chosen that was just okay. The left corner was ugly so it was selected and regenerated. Then a selection was made around part of a tree that ended up in the water – no prompt info, just generated it and it removed it from the image. Next the water looked awful in the middle so it was regenerated also. For post-processing info, check out my short A Halloween AI Party Tidbits Blog which lists the resources used to get the Halloween effect. I like to do Halloween images so this was really fun – the idea came from using a Color Lookup Adjustment Layer with the orange and gray tones. Color Lookup (LUT’s) Adjustment Layers can really help change the whole feel of these AI generated images and definitely worth trying out to see what effect they create.

METHOD 3: TOTALLY GENERATED AI PIX WITH NO BACK-UP IMAGE: This process is very similar to Method 2 above except it is a little different since you just adding colors to a New File and are letting AI Generative Fill do the rest. Found this technique from Adobe’s Rob de Winter using his very short 1:15 minute video called Totally Created AI Pix – this is on a Twitter feed. If you do not have Twitter, here are the descriptions as he basically writes them out on his feed:

1. On a New Layer draw a rough outline with the brush tool and use different colors for all parts – you do not have to be good at drawing for this (stick figures he was using for a tree as shown below).
2. Go to Quick Mask Mode (Q) and select Edit -> Fill – set to Color and in the swatch change it to H0/S0/B70 (70% brightness). The lower the brightness percentage, the more the result will resemble your original sketch (or more cartoon like in his case). Press Q to exit Quick Mask Mode and you now have a 70% opaque selection.
3. Click Generative Fill and type in your prompt (like tree, or landscape with mountains – his examples). And press Enter!

This is so easy and fun. Below is what I drew (Step 1 above) before going into Quick Mask. For Step 3 “Digital Oil Summer landscape with mountains, house, and tree” were used in the Generative Fill prompt field. Two of the resulting iterations are shown below – I find this totally Amazing!

The only changes created on the above were using a Smudge Brush in the sky to smooth out the area on the right that had some sharp stroke marks on it and one of the house windows was weird looking so it was painted in straight. That was it! Not even a Curves Adjustment Layer makes it look better!

This image was created using the same file as the one above, but pushing the Generate button a couple more times to get more choices. The only thing done to this image was where a weird edge of a tree showed up in the top right sky area – selected it and ran Generative Fill with the prompt empty to remove it – that was all that was done. I cannot believe it did this from that little crazy drawn layer above! Actually most of the iterations looked great!

I am not sure which method I like best. You can see that all the generated images in both Method 1 and Method 2 showed the rather iconic branch configuration that is seen in my original image. I am leaning towards liking Method One best as it seemed to create better choices using the Channel Method, but I am sure Brian’s Quick Mask Method is probably selecting pretty much if not exactly the same pixels – not sure why the brightness selection would be different. Both could be set up in an action. Rob’s Method is totally incredible – that’s what you get from an Adobe guy I guess – they know how to do this!

I hope this blog has demonstrated what can be done with AI images to make them somewhat your own. I still have a lot of concerns here for creatives when I see Method 3 results. Next time I will create some images using different art media and see how well these methods use them. Someone in a comment suggested trying these types: Watercolor, Acrylic, Pastel, Ink Wash, Spray, Fresco Secco, Gouache, Enamel, Tempera, and Sand to list a few. Will catch ya again soon!…..Digital Lady Syd