My flowpath for AI generated images to plasma cut files - a How-To

[Locust]

The setup details:
I use MidJourney for AI image creation. It's a really powerful tool and I think it's absolutely worth the subscription fee. At the time of writing the pricing is $10/month for a basic plan billed month to month which is good for about 200 images. They offer 20% off if you subscribe yearly. I haven't had any need to go above the basic plan. MidJourney has a few ways you can create images. There's an app called Create, you can log into the website and have essentially the same controls, or you can log in through Discord. I prefer the Discord option so it's what I'm going to show you, but see what you're most comfortable with. In Discord you can either create images that are immediately viewable by everyone else, or you can Direct Message the MidJourney Bot. If you don't want to immediately lose your work in a sea of other people's designs I suggest using the DM option.

I use Inkscape for my vector graphic work. It's free to use, and if you're just getting into vector graphic work, it's indispensable.

I use SheetCAM to lay out my cuts. I started with Fusion 360 but it sucks for artwork. Once the node count climbs up into the 4-digit realm Fusion craps out. It's a one-time cost of $145 for a SheetCAM license.

With that said, I'll walk through an exercise.
People like dogs, so I will create a design for a Basset Hound today.

Screenshot 2025-01-05 101627.png

I'm in my direct messages on Discord with the MidJourney bot. In the message field, I type /imagine to begin the prompt.

In the prompt, I'll describe what I want to see. To start I'm just going to describe the image I want but there is also the option to control things like aspect ratio by appending --ar[whatever aspect ratio you want] at the end of your description. I'm going to skip those details today.

My prompt today is "a black silhouette of a basset hound, no background". I'm looking for a nice high-contrast image to make it easier on myself when I move to Inkscape. "No background" has pretty consistently given me a plain white background. "Black silhouette" seems redundant but it has worked to reduce the occurrence of shading into grays.

Screenshot 2025-01-05 102417.png

After a few moments of processing, I'm given four images, below them are two rows of buttons, U1 thru U4 and V1 thru V4 as well as a blue icon with white arrows. Four images, four numbers. Top left is 1, top right is 2, 3 and 4 are the second row. The arrow button is a re-roll if you're not feeling good about what you've been given to start with but if you're not happy with the results you might want to try describing your prompt a little differently.

The V's are for variation. If one of the four options stands out but it's not quite what you want, click the number corresponding to the image you want to try to refine.

If there's an image you like and want to work with, click the U button for that image to upscale it and break it out of the four images.

I kind of like #4 in the image above, so I'm going to see what variations of that I get with "V4".

Screenshot 2025-01-05 103400.png

Not really what I'm looking for, I think I like the first one better, but there's one more thing I can try. I'm going to upscale #3 because I still think that has potential, and I'm going to be presented with more options.

Screenshot 2025-01-05 103844.png

I still have upscale and variation, but now I can also vary just a portion of an image.

Screenshot 2025-01-05 105032.png

I've selected the region I want to change. I want longer jowls and I want the white mask that I was shown in my first set of images. I think I got the jowls, but I probably needed to word my prompt better for the mask. I'll leave it at that for the sake of this instruction. You get the idea. I'm going to upscale #2 and the right click and copy image so I can move to Inkscape.

Screenshot 2025-01-05 104406.png

After pasting the image in Inkscape I go File>Document Properties, set my units to inches, link the width and heights and set my page height to the dimensions I want my final product to be. 16" works pretty well for me. Big enough that you can get some detail in the cut, still small enough for someone to hang on an interior wall if they want. Exit out of that window and ctrl and drag my image to the size of the page.

Screenshot 2025-01-05 105458.png

The graphic is already a bitmap so all I need to do is trace it. Path>Trace Bitmap. Because I was careful to set up my image with nice hard lines there's really no need to mess with the trace settings but just to make things a little bit cleaner I'm going to move that optimize slider all the way up. This should reduce the number of nodes in the image. I didn't select the image before I took this screen clip, but you'll need to do that before hitting Apply.

Screenshot 2025-01-05 110525.png

You may not think anything changed but you should have two images now, on top of each other. Drag one off. Delete the one that has the white background with it.

Screenshot 2025-01-05 110836.png

Select the Node Tool, click the image. It should look like this. If you didn't move that optimize slider up you might be looking at four times as many nodes and you really don't need them all.

Screenshot 2025-01-05 110941.png

[Locust]

I'll hover my mouse over the nose, hold ctrl, and use the mouse wheel to zoom in. You can see that the nose parts aren't connected and they'll fall out if you try to cut the job like this.

Screenshot 2025-01-05 111153.png

I'll drag the edge of the image pieces, the nodes, or the node legs that are visible once you click on a node and move the images around until I get enough of an overlap that I feel confident that the parts will be sturdy.

I've done that here by dragging the area under the nose until it overlaps with the nose.

Screenshot 2025-01-05 111646.png

You can do it as you go, or once at the end, but those are still separate lines even if they look like one solid area. Path>Union will make them into one part.

After I did that, I deleted two nodes I didn't think I needed and used ctrl and left mouse click on the nodes that were nearest the tightest angles to smooth/open them up.

Screenshot 2025-01-05 111940.png

I'll go around the rest of the image and do the same anywhere parts will be cut out, or where part of the finished work might be so thin that it bends easily I'll thicken areas. Narrow cutouts like around the left eye I know my jet will skip over for being too tight, so I'll move the lines a bit to make that open space a little more open.

When I'm happy with everything File>Export and send it as an SVG. SheetCAM will accept that, or if you want to bring a simple design like this into Fusion you can import it as an SVG there, just be sure to "Fix" your import in your sketch. An "unfixed" SVG imported into a sketch is green and a "fixed" import is blue.

Hope this helps, if anyone has any questions or needs more clarification, let me know.

Oh, I didn't explain the reason for working in Inkscape at the full scale I want my finished part and I forgot one other tip I wanted to share.

First, I work full scale and in the units I plan to work with in other programs so I can import into CAM programs at a 1:1 instead of trying to work out whether I need a 25:1 between units, or try to click and drag to make things match up. Maybe a 16" dog cutout doesn't need to be exactly 16" but if you're including mounting holes at a particular size or spacing, it matters.
Speaking of mounting holes, that's my other tip. In Inkscape grab the circle tool, pick a color that contrasts with your graphic and hold ctrl as you carefully drag your circle out. You can see on the bottom edge of your screen the size of the circle and if you're careful with the direction you drag your mouse the circle will stay.. circular.

Once you've got the size you want (I like something around .06") select your circle, then shift and click the body of your graphic to select both. Path>Difference will cut the circle out of the image. Repeat as needed.

Screenshot 2025-01-05 113116.png

Screenshot 2025-01-05 113208.png

It's pretty easy to set up holes in SheetCAM under a drill operation instead of a jet operation and they seem to come out a lot better than trying to path a tiny little circular cut.

[Locust]

This is what I settled on after connecting the loose bits, opening up the thin cuts, and cutting the dogs nails so he doesn't scratch anyone.

basset.png

[adbuch]

Locust - thanks for your comprehensive explanation. I will need to study it over in detail and possibly ask questions. I do have one question - I was under the impression that for the dog you showed in the other post - you started with an actual photo of your dog and used this AI program to convert it to the image you showed. I guess partly because I saw your image on Reddit with a caption saying something to that effect.

So - my question is - is it possible to use an existing photo and then convert it using the Midjourney AI to achieve a result similar to what you showed?
Here is the Reddit screen shot for reference.

Thanks,
David

dog.jpg

[adbuch]

By the way, I did create a Discord account and sign up for the $10/month subscription to Midjourney yesterday. This is what my home screen looks like.
David

discord.jpg

[Locust]

Locust - thanks for your comprehensive explanation. I will need to study it over in detail and possibly ask questions. I do have one question - I was under the impression that for the dog you showed in the other post - you started with an actual photo of your dog and used this AI program to convert it to the image you showed. I guess partly because I saw your image on Reddit with a caption saying something to that effect.

So - my question is - is it possible to use an existing photo and then convert it using the Midjourney AI to achieve a result similar to what you showed?
Here is the Reddit screen shot for reference.

Thanks,
David

dog.jpg

You can import an image but I haven't made it work as well as the fully generative stuff. I also haven't tried that hard. For the image in my Reddit post I started by asking for a picture of a Pitbull, then asked again, and again, until I got something that didn't look like a Boxer or a Catahoula, then refined it in Inkscape.

Here's an example of a recent attempt.

Screenshot 2025-01-05 173109.png

I'd rather keep going after the variations until I get something pretty close if I can, then do the Inkscape thing to polish it. In the case of the picture of the brown dog, I'm back to doing it the old-fashioned way with a manual trace in Gimp. That is for a commissioned piece and it needs to be exact.

To use an image as part of your prompt you need to start the prompt with /imagine then paste a link to the image you want to use.
You can send the image to the MidJourney bot if you're working from a photo on your desktop. Then, pull the link address from the image (right click, copy link) but when you paste it into your prompt you'll need to delete everything after the ".jpg" or whatever file extension your picture is. It'll start with a question mark and then a bunch of letters and numbers. The question mark and everything after needs to go.

Once the photo link is in the prompt, type your text prompt like usual.

The problem is, AI is still kind of dumb. It's been fed enough 1's and 0's to know what a dog is supposed to look like, but it doesn't know what YOUR dog is, so it's only going to get so close.

[Locust]

By the way, I did create a Discord account and sign up for the $10/month subscription to Midjourney yesterday. This is what my home screen looks like.
David

discord.jpg

Hope you get a lot of use out of it! Accept the TOS and start your /imagine prompt where it says "message @midjourney bot"

[Locust]

Oh buddy.. you got me thinking and digging into the instruction manual for MidJourney.

I remembered that you can basically reverse-engineer an image back into a prompt as well. So I took the pitty pic and had Midjourney turn that back into words. It spits out four prompts instead of four images. Here is prompt #4

"Black and white vector graphic of pitbull dog head on solid background, contour, flat illustration style, simple design with no shadow lines, high contrast, vector art by minimalist artist clean edges, uniform colors, professional t-shirt screen print, clean and smooth, no sharp details, white background"

I ran that text prompt with the original image but I still got something of a color image.

Black_and_white_vector_graphic_of_a_dog_head_on_ca38a8f4-ac78-4322-884f-df788a17664d.png

I upscaled that image and ran "Vary (Strong)" on it and ended up with this.

Black_and_white_vector_graphic_of_a_dog_head_on_f46fb09f-3d2e-40ee-a974-45ab00848a25.png

Original for comparison.

FullSizeR.jpg

Getting close, but I've spent too much time staring at this computer screen today so I'm tapping out. Really goes to show what I said early in my how-to. If you don't like what you're seeing, change what you're asking for.

[adbuch]

Thanks for the update. I have tried PortraitArt for s few things, and I think I read somewhere that they claim they are using Midjourney for their processing. They do have a tool where you can type in some sort of description and details. but I haven't tried it yet.

For certain things, it does a great job. Here is an example.
This one would need editing for cnc plasma cutting, but would work very well for engraving - and looks pretty much like the photo I input to the program.

cotton machine 1.jpg

cotton machine 2.jpg

1.jpg

2.jpg

But for other things like animals, people, etc. the results are not so great - at least for the ones I have done. But maybe I can experiment with giving it some instructions and see how that goes.

David

[Locust]

Yeah, the pet thing is where it gets difficult. Fuzzy (or furry) edges are difficult for a lot of software to detect it seems. I haven't tried Portrait Art, I'll have to take that for a spin.

[adbuch]

Does Gimp have any certain advantages over Inkscape for manual tracing of dogs, cats, etc? I was checking it out and I do see that it has a Bezier draw tool, but no option for B-Spline mode like Inkscape has.

Thanks,
David

gimp.jpg

[Locust]

It's a familiarity thing for me with the caveat that I'm not an expert in either. I've learned what I need to learn to do the things I need to do and then I usually stick with what I know until I'm introduced to something better.

[adbuch]

b-spline 1.jpg

b-spline 2.jpg

b-spline 3.jpg

b-spline 4.jpg

[Locust]

I've been using a Ctrl + mouse click to cycle thru node modes individually or selecting a number of nodes and using the "smooth" (or other modes) node selector just below the p in "help" in the menu list at the top of the screen to that effect.

Sorry, I'm on my phone. I see that's Inkscape now and we're talking about the same program.

[Locust]

Home again now.

So I guess that's one of the advantages of GIMP. I'll just draw. No b-curves. Basic pen tool, follow the edges of things, use the bucket fill to cover areas. I've actually got a graphics tablet on the way to make things a little easier for me. A stylus/pen is a lot easier to control for me than a mouse.

Brown dog project - I did a bunch of the groundwork in GIMP to fill the shadowed areas before I brought it over to Inkscape to flesh out the rest of the design. Again, I'm still learning a lot of this as I go and hand-tracing hasn't been my highest priority and this is a work in progress, and again this is without MidJourney. Only tonight did I make progress in taking an actual image and making it pretty close with MidJourney. I'll keep digging into that but from where I left off, some Gaussian blur in GIMP before importing into Inkscape will go a long way I think.

I've got more work to do on the shadowed side of the face for now, and the upper half of the mouth, but I'll have that dialed in by the the end of the week. Then I'll spend a little time on smoothing out the line widths before I'm ready to cut. Still not bad as-is given my limited experience in hand tracing.

dog.png

[adbuch]

Nice progress! Please keep us posted as you continue with it.
David

[adbuch]

By the way - I only brought up Gimp because you had mentioned it in your post and it is a program I had not used before. I'm always interested in learning about new methods for creating and editing drawings - so I thought I'd check it out. Here is a comparison of using your "Ctrl - click" method for smoothing vs. using the BSpline tool in Inkscape.

David

BSpline vs manually smoothing.jpg