Using Vision FX 1.0 AI Plugin with Paintshop Pro 2023

[Morticia]

I just installed the Vision FX 1.0 plugin for Paintshop Pro 2023 Ultimate. The good news: It works! I've already created some wonderful images and I'm really enjoying the product.

My question is this. I've looked for explanations of the sliders the Vision FX plugin uses, but there doesn't seem to be a manual or a help page available. There is a little 'information i' beside the slider names, but clicking it does nothing. Now I don't mind guessing -- it can be fun -- but it would be nice to have some idea of what the sliders (text guidance scale, strength, starting seed) do.

I bought the Vision FX plugin yesterday via the Corel store. Installation was easy.

I've checked the Vision FX website, but there's no mention of a user forum or other resources.

Any advice would be welcome. Thank you.

[LeviFiction]

It's not clear if DistinctAI (HappyTreeSolutions) is a Corel/Alludo company or just a third-party that's partnered with Corel/Alludo to provide the plugin in their store as it's advertised to work with Photoshop and Affinity as well. So....you can try emailing their support team.

[Morticia]

Thanks, I'll do that. Was also hoping that perhaps another wiser PaintShop Pro user of Vision FX was around so we could share questions and experiences.

[LeviFiction]

This plugin is based on stable diffusion, so this article on the Stable Diffusion parameters might help to explain things pretty well - https://blog.openart.ai/2023/02/13/the- ... arameters/

Are there more sliders than are visible in the screenshots on their website? I see Strength, Text Guidance Scale, Seed, and steps. If that's all, then using the article posted above I think these would be decent descriptors of those sliders.

Strength: As I don't have the plugin I don't know if this is the img2img strength (higher means more like the original and lower means use the original as a rough template for colors and appearance). Or if it's like pic2painting was and is a transparency/blending between the result and the original. If I had to guess I'd assume it's the img2img strength.

Text Guidance: most AI has something like this. AI doesn't actually understand human language so it puts weights and makes connections to giant datatbases of image features. This tells the AI how closely to match the prompt you gave it. And how many additional connected weighted words and features it can throw in or play with the prompt you gave it.

Steps: So the original stable diffusion AI works by creating a noisy image and then over several steps, slowly denoises the image to create a result. So my assumption is the steps are how clear, good, and detailed the end result will be. The higher the steps, the better the end result (in general) but the longer it takes as each new step adds detail and it takes time to calculate. 25 is considered a good starting point.

Seed is not a slider, but all of this works on a starting point and somewhat randomly. It's the initial random noise that the image starts out with. If you use the same prompt with the same seed, you should get the same result every time because the seed is no longer random. But if you let a random seed be generated with each image, you'll get a new result every time. Most people leave this alone unless they specifically want to achieve the same image again.

[Morticia]

Oh, thank you so much! I shall follow the link you provided and study that.

And the sliders make so much more sense now, after your explanations. Again, thank you.

Just now I generated an alien face that I think is truly amazing. Sure, I didn't intend to wind up with an alien face, but wow! I included the word 'photorealistic' in the prompt and what a difference.

Really appreciate the help.

[Jean-Luc]

Just now I generated an alien face that I think is truly amazing. Sure, I didn't intend to wind up with an alien face, but wow! I included the word 'photorealistic' in the prompt and what a difference.

Could you post the result here? (starting image and result)
It may help PSP users to discover that plugin.

[Morticia]

Going to post an image I just made using Paintshop Pro 2023 and the Vision FX V1.0 plugin. I hope that's okay.

Some background: The source image is a digital photo taken inside Second Life. I used my own props, which were a pair of fire wall displays from Hidden and a Halloween tree. I just plopped them down and snapped a quick photo.

The source image is on the left. It's nothing special; I just wanted something creepy.

I used the Vision FX plugin on that image, with the following settings:

Text Prompt: "Lovecraftian horror. Fires, tentacles, desolation, highly detailed."
Strength: 63.
Text Guidance: 18.
Steps: 30.
Seed Number: 2085682463

And here's one of the four images Vision FX returned, on the right:

AI Art Rendering Panel for Flickr.png

This could be an excellent background, for instance, to a Halloween event poster. Or any number of macabre subjects.

And it was fun to make!

[Kathy_9]

A fun image. Thanks for sharing.

[Jean-Luc]

Thank you for this demonstration.
What is the maximum possible size for the output image?
Is it possible to change the resolution and/or the sharpness? The resulting image is a bit blurry.

[Morticia]

Below are both the source and the manipulated images, each presented in larger format.

To answer Jean-Luc's question, the Vision AI image is returned the same size as the source image. There are no controls on the Vision user interface to change this. Of course you can resize the resulting image in PaintShop.

If there is a maximum source image size I haven't found it yet. Most of my inworld photography is done at 4000 by 4000, and I do a lot of cropping, etc. My promo posters are always shrunk down to 1024 wide by 2048 tall (2048 is the maximum height or width that can be uploaded into Second Life).

I have noticed that larger source image sizes result in longer processing times for Vision FX. Each of the four sample images it renders usually takes between 20 seconds and a minute on this laptop, with larger images requiring more time.

Anyway, on to the pics!

SOURCE IMAGE:

Lovecraft Original Image.png

AI IMAGE:

Lovecraft Image.png

Hope this helps!

[Morticia]

Here's another example of how I use PaintShop Pro 2023 and Vision FX.

I took a picture of Gothic cathedral I won in Second Life. I used that image inside PaintShop and Vision FX to generate a nice Gothic background. I wanted something with a softer focus, more evocative than crisply detailed.

INWORLD PHOTO:

Original Image For Post.png

AI IMAGE:

AI Image - For Post.png

Now all I needed was me.

I quickly learned that including a human image in the source photo would, at my current skill level, lead to hilarity. Arms everywhere, fire hydrants for noses -- amusing but not quite what I'm looking for. So instead, I took a quick snapshot of my Second Life avatar dressed in a Gothic dress and removed the background using PaintShop Pro.

Finally, I dropped the spooky Mort as a new layer into the AI generated image, and here is the result:

FINAL IMAGE:

Gothic Mort - For Post.png

Now normally I'd mess around with shadows and smooth things out and add text for an event poster.

I'm really pleased with the final product, even as is. The AI saved me a LOT of time; making the source photo look like I wanted would have taken forever. I made this whole thing in about 20 minutes, including the inworld photoshoot.

Just for fun, yesterday I took an inworld photo of myself perched on a Halloween jack-o-lantern. I ran that through Vision FX and asked for an image in the style of an old Weird Tales pulp magazine cover. I didn't yet realize that my image would be used like any other pixel on the screen, because this is the image I got. I love the image -- it looks like Weird Tales magazine art, even if my favorite person isn't shown. This is why I'll probably use Vision FX for backgrounds until I learn more about human images.

NOT MY BEST SIDE:

mortpumpkinFor Post.png

[Jean-Luc]

Your examples (very nice ones, by the way) confirm what I suspected. This plugin produces images that lack sharpness. But this slightly blurred rendering is perfect for your intended use (background).
I hope you enjoy using this plugin.
Thank you for sharing.
)

[Morticia]

You're welcome!

I'll post more as I go.

[Morticia]

Here's a quick one I made today.

It's a composite image. The spacecraft came from a public-domain photo of a single flower against a black background. I asked the AI to make that a photorealistic starship.

The space image, the background, came from a NASA Hubble image of the Fountains of Creation nebula. I just asked for a photorealistic rendering of a deep space nebula with colorful lighting.

Then I erased the background from the ship and overlaid that onto the new nebula image.

Here is the result -- I call it Sailing the Vast.

Sailing the Vast.png

[LeviFiction]

I wonder if the quality would increase if you ran it through PSP's upscaling AI in the Resize dialog.