Testing Models: Flonix's Dan Mumford Style V2

Today, I am going to test this new Model called Flonix's Dan Mumford Style V2 today.

The author of this model, floniixcorn, says that the model is trained with a dataset of Dan Mumford Style images and uses highres fix. Non square image sizes should produce the best results. Another info given on Civitai is that this version is Fine Tuned, has added safetensors and that we should use "strong colors" and "vibrant colors" in the prompt if we want more colors. 

So, first things first... who is Dan Mumford?

Dan Mumford is a freelance illustrator working in central London, UK, running Studio Dan Mumford Limited. Over the past 15 years, Dan has worked within the pop culture and music scene creating everything from album covers, branding and screenprint's to new interpretations of classic film posters and albums.Clients include Disney, Hasbro, Apple, Adobe, Sony, Wizards of the Coast, Icon Motoports, CBS, Iron Maiden, Metallica, and many many bands and record labels from around the world. 

You can find out more about him and his amazing work on his official site.

After studying his portfolio and some of his work, I had a pretty good idea what his style looks like. Now, my expectations for this the model were high.

As I was waiting for the download of the 4GB file, I started to think about my concepts. I wanted to try the model with a few of my favorite sample methods. So I was going with 2 images for each sample method and using always the same prompt on each of the two images.

For the first image I wanted to try an illustration of a woman walking her dog, using a red bandana covering her mouth. For the second one I am going for an illustration of a ninja hiding behind a tree at night.

The Test

The first generated image will have a random seed, that same seed will be then used with the other sample methods for that picture. All the configuration stays the same for each image also. In this case for the first image I used:

Steps: 27, CFG scale: 11, Size: 512x720, Model hash: 57f62f93d0, 
HirResFix:ON, Hires Steps: 12, Denoising strength: 0.7

For the first image the prompt is:  

By danmumford extremely epic illustration ((latin woman walking her dog)) ((front view)) ((using a (red) bandana covering mouth)), ((close-up))empty street, strong colors, Beautiful, psychedelic, Cinematic atmosphere,

I tested this first image with DPM++ SDE, DPM++ 2S a Karras, Heun and Euler a. Let's see the results:

Sampler DPM++ SDE:

Sampler DPM++ 2S a Karras:

Sampler Heun:

Sampler Euler a:

Conclusion for image 1:

I got a few nice outputs with this test, but looking closely we see that I never got a red bandana, having a Covid Mask instead . With DPM++ SDE we got a very strange face and a weird artefact looking like a big flag.  Heun got us a very weird looking dog with a tent, a strange hand and an overall grainy look.

DPM++ 2S a Karras and Euler a produced both very nice images. Both could definitely be used to start a journey of remixing and fine tuning until perfection. I would go with Euler's version, the dog just looks bad ass giving the image that extra punch.

Second Image

Let's see the second image, we want an illustration of a ninja hiding behind a tree at night and the configuration used was:

Steps: 27, CFG scale: 11, Size: 512x720, Model hash: 57f62f93d0, 
HirResFix:ON, Hires Steps: 12, Denoising strength: 0.7

For this image the prompt is:  

By danmumford extremely epic illustration ((ninja hiding behind tree)) ((front view)) ((ninjitsu)), ((close-up)), night ((moonlight)), Beautiful, Cinematic atmosphere, 4k uhd image

I tested this first prompt with DPM++ SDE, DPM++ 2S a Karras, Heun and Euler. Here are the results:

Sampler Heun:

Sampler DPM++ 2S a Karras:

Sampler Euler a:

Sampler DPM++ SDE:

Conclusion for image 2:

Like in the first test, I got some quite good images in my output folder. Heun produced a very camouflaged Ninja, I almost didn't see him. DPM++ 2S a Karras created a very nice image but it feels a bit to dark and the bit of the face that should be visible could look better.  Euler a again with an overall good looking output, but with some strange artifacts in the tree and deformed weapons. Have to say that DPM++ SDE produced a very surprising image. The ninja is not only hidden, but he also climbed the tree; amazing position, amazing picture. A few details that have to be adressed like the eyes and the right arm, but in conclusion, for a first output it lokos very impressive.

My expectations were high about this Flonix's Dan Mumford Style V2 model, but i must say that it is very capable of producing high quality illustration art. It's definitely going on my model shortlist, and I just can recommend anybody to try it out. A big thanks from me to floniixcorn for making this amazing model.

Download  Flonix's Dan Mumford Style V2 and other great models on Civitai.