Quicksand Fans

I've always strived for photorealism in my AI works, and I think I'm finally starting to get somewhere with quicksand stuff here are a few samples of what I've done recently; any feedback is appreciated!

Wonderful work. Mind if I ask what tool you used to create these?

These are very good. I particularly like the pose and expression in the third one, the look of glee and anticipation fits the scene perfectly. Looking forward to seeing more.

Duncan Edwards wrote:Wonderful work. Mind if I ask what tool you used to create these?

From the names of the second and third picture, I'd guess Masteriti is using ComfyUI/Stable Diffusion. ComfyUI is a GUI for Stable Diffusion (I'm using the Automatic1111 GUI to control SD, basically because that's the one I came across first). I'd be interested to know which Checkpoint model is being used, this one is producing some nice realistic subjects.

Duncan Edwards wrote:Wonderful work. Mind if I ask what tool you used to create these?

Thank you! And of course I don't mind, I think these tools should be accessible to everyone. I use a tool named ComfyUI to construct my workflow for AI image creation. Here's what it looks like:



But that's only the workflow; there's a lot going on inside of it, and I'd be happy to answer any more questions about what it involves. But for starters, if you're wondering what model I'm using, it's called "EpicPhotoGasm" by Epinikion. It's the best model for photorealism out there right now, and with the right prompting, it can make some nice mud/quicksand as well. I'm also using a LoRA called "EpicRealismHelper" by the same person to add finer details, and I often have a ControlNet setup to control poses and the general composition of the image. In the above example, I'm using an image of Pippa sinking in a mud flat from her QS site (http://www.users.globalnet.co.uk/~pippa/, all credit goes to Pippa's Productions, of course) as the input for my ControlNet.

Ah, EpicPhotoGasm, that's one I have as well. I've been mostly using AbsoluteReality, but switch when I want a change of appearance or am not happy with the results I'm getting. If I'm using the "Refiner" I'll use a different model for that. I haven't tried Loras yet, you don't want too much new stuff at once or you end up not knowing which effect comes from where, and there is definitely a lot to experiment with in these AI tools. I shall watch your work with ComfyUI with interest, it is a very different approach to Automatic1111, one that I may wish to move to in time.