phosphorescence
"luminescence that is caused by the absorption of radiations (such as light or electrons) and continues for a noticeable time after these radiations have stopped"
This model is the result of a few weeks of experimentation and research. I wanted to make a model capable of creating striking visuals with great detail and contrast. I believe this is the best iteration so far of my experiments. Let's get into the breakdown!
Base Models used:
Yiff In Hell Dune Black by Latte Leopard (Part A) model link
Azure Starfire Void V6 by BastionExa (Part B) model link
These models were selected as I love the anatomy, glossy texture, and strong, expressive posing of Yiff in Hell, as well as the intense visual effects and style of Azure Starfire. While extremely capable and wonderful models to use on their own, I wanted to try to bring the best of both of them in one package. That leads me to my merge recipe.
Merge Recipe and Technique:
I merged this models in ComfyUI using a 60/40 ratio on the UNet, followed by a 50/50 ratio on the CLiP to preserve knowledge and features from both models. I baked in a custom VAE by Arctenox called Crystal. I really like the way it works, it adds and extra bit of personality to the images I think.
VAE link here
The method I used for the UNet portion is experimental, being a noise injection merge. This is possible by a node pack on ComfyUI called Tensor Prism, which is also made by Arctenox. I haven't seen anyone really test these nodes out so I figured I would try it out, and IT WORKED!
Quick breakdown of how noise injection merging works. From what I have seen it works by interpolating tensors like a normal merge, BUT it also injects a controlled dose of noise to offset the tensors. This has been used before in the LLM space, but only in small exercises. In theory the noise offset will reduce averaging and muddiness due to having too many similar tensors. This offset can also increase sharpening and add depth to the tensor space by increasing variation, which would improve the model's output.
Long story short, this model is a blend of the original models with a bit of controlled chaos added. And I baked a couple LoRAs to further enhance the experience.
LoRA 1: PTD Style: Moody Dark & Glow by Desync link
(this increases the contrast and the vivid colors.)
LoRA 2: Adjust details AND photorealism V9 ULTRA by maxxxx_civitai link
(this one boosts the scenery and details, it also makes the poses even more dynamic and stuff. this one is STRONG. and it still does more digital and toony stuff btw!)
LoRA 3: Furry-Artistic Refiner by daspin335 link
(this one increases the detail in the fur and helps with furry anatomy and stuff, and adds texture.)
LoRA 4: Azure Void Ink Dark by BastionExa link
(this one amplifies the contrast enhancement and adds an inky texture to the style, this one has a lower influence as it's pretty strong and I still want the model to be customizable style-wise.)
I hope you guys enjoy this model. It was really fun making this even though it took a while due to limited GPU power. I wanted to do something to show my gratitude for how welcoming and helpful you all have been.







