Quillworks Simple Shade V3
Usage Guide
Best results are found between 3-6 cfg, higher cfgs have more intense colors (often too much) and lower numbers are more likely to be coherent at the expense of color. I recommend running between 4 and 5.
The best Scheduler and Sampler are Dpm++ 3m with Karras but it does work in things like euler, but you lose a bit of the polish on edges and the images don't pop as well. It also works well in DPM++ 2m with Karras.
I run it at 30 steps but the model works well between 20-30 steps.
Prompting on this model is typical illustrious booru mixed with a bit of my own personal style that uses things like "Warm Color Palette" and "Cool Color Palette". My dataset was tagged using Qwen running locally and the training was run locally on my 4070 on illustrious 2.0 then merged into my last version. I will also be releasing the lora used to create this on the side for anyone that wants to experiment with it.
Negative Prompting "Bad Quality, Low Quality, colored sclera, " is my go to prompt but I'm not even sure it is needed.
This checkpoint explores a painterly 2.5D aesthetic that bridges the gap between stylized animation and cinematic realism. Each image aims to capture the tactile richness of a hand-painted illustration while preserving the dynamism of modern digital art. The style emphasizes vivid color palettes, expressive lighting, and soft depth cues that draw the viewer’s eye through the scene as if peering into a moving frame.
At the core of the design philosophy is the illusion of dimensionality. Subtle tilt-shift effects are used to blur foreground and background elements just enough to create a false but convincing sense of depth, making the midground subjects appear strikingly alive. The result feels like a digital painting that has stepped into motion, flat brushwork and sculpted form coexisting within the same visual space.
The aesthetic thrives on confident composition and energy rather than perfection. Brush-like edges, painterly gradients, and intentional texture variance give each image a handcrafted quality that invites closer inspection. Where visible brush strokes and stylized light bloom become part of the storytelling rather than details to be polished away.
Every image in this checkpoint is presented in its raw, unedited form. No hand or face corrections were made, allowing the underlying generation process to remain visible and authentic. This approach highlights the model’s capacity to produce striking, artful results without heavy post-production, offering a direct look into the creative potential of painterly 2.5D rendering.