FLSampler (BSS) for ComfyUI 🔬
A high-performance custom sampler inspired by the physiology of human sight (**Foveated Vision**). Engineered to deliver stunning cinematic clarity, rich micro-textures (skin, hair, fabrics), and precise depth-of-detail control inside the latent space—with zero performance overhead.
> ⚡ Ultimate Speed & Quality Combo:
> To achieve the ultimate balance of breathtaking visual clarity and extreme speed, we highly recommend pairing this node with our optimization suite ANIMA_BOOSTER
🔗 GitHub Repository (Full Guide & Physics): https://github.com/BlackSnowSkill/ComfyUI-BSS_FLSampler
🔬 How it Works
Instead of uniform denoising, FLSampler selectively enhances latent focus zones:
Differential Latent Analysis: Dynamically tracks where structures and object boundaries are forming.
Local Contrast (Sharpness Boost): Unsharp-masks active focus zones directly in latent space for ultra-sharp details (safe default 0.5, range 0.0–3.0).
Stochastic Texture Injection: Adds controlled micro-noise to focus areas, naturally resolving into skin pores, clothing fabrics, wood grains, or hair, preventing flat plastic surfaces.
BSS Premium UI: Integrated dark-matte & gold custom theme.
📂 Quick Installation
1. Open ComfyUI Manager -> Install via Git URL.
2. Paste this URL: https://github.com/BlackSnowSkill/ComfyUI-BSS_FLSampler
3. Click Install and restart ComfyUI.
For detailed parameters, mathematics, and recommended workflows, please check the GitHub README
☕ Support & License
Support Development: Support me and get exclusive models on Boosty!
License & Usage: © 2026 blacksnowskill (BSS). All rights reserved. This project is protected by copyright. Copying, distribution, merging, or use on other websites/repositories without explicit written permission is prohibited.
Description
FAQ
Comments (11)
Ok, but why released after 1 month?
because at first it was exclusive to my boosty subscribers.
@llikswonskcalb I didn't get it then
Hi, I like this approach.
But I got this error:
Given groups=1, weight of size [320, 4, 3, 3], expected input[2, 16, 94, 167] to have 4 channels, but got 16 channels instead
Channel must be 4, not 16
Which model are you using? Most likely, this error occurs because a model other than SDXL architecture is used.
@qek what channel are you talking about?
@llikswonskcalb I used BSS Realluxura IL
Your idea and implementation significantly improve the generation result. Thanks for sharing!
Here's what I noticed: in your Gothic render below, the area behind the nose was initially flat, but you still added contrast with the Sharpness: 2.0 parameter.
When I exactly replicated this example, the standard KSampler produced a gradient in this area, and the cross turned out textured (not flat).
In my case, a better result was achieved with the following FLS parameters:
Fovea Strength: 42 (Texture Injection)
Sharpness: 0.3 (Contrast Boost)
Mask Inertia: 0.78 (Mask Stability)
非常好的K采样器，显著提升了图像生成结果
Previews look Amazing, unfortunately it doesn't seem to work with AMD GPU's