ComfyUI-BSS_FLSampler
High-performance foveated sampler (FLSampler) for ComfyUI inspired by human vision; selective latent modification delivers cinematic clarity and micro-textures without generation overhead.
Nodes (1)
FLSampler (Foveated Latent Sampler) (BSS)
FLSampler (BSS) is a high-performance custom sampler for ComfyUI, inspired by the physiology of human sight (Foveated Vision). It is engineered to deliver stunning cinematic clarity, rich micro-textures, and precise depth-of-detail control directly within the latent space, without adding any overhead to generation times.
[!IMPORTANT] Author and Developer: blacksnowskill (BSS)
© 2026 blacksnowskill (BSS). All rights reserved.
This project is protected by copyright. Any unauthorized copying, modification without attribution, or representing this code as your own product is strictly prohibited.
[!TIP] Ultimate Speed & Performance Booster:
To achieve the ultimate balance of breathtaking visual clarity and lightning-fast generations, we highly recommend pairing FLSampler (BSS) with our core optimization suite ANIMA_BOOSTER. It delivers a massive 3.5× to 5.0× speedup on Anima DiT models, while FLSampler perfectly preserves and enhances every tiny high-frequency detail!
🔬 Physics & Mathematics of the Process
Standard samplers (KSampler) apply noise injection and denoising uniformly across the entire latent space, which often results in either blurred micro-textures (plastic look) or over-sharpened artifacts (burnt images).
FLSampler resolves this bottleneck through intelligent, selective latent modification:
- Differential Latent Analysis (Focus Detection): At each denoising step, the sampler computes the absolute difference $\Delta = |x_{0, t} - x_{0, t-1}|$ between the current predicted clean image $x_0$ and its state from the previous step. This mathematically pinpoints the exact regions where geometry, object boundaries, and fine structures are currently forming.
- Adaptive Fovea Map with Inertia (Momentum Mask):
The raw difference map is smoothed using a 2D average pooling operator (
avg_pool2d), after which a dynamic threshold is calculated based on its mean and standard deviation: $$Threshold = \mu + 0.5 \cdot \sigma$$ A soft, organic focus mask (Fovea Mask) is then generated using a sigmoid activation. To prevent erratic focal jumps between steps, a momentum coefficient (Momentum /mask_inertia) is applied, ensuring a smooth, natural transition of detail zones. - Local Contrast (Sharpness Boost): Inside the high-activity Fovea zone, an Unsharp Masking operation is performed directly on the latent tensor. High frequencies are extracted by subtracting a blurred version of the latent from the original, and then selectively amplified. This produces incredibly crisp object boundaries and realistic cinematic depth of field.
- Stochastic Texture Injection (Micro-Grain): During the early-to-mid stages of generation, a controlled micro-noise tensor is injected solely into the active Fovea zones. The subsequent denoising steps naturally resolve this noise into ultra-fine high-frequency textures (skin pores, clothing fabrics, wood grains, foliage, hair, or fur), preventing flat plastic surfaces. The injection smoothly decays towards the final steps, guaranteeing a clean, high-fidelity output.
🎨 BSS Premium UI
The node comes integrated with the luxurious BSS Premium UI design system, fully synchronized with the ANIMA_BOOSTER aesthetic:
- Luxury Matte Theme: A deep, matte ultra-dark graphite body (
#0f0f0eff) that seamlessly blends inputs, outputs, and the node itself. - Rich Gold Accents: Harmonious gold sliders (
#d4af37), gold knobs (#e5c158), gold dropdown arrowheads, and gold-trimmed numeric input borders. - Complete Silence: Absolute suppression of annoying hover tooltips during widget adjustment and node dragging.
- Branding: A fine gold divider bar situated above the first widget and a sharp, metallic
"BSS OPTIMIZED"branding engraving in the bottom-right corner.
🎛️ Parameters & Impact Guide
This node completely replaces the standard KSampler, introducing both classic controls and proprietary physical regulators:
| Parameter | Range | Recommended | Description |
| :--- | :--- | :--- | :--- |
| fovea_strength | 0.0 – 10.0 | 3.0 – 8.0 | Texture Strength: Controls high-frequency detail, skin porosity, and fine textures. Set to 0.0 to disable noise injection entirely. Default: 3.0. |
| sharpness | 0.0 – 3.0 | 0.3 – 1.0 | Local Contrast: Dictates edge definition and object boundary clarity in focus zones. Delivers a "tack-sharp" look without halo artifacts or "frying" the image. Default: 0.5. |
| mask_inertia | 0.0 – 0.99 | 0.80 – 0.90 | Mask Inertia: Focus movement smoothness between steps. Values >0.85 stabilize detail zones, while lower values force the focus to react instantly to sudden changes. Default: 0.85 (step 0.01). |
| denoise | 0.0 – 1.0 | 1.0 | Standard denoising strength. |
| steps / cfg | - | - | Standard generation steps and prompt guidance scale. |
📊 Node Outputs
The node provides two outputs:
- 🖼️ latent: The optimized latent tensor ready for VAE decoding.
- 👁️ fovea_mask: The accumulated Fovea focus map visualized as an RGB image. You can pipe this directly into a Preview Image or Save Image node to visually inspect exactly where the sampler focused its computation during the run—an unprecedented debugging tool for complex workflows!
📋 Recommended Workflow
[ MODEL (e.g., Anima / SDXL) ] ──> [ positive ]
──> [ negative ]
↓
[ Latent Image (resolution) ] ─────> [ latent_image ] ──> [ FLSampler (BSS) ] ──(latent)──> [ VAE Decode ] ──> [ Save Image ]
└──(fovea_mask)──> [ Preview Image ]
📥 Installation
- Open your terminal in the
ComfyUI/custom_nodes/directory. - Clone the repository:
git clone https://github.com/BlackSnowSkill/ComfyUI-BSS_FLSampler.git - Restart ComfyUI and refresh your browser cache (
Ctrl + F5) to enjoy the gorgeous BSS Premium UI.
☕ Support & Development
If you love my work and want to support the development of future optimizations, nodes, and custom models, please consider supporting me:
- Boosty: Support & Exclusive Models
📄 License & Usage
© 2026 blacksnowskill (BSS). All rights reserved.
This software is an experimental release. Feedback is highly welcome. Notice: This project is protected by copyright. Any unauthorized copying, distribution, merging with other projects, or hosting on other repositories/websites without the explicit written permission of the author is strictly prohibited.