ComfyUI Mohseni Kit
A collection of useful nodes for ComfyUI, including Float Preview for live image visualization.
⚡ ComfyUI Mohseni Kit
A collection of professional custom nodes for ComfyUI, focused on precise sampling control and a smoother preview workflow.
Nodes
| Node | Category | Purpose |
|---|---|---|
| 📈 Mohseni Scheduler | ⚡ Mohseni Kit / Sampling | Karras polynomial sigma scheduler tuned for two-stage sampling |
| 🖼️ Float Preview | ⚡ Mohseni Kit / Image | Floating, always-on-top image preview window |
📈 Mohseni Scheduler
A custom sigma scheduler that outputs a SIGMAS curve using Karras polynomial spacing with a tunable shape. It is built for two-stage sampling — a composition/color pass followed by a detailing pass — letting the same formula bias steps toward either end of the noise range.
Output:
SIGMAS— connect it toSamplerCustom/SamplerCustomAdvanced(with aKSamplerSelect), not the standardKSampler.
Parameters
| Parameter | Description | Typical |
|---|---|---|
| scheduler | Spacing formula. Currently karras_polynomial (extensible). | — |
| steps | Sampling steps for this stage. | 15–30 |
| denoise | 1.0 starts from pure noise; lower values start partway down the curve. | detail: 0.35–0.5 |
| rho | Curve shape. < 1 = composition (steps held high), 1 = linear, > 1 = detail (steps packed low). | compose 0.4–0.8 / detail 7–15 |
| shift | Adds high-noise step density on top of rho (resolution-aware). 1.0 = off. | compose 2–4 / detail 1.0 |
| sigma_max | Top of the noise range. 0 = auto from model. | 0 |
| sigma_min | Bottom of the noise range. 0 = auto from model. | 0 |
Why polynomial spacing
Composition and color form at high noise; fine detail forms at low noise. rho decides where the steps are spent:
- Composition —
rhobelow1(e.g.0.6) holds steps in the high-noise band; addshift2–4for more resolution there. - Detail —
rhoof7–15concentrates steps in the low-noise band, which adds detail far more cleanly than exponential spacing.
Two-stage detailing recipe
| | Pass 1 — composition | Pass 2 — detail |
|---|---|---|
| steps | 24 | 18 |
| denoise | 1.0 | 0.45 |
| rho | 0.6 | 10.0 |
| shift | 3.0 | 1.0 |
| sigma_max / sigma_min | 0 / 0 | 0 / 0 |
Per pass, wire Mohseni Scheduler → SamplerCustom (sigmas) and KSamplerSelect → SamplerCustom (sampler). Pass 1 runs on an empty latent (add_noise on); feed its LATENT into Pass 2 (add_noise on), then VAE Decode.
Tip: for a detail pass without splitting, use
denoise 1.0withsigma_max ≈ 2.5andsigma_min 0to keep the whole schedule in the low-noise band.
KSampler dropdown presets
Two fixed presets are added to the standard KSampler scheduler dropdown automatically:
mohseni_compose— composition-weighted (rho 0.5,shift 3)mohseni_detail— detail-weighted (rho 7)
They are fixed-value (no live controls) — use the node to tune, the presets for speed. On an incompatible ComfyUI build they are skipped without affecting the rest of the kit.
🖼️ Float Preview
A floating, resizable, always-on-top image preview window that lives outside the ComfyUI interface — ideal for multi-monitor setups.
Features
- Live preview that updates as the workflow runs
- Batch support — scroll through multiple images
- Save to disk or copy to clipboard
- Always-on-top toggle and free window dragging
- Persistent window size, position, and state across sessions
- Keyboard shortcuts and a right-click menu
Controls
| Action | Description | |---|---| | ← / ↑ | Previous image | | → / ↓ | Next image | | Ctrl + T | Toggle always on top | | Ctrl + C | Copy current image to clipboard | | Ctrl + S | Save current image (PNG / JPEG) | | Esc | Close the window | | Right click | Open the context menu |
Connect any IMAGE output to the node and run the workflow; the window appears and tracks new outputs.
📥 Installation
ComfyUI Manager
Search for ComfyUI Mohseni Kit in ComfyUI Manager, click Install, then restart ComfyUI.
Manual
cd ComfyUI/custom_nodes
git clone https://github.com/mohsenicreative/ComfyUI-Mohseni-Kit
Restart ComfyUI. Dependencies are installed automatically; if needed, install them manually:
pip install PyQt6 ftfy psutil
📜 License
Licensed under Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0).
- Credit the original author: Mohammadreza Mohseni.
- Modifications must be shared under the same license.
- Commercial use is allowed with attribution.
Full text: CC BY-SA 4.0