ComfyUI Extension: TBG’s ComfyUI Development Takeaways

Authored by Ltamann

Created 4 months ago

Updated 8 days ago

10 stars

A curated collection of reusable ComfyUI nodes developed by TGB. These sidecodes encapsulate key breakthroughs in model sampling, noise scheduling, and image refinement for enhanced stable diffusion workflows.

Custom Nodes (0)

README

ComfyUI-TBG-Takeaways

TGB’s ComfyUI Development Takeaways

A curated set of reusable ComfyUI nodes created by TGB, these sidecodes capture major advances in model sampling, noise scheduling, and image refinement to enhance stable diffusion workflows. All nodes in this pack are extracted from micro-developments created during the development of TBG ETUR. They’ve been turned into standalone “micro nodes” for testing purposes and for anyone who enjoys experimenting with them.

Visit TBG enhanced tiled upscaler and refiner pro

Included Nodes
Overview
License
Support

Included Nodes

VAE Decode ColorFix (New!)
VAE Decode (ColorFix) - Fast, Color-Accurate Decoding for Flux Models Solves the brightness shift and washed-out color problem that affects Flux models when using the standard VAE Decode node. WHY THIS NODE? Standard VAE Decode produces washed-out, brightened images with Flux models because it processes the entire image at once, causing normalization drift in the VAE decoder. This node uses tiled processing to maintain accurate colors while offering speed optimizations not available in the standard VAE Decode (Tiled) node. HOW IT WORKS The node breaks your image into smaller tiles during decode. Smaller tiles keep the VAE's normalization statistics closer to training values, preventing color shift. By default, it uses a single-pass approach (3x faster than standard tiled decode) while maintaining color accuracy.
PromptBatchGenerator (New!)
Ever get super frustrated with those 81-frame batch videos where the same prompt just keeps repeating over and over? 😫 I know I did! Like, every 81 frames I’d have to smile or move my hands exactly the same way because the prompts were crazy repetitive. No no, that’s not fun for me. But here’s a cool trick: did you know you can enter as many prompts as you like, separated by |, and the node will automatically calculate the time per prompt? That means you can actually tell a story! 🎉 Still, I wanted more… so I asked my team to build a nicer random prompt generator. Now you can: Input multiple prompts with different behavior strengths Mix them all across your video Let the node decide how the behaviors combine, so it feels natural and dynamic
TBG_FluxKontextStabilizer Developed specifically for the TBG ETUR (Enhanced Tiled Upscaler and Refiner), this node maintains exact positioning of reference images in final outputs. It stabilizes spatial context during tiled upscaling and refinement to ensure high-fidelity alignment and image coherence. Stay with euler beta and between 16 and 30 steps. Add to Promt: Repair and enhance this this this photo.
ModelSamplingFluxGradual
Implements gradual flux-based sampling control for smoother transitions during model sampling - ModelSamplingFluxGradual interpolates between ModelSamplingFlux and ModelSamplingFlux Normalized. This allows for the best of both approaches. Detailed Inforamtion here: https://www.patreon.com/posts/125571636/edit
PolyExponentialSigmaAdder Highres Fix Flux
The PolyExponential Sigma Adder node adds the ability to manipulate curve parameters, such as adjusting the curve’s rigidity, and allows for the application of a negative poly-exponential curve to the Sigmas. The PolyExponential Sigma Adder introduces a resolution-independent curve, ensuring a consistent adjustment to img2img processing across different resolutions. Sustitución for resolutions depending ModelSamplingFlux
BasicSchedulerNormalized
A scheduler node with built-in denoise normalization to ensure stable consistent sampling results across schedulers.
LogSigmaSamplerNode & LogSigmaStepSamplerNode These nodes offer direct access to the internal model noise curve—the curve the model expects from sigma values during diffusion. LogSigmaSamplerNode enables manipulation of this curve directly, allowing users to enhance fine details, introduce natural imperfections, or soften the final image by shifting how the model interprets noise over time. This approach gives precise control over the model's behavior—similar in effect to techniques used by Detail Deamon or Lying Sigmas. These nodes are ideal for users looking to experiment with or customize the core diffusion response for artistic or technical purposes.
TBG_Preview_Sender_WebSocket This node lets you send images directly from memory to memory, right through ComfyUI’s websocket no need to save anything to disk. Just plug it in as an output for your images, and you can instantly fetch the previews from another web interface or app connected to your ComfyUI websocket. TBG_Preview_Sender_WebSocket
TBG Hex Cone Mask Generates a hexagonal cone pattern mask for variable denoise strength across an image. Creates gradients from center (maximum denoise) to edges (minimum denoise) of each hexagon. Control the gradient zones with inner_fraction (start of gradient), outer_fraction (end of gradient), and curve_power (gradient steepness). Outputs both a visual IMAGE of the pattern and a MASK for denoise control. Perfect for creating organic, non-uniform enhancement patterns. TBG Hex Cone Mask

Detailed Inforamtion on my Patreon:

Overview

TGB’s sidecodes provide practical solutions distilled from complex development work into simple, easy-to-integrate nodes. These tools offer enhanced control over sampling dynamics, noise management, and image refinement, making them valuable assets for artists, developers, and researchers using ComfyUI and stable diffusion workflows.

Explore this repository to discover how these nodes can streamline your image generation pipeline and help push the boundaries of creative and technical possibilities.

License

Copyright (C) 2025 Tobias Laarmann This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License.

This program comes with ABSOLUTELY NO WARRANTY; 
This is free software, and you are welcome to redistribute it
under certain conditions;

Support

For updates and detailed development insights, visit TGB’s Patreon page).