ComfyUI-easygoing-nodes
Enhanced Text Encoder modules, acd Custom nodes for ComfyUI, device-select CLIP loaders, providing HDR effects, image saving with prompt metadata.
Nodes (31)
Save a Qwen text encoder without the key-prefix mess
Scale each of SDXL's 44 text-encoder layers
Scale the Qwen-VL text encoder layer by layer — 60 sliders, one idea
Free your text encoder from VRAM the moment it's done
Punchy HDR look without re-rolling the generation
Prove which image is better, not just feel it
Peek inside a model's weights when the merge nodes make no sense
A checkpoint loader that also names your output folders for you
Load a UNet from either model folder, with fp8 options and a clean name
The loader that remembers what your model looked like before you edited it
Merge HiDream models block by block, not as one blob
Merge Z-Image models the way people actually use them
Re-save an edited model without mangling its keys or dropping tensors
Zero out or amplify specific ERNIE-Image layers, down to the attention heads
Scale Flux 2 Klein's blocks to tune a model that's actually runnable
Scale HiDream's double- and single-stream blocks independently
Scale a pixel-space model that has no VAE and no text encoder
Scale Krea 2 with slider names that match the loader, not the file
Scale a 60-block Qwen-Image model one transformer block at a time
The SDXL block scaler that started this pack's merge family
Per-block dials on the 6B model that became SDXL's successor
PNGs that remember what you told the model
The blunt instrument of VAE arithmetic
Mix two SDXL VAEs stage by stage, not all at once
Blend two VAEs when one's color and the other's texture both need to win
Carving one VAE's fingerprint out of another
Per-stage dials on Black Forest Labs' retrained autoencoder
The most fine-grained knob in this VAE-surgery kit
Dialing the decoder behind the 'too smooth' complaint
Turn one stage of your SDXL VAE up, down, or off
Scaling the 3D causal VAE that compressed open video
ComfyUI-easygoing-nodes
Custom nodes for ComfyUI focused on practical workflow improvements: workflow screenshots, model loading helpers, prompt metadata embedding, VRAM/RAM control, HDR color adjustment, and image difference analysis.
v2.0.0 supports ComfyUI Nodes 2.0 and the V3 node schema.
Updating from v1.x.x?
Nodes are designed to migrate automatically when opening old workflows. If migration fails in your environment, delete the old nodes and place the new ones again.
Features
1. Workflow Screenshots (Nodes 1.0 only)
<div align="center"> <img width="800" height="535" alt="workflow image screenshot" src="Images/workflow_image_screenshot_with_box.png"> </div>Right-click the canvas → Workflow Image → Export PNG to save a high-resolution screenshot of the current workflow with embedded metadata.
- Improved implementation based on the original from ComfyUI-Custom-Scripts, fixed for recent ComfyUI frontend changes.
- Exports at 4× resolution for clean display on high-DPI screens (4K monitors, tablets, etc.).
Note: Workflow export is currently supported only in Nodes 1.0. Nodes 2.0 (Vue + litegraph rendering) is not supported yet. Switch to Nodes 1.0 when using this feature.
2. Improved Model Loading
<div align="center"> <img width="800" height="281" alt="load model with name clean nodes" src="Images/load_model_with_name_clean.png"> </div>Load Checkpoint (with Name Clean) and Load Diffusion Model (with Name Clean)
- Lists models from both
models/checkpointsandmodels/diffusion_models. - Outputs a cleaned model name (
model_name_clean) with the following removed (case-insensitive):- Precision / format markers:
fp32,fp16,bf16,fp8,svd,scaled,mxfp8,nvfp4,int8,convrot,hq,e4m3fn,e4m3,e5m2fn,e5m2 - Extensions:
.safetensors,.pth,.pt - Trailing
-and_
- Precision / format markers:
Use the cleaned name with a trailing / in a Save Image node’s filename_prefix to automatically organize outputs into per-model subfolders.
3. CLIP Text Encode (with Offload)
<div align="center"> <img width="320" height="242" alt="clip text encode with offload node" src="Images/clip_text_encode_with_offload_node.png"> </div>Replaces the standard CLIP Text Encode node and explicitly frees the text encoder from VRAM (and optionally RAM) after encoding.
Useful on systems with limited VRAM/RAM where ComfyUI’s default model retention can reduce available memory. Performance impact depends on your hardware — test both ways.
4. Save Image With Prompt
<div align="center"> <img width="800" height="556" alt="Images/save image with prompt sample workflow" src="Images/save_image_with_prompt_sample_workflow.png"> </div>Saves images with the following fields stored as separate PNG metadata (in addition to the full workflow JSON):
positive_promptadditional_promptnegative_promptcaptionseed
Makes prompts easy to view and copy in image viewers such as XnView MP or digiKam.
Tip: Seeds are integers in ComfyUI. Pass them through a “Preview as Text” (or equivalent) node to convert to string before connecting to this node.
5. HDR Effects with LAB Adjust
<div align="center"> <img width="800" height="362" alt="hdr lab color adjusts sample workflow.png" src="Images/hdr_lab_sample_workflow.png"> </div>Advanced tone-mapping and color adjustment in LAB color space (luminance and chrominance separated). Controls include shadows, highlights, gamma, contrast, color boost, and per-channel LAB adjustments.
Based on the HDR processing from ComfyUI-SuperBeasts with additional color controls. Well suited for high-contrast looks.
6. Image Difference Checker
<div align="center"> <img width="800" height="594" alt="image difference checker sample workflow" src="Images/image_difference_checker.png"> </div>Compares two similar images and outputs:
- Difference map
- Similarity metrics (MAE, SSIM)
- Tone-curve / color analysis
Useful for evaluating VAE reconstruction accuracy, upscaler differences, etc. Use ComfyUI’s partial execution (blue play button) to re-run only from this node onward after changing settings.
Model Merging
Includes hierarchical model merging nodes (e.g. Model Scale SDXL) and related utilities for advanced blending while preserving structure.
Removed in v2.0.0
The experimental sdxl_clip.py replacement (enhanced attention mask / tokenization) has been removed. Recent ComfyUI core updates made it ineffective.
Installation
Recommended: ComfyUI Manager / Registry
<div align="center"> <img width="640" height="343" alt="comfyui nodes manager easygoing search with comment" src="Images/comfyui_nodes_manager_easygoing_search_with_comment.png"> </div>The pack is registered in the ComfyUI Registry. Search for Easygoing in the Nodes Manager and install.
Manual
cd ComfyUI/custom_nodes
git clone https://github.com/easygoing0114/ComfyUI-easygoing-nodes.git
Restart ComfyUI. Nodes appear under their respective categories (model loaders, image, etc.).
Requirement: A ComfyUI build that supports the V3 node API (comfy_api.latest). Older ComfyUI versions without V3 support will not register the nodes.
Update Notes (v1.8.5 → v2.0.0)
- Full support for Nodes 2.0 and V3 schema.
- Automatic node migration for existing workflows is implemented, but may not succeed in every environment.
- If nodes appear broken or missing after the update, delete the old instances and re-add the new ones.
- Updating ComfyUI itself is also recommended if you encounter issues.
Links
Credits
- HDR Effects based on ComfyUI-SuperBeasts
- Workflow screenshot improvements based on ComfyUI-Custom-Scripts