ComfyUI Progress Bar and Day-One Settings

TL;DR

• Install Crystools via ComfyUI Manager to get a progress bar, time elapsed display, and real-time CPU/GPU/RAM/VRAM monitoring
• Configure startup flags like --preview-method, --cache-ram, and --async-offload on first launch
• Use Python 3.13 for best compatibility with current ComfyUI and custom nodes

Related Content: See ComfyUI Beginner Setup, ComfyUI Installation Import Errors and Benchmark ComfyUI Setup for more on getting started.

How to See Progress While Generating

ComfyUI's default interface shows minimal feedback during generation—you'll see a spinner and the queue count, but not which node is currently executing or how much time remains. The Crystools extension solves this with a progress bar and timing display (source).

Installing Crystools

Install Crystools through ComfyUI Manager by searching for "crystools". The extension requires ComfyUI build 1915 or higher.

Progress Bar and Time Elapsed Features

Once installed, Crystools adds:

1. A progress bar in the menu showing workflow completion percentage
2. Time elapsed display that appears when your workflow finishes
3. Click-to-reveal functionality—clicking the progress bar shows which node is currently executing

You can see the progress of your workflow with a progress bar on the menu! Additionally, it shows the time elapsed at the end of the workflow, and you can click on it to see the current working node.

The progress bar can be disabled from settings if you prefer the default interface.

Resource Monitoring

Crystools also includes a real-time hardware monitor showing CPU, GPU, RAM, VRAM usage, and temperature. This overlay appears in the menu and updates at configurable intervals.

The monitor has minimal overhead (0.1 to 0.5% utilization) and works with NVIDIA GPUs via CUDA.

What Settings to Change on First Install

Python Version

Use Python 3.13 for best compatibility (source). Python 3.14 works but some custom nodes may have issues. Python 3.12 is a good fallback if you encounter dependency problems.

Default Port

ComfyUI runs on port 8188 by default. If another service uses this port, start ComfyUI with --port 8189 (or any available port).

Installation Methods

ComfyUI offers three installation paths (source):

For full control, use the comfy-cli tool with comfy install. Use comfy install --fast-deps to leverage uv for faster dependency resolution (source).

Essential Startup Flags

For day-one configuration, consider these flags in your launch command (source):

Model Folder Structure

ComfyUI expects models in specific subdirectories (source):

ComfyUI/models/
├── checkpoints/     # Main diffusion models (.ckpt/.safetensors)
├── diffusion_models/ # Alternative model location
├── loras/           # LoRA adapters
├── vae/             # VAE models
├── clip/            # CLIP text encoders
├── controlnet/      # ControlNet models
└── embeddings/      # Textual inversions

Custom paths can be added via extra_model_paths.yaml.

Output Directory

By default, generated images save to output/. Override this with --output-directory /path/to/outputs (source).

Day-One Performance Settings for Low VRAM

ComfyUI includes smart memory management that can run large models on GPUs with as little as 1GB VRAM (source).

Essential Flags for Limited VRAM

Dynamic VRAM Management

By default, ComfyUI enables dynamic VRAM on NVIDIA GPUs. This loads and unloads model parts based on memory pressure. You can explicitly enable it with --enable-dynamic-vram.

Async Offloading

Async weight offloading is enabled by default on NVIDIA cards. This moves model weights to CPU when not actively used, freeing VRAM for the current operation.

Disk-Backed Loading

With fast NVMe storage, use --fast-disk to prefer disk-backed dynamic loading over RAM buffers.

Working with Qwen Image Models

Qwen Image is a 20B diffusion model supported natively in ComfyUI (source).

Required Model Files

Download and place these files in their respective folders (source):

For image editing workflows, use qwen_image_edit_2509_fp8_e4m3fn.safetensors instead of the base model.

Next Scene Generation with LoRAs

For sequential scene generation (storyboards, comics), the next-scene LoRA works with Qwen Image Edit (source).

Model Versions

The LoRA exists in two versions:

Installation

Download the V2 LoRA from HuggingFace and place in models/loras/.

Usage

1. Load Qwen Image Edit 2509 as your base model
2. Add a LoRA Loader node with strength 0.7–0.8 (source)
3. Structure prompts with "Next Scene:" prefix for cinematic continuity (source)

Common Errors and Fixes

GPU Monitor Shows No Data

Cause: Running on AMD/Intel/Apple Silicon

Fix: Crystools resource monitor only works with NVIDIA CUDA GPUs. Disable the monitor if you're not on NVIDIA hardware.

Crystools Monitor Not Appearing

Cause: ComfyUI version too old

Fix: Crystools requires ComfyUI 1915 or higher. Update ComfyUI via ComfyUI Manager or git pull.

Chrome Visual Glitches

Cause: Using Chrome version 142 or below

Fix: Update to Chrome 143 or later for best ComfyUI experience.

Python Version Issues with Custom Nodes

Cause: Using Python 3.14 with incompatible dependencies

Fix: Fall back to Python 3.12 if you encounter issues.

"Failed to load checkpoint"

Cause: Model file corrupted or incompatible architecture

Fix: Re-download the checkpoint and ensure all components match (e.g., don't mix SDXL models with 1.5 components).

CUDA Out of Memory

Fix: Reduce image resolution, lower batch size, enable --lowvram or --novram, or restart ComfyUI to clear memory.

FAQ

How do I see progress while generating in ComfyUI?

Install the Crystools extension via ComfyUI Manager (source). It adds a progress bar to the menu that shows completion percentage and elapsed time. Click the bar to see which node is currently executing.

How do I know which node is currently running in ComfyUI?

With Crystools installed, click on the progress bar in the menu. It will highlight the currently executing node. Alternatively, watch the console output or enable preview images on sampler nodes.

What settings should I change in ComfyUI on first install?

Set Python 3.13 as your Python version, verify your port (default 8188), consider enabling --preview-method latent2rgb for latent previews, and configure caching with --cache-ram. For low VRAM systems, add --lowvram or --novram.

How do I optimize ComfyUI for low VRAM?

Use the --lowvram flag to run text encoders on CPU, --novram for minimal VRAM usage, --force-fp16 to halve model memory, and --async-offload for weight offloading. Enable dynamic VRAM with --enable-dynamic-vram.

What is the default port for ComfyUI?

ComfyUI defaults to port 8188. Change it with --port 8189 or any available port.

How do I install ComfyUI dependencies faster?

Use comfy install --fast-deps with comfy-cli, which leverages uv for faster dependency resolution compared to pip (source).

Does ComfyUI work on Apple Silicon Macs?

Yes, ComfyUI supports Apple Silicon (M1/M2/M3/M4) via the MPS (Metal Performance Shaders) backend. Use the Comfy Desktop application for the easiest setup on macOS ARM.

Can I run ComfyUI on CPU only?

Yes, use the --cpu flag when launching ComfyUI. Expect significantly slower generation times compared to GPU acceleration.

How do I enable preview images during generation?

Enable preview images during generation with --preview-method latent2rgb for latent previews or --preview-method taesd for TAESD-based previews. You can also right-click sampler nodes and select "Preview" during execution.

Sources