Start here when something's broken and you're not sure why
#Step 1 — When does it break?
Before you can even open ComfyUI
Something failed during install or launch. Jump to Installation if python main.py won't start at all, or to Fix Missing Dependencies if it starts but crashes on a specific import error.
When you load a workflow
The workflow itself won't open, shows red/broken nodes, or throws an error the moment it loads (before you click Queue Prompt). That's Workflow Errors — missing custom nodes and wrong model types live there.
When you click "Queue Prompt"
The workflow loaded fine, but generation itself fails, crashes, or runs out of memory. That's GPU Errors for CUDA/VRAM issues, or Fix Missing Dependencies if the error is about a Python package rather than memory.
It runs, but the output is wrong or too slow
Black images, garbled output, or generation that "works" but takes forever — see the decision table below.
Hardware Partner
Running these workflows? ComputeAtlas.ai helps you find the right GPU
Optimization is only half the battle. Get precise VRAM benchmarks and hardware recommendations tailored for ComfyUI.
Check GPU Prices →#Step 2 — Match your symptom
| Symptom | Most likely cause | Fix guide |
|---|---|---|
| ComfyUI won't launch at all | Python/CUDA/venv setup issue | Installation |
| "Node not found" when loading a workflow | Missing custom node | Workflow Errors |
| "Input type mismatch" | Wrong checkpoint for the workflow (e.g. SD1.5 checkpoint in an SDXL workflow) | Workflow Errors |
| Output is solid black or garbled noise | Wrong VAE, bad sampler/scheduler pair | Workflow Errors |
| "CUDA out of memory" mid-generation | VRAM exceeded for your resolution/batch size | GPU Errors |
| Generation is technically working but painfully slow | No --lowvram/--highvram flag set for your tier, or missing xformers | Optimize Performance On Any GPU |
ModuleNotFoundError or a custom node crashes on import | A Python dependency didn't install | Fix Missing Dependencies |
| Video export fails but images generate fine | FFmpeg not installed or not on PATH | Fix Missing Dependencies |
| Can't decide between the portable build and desktop app | Not a bug — a setup choice | Portable vs. Desktop |
#Step 3 — Still stuck? Gather this before asking for help
Whether you're posting in a Discord, a GitHub issue, or asking an AI assistant, these five things resolve 90% of "it doesn't work" reports instantly:
- →The exact error text from the ComfyUI console window, not just the red banner in the browser UI — the console has the real stack trace.
- →Your GPU and VRAM (
nvidia-smiin a terminal shows both). - →Whether it ever worked — did this break after installing a new custom node or model, or has it never worked?
- →Your launch command (
python main.py --lowvram, etc.) — half of "won't run" reports are missing the flag their hardware needs. - →What changed most recently — a new custom node, a driver update, a Windows update. Recency is the single best predictor of the cause.
#Common false leads
- →"My GPU must be too weak" — check GPU Errors and the VRAM tables in Optimize Performance On Any GPU first. Most "too weak" cases are a missing flag, not the hardware.
- →"The workflow JSON is corrupted" — 9 times out of 10 this is a missing custom node or a model swapped for the wrong version, not a broken file. See Workflow Errors.
- →"I need to reinstall everything" — almost never necessary. Dependency conflicts (see Fix Missing Dependencies) look identical to a broken install but are fixed in one
pip install, not a full wipe.
#Related Guides
- →Fixing GPU Errors
- →Understanding and Fixing Workflow Errors
- →Fixing Missing Dependencies
- →Optimize Performance On Any GPU
Hardware Partner
Running these workflows? ComputeAtlas.ai helps you find the right GPU
Optimization is only half the battle. Get precise VRAM benchmarks and hardware recommendations tailored for ComfyUI.
Check GPU Prices →