Sign In
homemodelsimagesvideos3D Modelshubsarticlescomicscruciblesupdatesshop

Red nodes on a downloaded ComfyUI workflow? The complete fix guide

0

Musetta's Avatar

Musetta

Oct 17, 2026

workflows
comfyuicustom nodestutorialguideworkflow
Red nodes on a downloaded ComfyUI workflow? The complete fix guide

You download a workflow from Civitai, drop it into ComfyUI, and half the graph turns red. Or it loads fine and dies on Queue with "Prompt outputs failed validation". Almost every broken workflow fails for one of six reasons. Here is how to find which one, and how to fix each by hand.

First, read what ComfyUI is telling you

  • Red nodes on load: a custom node pack is missing. Go to step 1.

  • "Value not in list" on Queue: a model file is missing or named differently. Go to step 2.

  • A node loads, but its widgets show strange values: the pack version changed. Go to step 3.

  • A traceback in the console mentioning a Python package: a dependency problem. Go to step 4.

  • "Allocation on device" or "out of memory": the workflow is heavier than your GPU. Go to step 5.

The console window, not the browser, has the real error. Scroll to the last lines of the traceback: the node name and the message are there.

1. Missing custom nodes (red nodes)

  1. Open ComfyUI Manager and press Install Missing Custom Nodes. It reads the workflow and lists the packs it knows.

  2. Install, then restart ComfyUI fully, not just refresh the browser.

  3. Still red? The pack is not in the Manager's list. Copy the node's type name (shown on the red node) and search for it on GitHub or the Comfy registry. The pack's README tells you how to install it.

  4. Installing by hand: git clone the pack into ComfyUI/custom_nodes, then install its requirements with ComfyUI's own Python. On the portable build that is python_embeded\python.exe -m pip install -r requirements.txt, run from the ComfyUI folder.

Watch for: two packs that register the same node name, and node names that changed between pack versions. The workflow's notes or the Civitai page usually say which pack the author used.

2. Missing models ("Value not in list")

The workflow stores model filenames. If your file is named differently, or lives in another folder, validation fails.

Find the exact file in the workflow's notes, its Civitai page (Resources used), or the model's Hugging Face page. Put it in the folder the loader node reads from:

  • Load Checkpoint: models/checkpoints

  • Load LoRA: models/loras

  • Load VAE: models/vae

  • Load Diffusion Model / UNET: models/diffusion_models (older installs: models/unet)

  • Load CLIP / text encoders: models/text_encoders (older installs: models/clip)

  • Upscale models: models/upscale_models

  • ControlNet: models/controlnet

  • CLIP Vision: models/clip_vision

Then press R in ComfyUI to refresh the model lists, and pick the file in each loader's dropdown.

Watch for: the same model in fp16, fp8 and GGUF variants. A GGUF file needs a GGUF loader node, not the standard one. Match the variant the workflow expects, or switch the loader.

3. Version breaks

A node pack updated and its inputs moved, so saved values land in the wrong widgets: a seed in a steps box, a sampler name where a scheduler should be.

  • Check each widget on the affected node against the workflow's screenshots or notes, and reset the wrong ones.

  • If a whole pack breaks, look at the pack's releases and install the version from around the workflow's date.

  • New-style workflows with subgraphs need a recent ComfyUI frontend. Update ComfyUI itself if a workflow opens with empty groups.

4. Dependency conflicts

One pack installs a package version another pack cannot use. The console shows an ImportError or a version complaint when ComfyUI starts, and the pack shows IMPORT FAILED in the Manager.

  • Read which package and which version the failing pack wants.

  • Install that version with ComfyUI's own Python, then restart.

  • If two packs truly need incompatible versions, keep the one the workflow uses and disable the other.

5. Out of memory

  • Lower the resolution or the batch size first.

  • Use an fp8 or GGUF variant of the big model.

  • Swap the VAE decode for a tiled VAE decode on large images and video.

  • Close other GPU apps. As a last resort, start ComfyUI with --lowvram.

6. Settings your install does not have

A sampler or scheduler name the workflow saved may not exist in your version, or comes from a custom pack. Pick the closest one your install lists, or install the pack that provides it.

A checklist before you run any downloaded workflow

  1. Read the notes on the canvas and the Civitai description.

  2. List every model the loaders ask for, with sizes. Big video models can be 10 GB or more.

  3. Install missing nodes, restart, then check the console for IMPORT FAILED.

  4. Queue once at a small size before the real render.

Or let the Brain do it

This is the job Musetta Studio was built for. Import a workflow from Civitai and the Brain opens it, finds every missing node pack and model, installs the packs and downloads the exact model files, on your machine or a rented GPU. It repairs what is broken until the workflow runs, and only says ready once it does. It explains each fix in plain language, gives the workflow a simple control panel and keeps it in one library with your models, matched against Civitai with their trigger words.

Musetta Studio is in early access: free to join, no card.

Stuck right now? Drop the workflow .json in #comfyui-help on the Musetta AI Discord or run /diagnose: the bot reads the node list and tells you the likely missing packs and where it breaks. People there help too.

0