TROUBLESHOOT
Turn an error log into a reading path.
Paste an error or describe a symptom. Matching stays in your browser; no log upload or repair execution.
Local matching · Input is not put in the URL, browser storage or analytics events. Clear or leave this page to discard it. Exports contain article pointers only. Third-party browser extensions are outside this site’s control.
Article retrieval only—not an inspection of your Python, models or GPU. Ranking is not a diagnosis or repair success rate.
BROWSE BY CATEGORY
Or browse troubleshooting articles
Every published error article in this language, grouped by the kind of problem. Open a group to see its articles.
Environment and devices18
- Bad linked input: represent API connections with a source node and output slot
- ComfyUI archive will not extract: check the archive before the destination
- CUDA out of memory: separate model loading, sampling and VAE decoding
- Float8 is unsupported on MPS: distinguish stored weights from compute dtype
- Hugging Face 401 or 403: browser access and download-process authorization differ
- MPS backend out of memory on Mac: account for unified-memory pressure
- No space left on device: check caches, temporary files and output volumes
- Numeric input validation failed: use the node's actual types and bounds
- Prompt has no outputs: provide an actual execution endpoint
- Required input is missing: identify the exact node and field
- Return type mismatch: inspect the source output slot and destination input
- Torch not compiled with CUDA enabled: verify the build and the intended backend
- torch.cuda.is_available() is False: inspect the build, driver and runtime context
- VAE out of memory after sampling finishes: diagnose decoding separately
- Windows blocked the launcher: verify the publisher before adding exceptions
- Windows error 1455: paging-file exhaustion is not CUDA out of memory
- Windows path too long: distinguish a shorter install root from long-path support
- xFormers is installed but unavailable: inspect its native extension and operators
Dependencies and custom nodes50
- aiohttp and requests: a missing library is not a connection failure
- argostranslate is missing, or the language pair is missing: diagnose the right layer
- blend_modes imports but blending fails: inspect the image-array contract
- Cannot import name after an update: check the importing code and dependency together
- ComfyUI Manager button missing: distinguish UI changes from load failures
- ComfyUI missing nodes: distinguish absent packages, failed imports and unregistered classes
- Custom validation failed: read the node's own error details
- decord reads video but cannot use the GPU: distinguish the installed build
- deepdiff failures: imports, object comparison and result serialization
- dlib installation failed: separate the binding, native build and model resources
- DLL load failed: identify the native extension instead of downloading loose DLLs
- Fix cv2 imports without stacking OpenCV distributions
- flash_attn will not install: check the implementation, platform and build requirements
- InsightFace installation or initialization fails: separate packages from FaceID assets
- Kornia import errors: check the base package, Torch and optional extras
- llama_cpp failures: distinguish installation, native builds and model loading
- Manager cannot find Git: check the environment that launches ComfyUI
- Manager security policy blocks an action: confirm authorization and scope
- MediaPipe and NumPy: distinguish dependency constraints from binary import failures
- Missing safetensors package or unreadable weights? Diagnose them separately
- MMCV installation and missing-operator errors: check the build combination
- ModuleNotFoundError in ComfyUI even though the package is installed
- No matching distribution found: check Python tags, platform and package identity
- No module named hydra: check hydra-core, then the configuration search path
- No module named yaml, PIL or pydantic_settings: check the distribution name
- NumPy ABI errors: _ARRAY_API and dtype size changed
- Nunchaku still fails after installation: align the node, wheel and model
- nvvfx is missing: identify the wrapper before treating the import as a package name
- ONNX Runtime imports, but GPU execution fails: check providers and sessions
- OpenGL_accelerate is missing: a Python accelerator is not a graphics-driver repair
- pip certificate verification failed: repair trust without disabling TLS
- pip ResolutionImpossible: resolve conflicting requirements, not the download speed
- Protobuf Descriptors cannot be created directly: identify the generated code's owner
- pytorch_lightning or lightning is missing: follow the node's actual import path
- Qwen Image 2.1 missing nodes: separate core support, frontend support and subgraphs
- Qwen TTS conflicts with Transformers: inspect the pins before downgrading ComfyUI
- rotary_embedding_torch is missing, or the rotary tensor dimensions do not match
- SageAttention or Triton import failure: match the platform and build combination
- SAM and GroundingDINO failures: separate code, checkpoints and extensions
- SciPy, einops and torchsde: distinguish missing imports from failed operations
- SenseVoice-ONNX and setuptools: compare the installed distribution with its source
- simpleeval: distinguish a missing module from an invalid expression
- soundfile errors: separate the Python package, libsndfile and audio input
- Spandrel is installed but the model will not load: check architecture support
- SQLAlchemy and Alembic failures: preserve local data before changing anything
- timm is missing, a model name is unknown, or weights will not download
- TorchScale conflicts with timm: exact consumer pins are not fixed by upgrading everything
- Transformers errors: separate imports, API changes and model downloads
- webcolors is installed but CSS3_HEX_TO_NAMES is missing: check API changes
- Windows Overlapped operation errors in Manager: check the documented event-loop option
Models, files and workflows16
- A GGUF file will not load: match the architecture and the specialized loader
- CLIP input is invalid: None — trace where the text encoder should come from
- CLIP Vision missing or incompatible: it is not the prompt's text encoder
- ControlNet fails or has no visible effect: check the base, control image and apply path
- Dependency cycle detected: trace the reported node IDs through the graph
- Dragging an image does not restore a workflow: verify that metadata survived
- Expected all tensors to be on the same device: inspect branch boundaries
- Float, Half and BFloat16 mismatch: trace the failing operation
- IPAdapter model not found: unified presets and manual loaders use different rules
- LoRA has no effect or reports size mismatch: check the base model and both output branches
- mat1 and mat2 shapes cannot be multiplied: inspect the model component chain
- Node has no class_type: distinguish API export errors from missing registration
- Qwen Image 2.1 tensor or decode failures: check the component set
- Safetensors header errors: verify the artifact before changing loaders
- VAE and latent channel mismatch: verify the exact variant, not just its family name
- Value not in list: why a downloaded model does not appear in ComfyUI
Interface, connections and APIs10
- Address already in use: identify the process before stopping anything
- Blank ComfyUI screen after an extension update: separate frontend and backend failures
- Cannot connect to 127.0.0.1:8188: locate the running ComfyUI service
- ComfyUI Failed to fetch: identify the request before reinstalling
- ComfyUI workflow save failed: preserve the canvas before troubleshooting
- Core updated but nodes or templates are missing: check component versions separately
- Manager cannot load the node list: investigate retrieval before reinstalling nodes
- POST /prompt returns 400: inspect error and node_errors before retrying
- Queue finishes quickly but no new file appears: check cache and output location
- WebSocket connects but no result arrives: track prompt-specific terminal states
Generation, video and output7
- Black images or invalid colors: separate sampling, VAE and display failures
- FFmpeg audio contains NaN or Inf: trace effects and decoding upstream
- FFmpeg not found: test the executable in ComfyUI's process environment
- No kernel image is available: check GPU architecture support in each binary
- No module named av: why installing FFmpeg may not fix the Python import
- VHS output is nearly silent: inspect audio dtype before increasing gain
- Video export fails or plays incorrectly: check frames, encoding and audio separately