ComfyUI のエラーと問題(ノード別)
ガイドで扱っているすべてのエラーと問題を、原因となるノード別にまとめました。各リンクから説明と対処法に移動します。
このページはまだ翻訳されていないため、英語で表示しています。
AD
パソコンのいらない ComfyUI
グラボがなくても大丈夫。BitVector は同じオープンソースモデルと ComfyUI ワークフローをクラウドで動かします。スマホ、タブレット、どの PC からでも。プリインストール、セットアップゼロ、定額。
CLIP Text Encode (Prompt)
- Token indices sequence length is longer than the specified maximum (77)
- The prompt seems ignored and the image barely changes
- Error: embedding not found / "embedding:xyz" printed in the output
- Weighted words like (red:1.5) produce artifacts
ComfyUI Manager
- Red nodes remain after installing missing nodes
- IMPORT FAILED for a pack
- Manager button missing
- Update All broke everything
- This action is not allowed with this security level
KSampler
- Noise, static or a blurry blob instead of an image
- Deep-fried, oversaturated, high-contrast output
- torch.OutOfMemoryError: CUDA out of memory during sampling
- Error: mat1 and mat2 shapes cannot be multiplied
- Images change every run although the seed is fixed
- The same image repeats across prompts (seed stuck)
Load Checkpoint
- ckpt_name shows "undefined" or the dropdown is empty
- Error while deserializing header: HeaderTooLarge or MetadataIncompleteBuffer
- Images come out as noise or grey squares with a valid checkpoint
- Clip missing: ["text_projection.weight"] or "clip missing" warnings in the console
- Out of memory as soon as the checkpoint loads
Save Image
- Images are not in the output folder
- PermissionError when saving
- Dragging the PNG into ComfyUI does not restore the workflow
- Colours are washed out compared with the preview
VAE Decode
- Black image after decode
- Given groups=1, weight of size [...] expected input to have 4 channels, but got 16
- Out of memory in VAE Decode
- Visible grid seams in the decoded image
AD
このワークフローをスマホから実行
このページのワークフローはすべて、モデルとカスタムノードを揃えた状態で BitVector にプリインストール済み。選んでプロンプトを入力するだけ。セットアップゼロ、ダウンロード不要。
Empty Latent Image
- Flux / SD3 / Wan produce garbage with this node
- Two heads, duplicated people or tiling patterns
- Error: height and width must be divisible by 8
- Out of memory with a batch
Load Image
- Prompt outputs failed validation: Invalid image file
- MASK output is empty / inpainting changes nothing
- The whole picture is masked instead of the painted area
- Upload button does nothing or the browser hangs
Load LoRA
- lora key not loaded: lora_unet_... (dozens of lines)
- The LoRA has no visible effect
- Images are oversaturated, deep-fried or full of artifacts
- Error: Expected size for first two dimensions of batch2 tensor to be... (shape mismatch)
Load Diffusion Model
- ERROR: Could not detect model type
- Flux output is a blurry mess or pure noise
- torch.OutOfMemoryError when loading a 24 GB Flux file
- fp8_e4m3fn_fast is slower or errors on my GPU
Apply ControlNet
- The result ignores the guide
- Error: mat1 and mat2 shapes cannot be multiplied / ControlNet mismatch
- Burned, high-contrast, posterised image
- Old workflow shows "Apply ControlNet (OLD)"
- Flux ControlNet error about missing vae
Preview Image
- Preview disappeared after restart
- The graph runs even though the preview is not connected to the sampler
- Preview shows a black square
FaceDetailer (Impact Pack)
- No face detected, image passes through unchanged
- The new face does not match the body (colour or style)
- Visible square around the face
- ModuleNotFoundError: ultralytics / Impact Subpack missing
- Faces get a second set of eyes or a different person
Upscale Image (using Model)
- torch.OutOfMemoryError during the upscale
- Error: upscale model is not a supported architecture
- Result looks oversharpened or plastic
- Visible grid lines in the output
Ultimate SD Upscale
- Faces or objects appear in textures (grass, walls)
- Visible grid of tiles
- Very slow: hundreds of tiles
- The node is missing after import (red node)
Video Combine (VideoHelperSuite)
- Video formats are missing, only gif/webp are listed
- Playback is too fast or too slow
- The preview shows one frame or a still image
- Error: height not divisible by 2 (h264)
DualCLIPLoader
- The dropdown does not show my t5xxl file
- ERROR: clip input is invalid: None, or "shape mismatch" during text encoding
- Prompt encoding takes a very long time or spills to system RAM
IPAdapter (IP-Adapter Plus)
- ClipVision model not found / IPAdapter model not found
- Error: insightface model is required for FaceID models
- The prompt is ignored, output is a copy of the reference
- Error: Expected size 1280 but got 1024 (or similar)
VAE Encode and VAE Encode (for Inpainting)
- The inpainted region ignores the surroundings and looks pasted in
- Hard edges around the repainted area
- Nothing changes in img2img
- Error: expected 4 channels but got 16 (or reverse) when encoding
Upscale Image and Upscale Image By
WanImageToVideo and Wan 2.x video nodes
- Error: length must be ... / shape mismatch in the latent
- Output is a still image with no motion
- clip_vision_output is required / wrong CLIP Vision model
- Out of memory at the VAE decode
- Load CLIP fails with the umt5 file
Load ControlNet Model
- Error: ControlNet model type mismatch / shape mismatch
- The ControlNet has no visible effect
- Dropdown is empty
Load VAE
- Images are washed out, grey or low contrast
- The decoded image is a block of colour noise
- Black image after decode on SDXL
- VAE decode runs out of memory on large images
Flux Guidance
- Flux images are washed out and blurry
- Images look plastic, oversaturated or have grid artifacts
- The value does nothing on Flux Schnell
Power Lora Loader (rgthree)
Upscale Latent and Upscale Latent By
- The upscaled result is soft or blurry
- The second pass changes the image completely
- Extra limbs or doubled details appear after upscaling
Load Upscale Model
- Error: Unsupported model type / spandrel could not load
- The dropdown is empty
- Out of memory during Upscale Image (using Model)
Unet Loader (GGUF)
- The .gguf file does not appear in the dropdown
- Error: unknown GGUF architecture / unsupported quant type
- LoRA produces warnings about missing keys
- Still out of memory at Q4
KSampler (Advanced)
- Second stage output is pure noise
- Second stage looks over-smoothed or ignores the prompt
- Nothing happens in the second sampler (output equals input)
ReActor Fast Face Swap
- ModuleNotFoundError: insightface or onnxruntime
- No face detected in source image
- The swapped face is blurry or plastic
- The node is red / NSFW filter error
AnimateDiff Loader (Evolved)
- Frames are unrelated pictures, no motion
- Error: motion module is incompatible with the model (SD1.5 vs SDXL)
- Flickering and colour shifts between frames
- Out of memory with 32 frames
CLIP Set Last Layer
- Anime model looks muddy, generic or off-style
- The setting has no effect
- Error on Flux or SD3 workflows
EmptySD3LatentImage
- Flux output is noise although everything else is right
- RuntimeError about channel sizes (expected 16, got 4)
Load Video (Upload) (VideoHelperSuite)
- Out of RAM / the browser tab crashes on load
- Error: could not open video / no frames
- Audio output is None
ModelSamplingFlux, ModelSamplingSD3 and ModelSamplingAuraFlow
Efficient Loader and KSampler (Efficient)
- The pack fails to import after a ComfyUI update
- XY Plot produces a single image
- Prompt weights like (word:1.3) behave differently
Set Latent Noise Mask
- The whole image changes, not only the masked part
- A visible rectangle or outline around the edit
- The masked region stays almost identical
WAS Node Suite
- IMPORT FAILED: No module named cv2 / skimage / numba
- Startup takes a long time after installing WAS
- Image Save writes nowhere I can find
Conditioning (Combine) and (Concat)
- Subjects bleed into each other with Set Area or Set Mask
- Error: cond shapes do not match when using Concat
- Nothing changes when combining
Load Image (as Mask)
- The mask selects the wrong area (inverted)
- Mask is all zeros
- Hard, visible seam at the edge of the inpainted region
Primitive, Reroute, Note and Group nodes
Grow Mask, Feather Mask and Invert Mask
- A soft halo appears around the inpainted object
- Grow Mask changes nothing
- Edits leak outside the painted area
AD
ノード修理に疲れた?クラウドに任せよう
BitVector は数百の ComfyUI ワークフローを高速クラウド GPU 上でインストール・更新・検証済みの状態に保ちます。Python も CUDA エラーも VRAM 制限もなし。どのブラウザでも動きます。
Model
Trends
.ai
© ModelTrends.ai
|
日本製
|
© 2026