If you use tools such as ComfyUI, it is remarkably easy to accumulate models that consume massive amounts of disk space.
You might download a model only to discover that it is too large for your graphics card. You might be experimenting with different models, downloading several alternatives before settling on one. Sometimes you download a model into the wrong folder, or import a workflow that expects a model in a particular subdirectory, only to end up with two identical copies in different locations.
Over time, the situation gets worse. You stop using certain models, delete the workflow JSON files that reference them, but leave the models themselves behind. Eventually, you have hundreds of gigabytes of models taking up disk space.
ComfyUI-Cleaner
I use ComfyUI Easy Install and wanted a practical way to identify models I probably no longer needed, without manually examining hundreds of files.
So I vibe coded ComfyUI-Cleaner, a Python utility designed to help identify unused, duplicated and misplaced models.
The project consists of two separate utilities.
1. Finding unused models
find_unused_models.py examines the models installed in ComfyUI and compares them against references in your saved workflows.
It identifies:
- Orphaned models: Models not referenced by any of your saved workflows.
- Duplicates: Models appearing in multiple locations.
- Misplaced models: Models that workflows expect to find somewhere other than their current location.
The results are displayed in an interactive list, showing model names, sizes and their classification. You can review the findings, decide which models to retain or delete, and identify models that should be moved.
Your selections are saved to unused_models.txt, a plain-text file that you can also review and edit using your text editor.
# ComfyUI-Cleaner 1.0.0-beta - review list
# Created 2026-11-11 14:10 for D:\ComfyUI-Easy-Install\ComfyUI\models
#
# !!! ==========================================================================
# !!! ATTENTION: models outside your workflows were found. 1 model (19.53 GB) is
# !!! not used by any of your workflows, only by ComfyUI's own templates or
# !!! blueprints (see "template model" below). It is marked for deletion by
# !!! default and will be deleted. To keep it by default, run
# !!! find_unused_models.py with --keep-template-models.
# !!! ==========================================================================
#
# delete_unused_models.py carries out every line that does NOT start with '#':
# a path the model is deleted
# MOVE a -> b the model is moved from a to b
# To skip a line, delete it or put '#' in front of it. To carry out a line
# that starts with '#', remove the '#'.
#
# orphan no workflow uses it. Only orphans verified online are marked
# by default.
# template model an orphan that only ComfyUI's own templates or blueprints load.
# Marked like other orphans unless --keep-template-models is used.
# duplicate the same name exists in more than one place. Choose which copies
# to delete. If a workflow uses the file, one copy has to stay.
# misplaced a workflow names it but looks for it in another folder. It is not
# deleted, but it can be moved to where the workflow expects it.
# custom node a custom node may load it without the workflow naming it - check
# before deleting.
#
# Marked: 1 to delete (19.53 GB), 0 to move
# ===== Duplicates - the same name in more than one place =====
# --- Duplicate 1: minimax_h3_fl2va_pruned_int8_convrot.safetensors (same size)
# 19.53 GB | file | used by 3 workflows | HF: Comfy-Org/MiniMax-H3 | https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/diffusion_models/minimax_h3_fl2va_pruned_int8_convrot.safetensors
# D:\ComfyUI-Easy-Install\ComfyUI\models\diffusion_models\h3\minimax_h3_fl2va_pruned_int8_convrot.safetensors
# 19.53 GB | file | used by 9 workflows | also loaded here by 1 blueprint and 6 templates | HF: Comfy-Org/MiniMax-H3 | https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/diffusion_models/minimax_h3_fl2va_pruned_int8_convrot.safetensors
# loaded by: blueprint "Image to Video (MiniMax H3)", template "image_minimax_h3_t2i", template "video_minimax_h3_i2v" (+4 more)
# D:\ComfyUI-Easy-Install\ComfyUI\models\diffusion_models\minimax_h3_fl2va_pruned_int8_convrot.safetensors
# --- Duplicate 2: minimax_h3_ref2va_pruned_int8_convrot.safetensors (same size)
# 19.53 GB | file | used by 3 workflows | HF: Comfy-Org/MiniMax-H3 | https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/diffusion_models/minimax_h3_ref2va_pruned_int8_convrot.safetensors
# D:\ComfyUI-Easy-Install\ComfyUI\models\diffusion_models\h3\minimax_h3_ref2va_pruned_int8_convrot.safetensors
# 19.53 GB | file | orphan (the workflows that name it use another copy; template model: only 4 templates load it) | HF: Comfy-Org/MiniMax-H3 | https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/diffusion_models/minimax_h3_ref2va_pruned_int8_convrot.safetensors
# loaded by: template "video_minimax_h3_fun_controlnet_union", template "video_minimax_h3_multiframe_reference", template "video_minimax_h3_r2v" (+1 more)
D:\ComfyUI-Easy-Install\ComfyUI\models\diffusion_models\minimax_h3_ref2va_pruned_int8_convrot.safetensors
# --- Duplicate 3: krea2_turbo_int8_convrot.safetensors (same size)
# 12.57 GB | file | used by 4 workflows | also loaded here by 1 blueprint and 2 templates | HF: Comfy-Org/Krea-2 | https://huggingface.co/Comfy-Org/Krea-2/resolve/main/diffusion_models/krea2_turbo_int8_convrot.safetensors
# loaded by: blueprint "Image Style Reference (Krea-2 Turbo)", template "image_krea2_turbo_int8_image_style_reference", template "image_krea2_turbo_t2i_int8"
# D:\ComfyUI-Easy-Install\ComfyUI\models\diffusion_models\krea2_turbo_int8_convrot.safetensors
# 12.57 GB | file | misplaced, expected at diffusion_models/krea2/krea2_turbo_int8_convrot.safetensors | also loaded here by 1 blueprint and 2 templates | HF: Comfy-Org/Krea-2 | https://huggingface.co/Comfy-Org/Krea-2/resolve/main/diffusion_models/krea2_turbo_int8_convrot.safetensors
# moving it breaks: blueprint "Image Style Reference (Krea-2 Turbo)", template "image_krea2_turbo_int8_image_style_reference", template "image_krea2_turbo_t2i_int8"
# D:\ComfyUI-Easy-Install\ComfyUI\models\vae\krea2_turbo_int8_convrot.safetensors
# MOVE D:\ComfyUI-Easy-Install\ComfyUI\models\vae\krea2_turbo_int8_convrot.safetensors -> D:\ComfyUI-Easy-Install\ComfyUI\models\diffusion_models\krea2\krea2_turbo_int8_convrot.safetensors
# ===== Misplaced models - a workflow names them but looks in another folder =====
# 5.87 GB | file | misplaced, expected at diffusion_models/z-image/z_image_turbo_int8_convrot.safetensors | also loaded here by 1 template
# moving it breaks: template "image_z_image_turbo_int8"
# MOVE D:\ComfyUI-Easy-Install\ComfyUI\models\diffusion_models\z_image_turbo_int8_convrot.safetensors -> D:\ComfyUI-Easy-Install\ComfyUI\models\diffusion_models\z-image\z_image_turbo_int8_convrot.safetensors
An additional feature is online verification. The utility attempts to locate unused models on Hugging Face or Civitai, helping establish whether they could be downloaded again if needed.
Nothing is deleted during this stage.
2. Removing unwanted models
delete_unused_models.py processes the reviewed list.
Before making changes, it rescans the installation and validates the proposed operations, including checks intended to protect models still required by workflows.
You can run it in dry-run mode to preview everything before committing to any changes.
Deletion requires explicit confirmation, and the utility maintains a log of deleted or moved models, including download links where available.
2026-11-11 14:46:27 MOVED 13492686496 D:\ComfyUI-Easy-Install\ComfyUI\models\vae\krea2_turbo_int8_convrot.safetensors D:\ComfyUI-Easy-Install\ComfyUI\models\diffusion_models\krea2\krea2_turbo_int8_convrot.safetensors ComfyUI-Cleaner 1.0.0-beta
2026-11-11 14:46:27 MOVED 6299127240 D:\ComfyUI-Easy-Install\ComfyUI\models\diffusion_models\z_image_turbo_int8_convrot.safetensors D:\ComfyUI-Easy-Install\ComfyUI\models\diffusion_models\z-image\z_image_turbo_int8_convrot.safetensors ComfyUI-Cleaner 1.0.0-beta
2026-11-11 14:46:58 DELETED 20970379616 D:\ComfyUI-Easy-Install\ComfyUI\models\diffusion_models\minimax_h3_fl2va_pruned_int8_convrot.safetensors https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/diffusion_models/minimax_h3_fl2va_pruned_int8_convrot.safetensors ComfyUI-Cleaner 1.0.0-beta
2026-11-11 14:46:58 DELETED 20970379616 D:\ComfyUI-Easy-Install\ComfyUI\models\diffusion_models\minimax_h3_ref2va_pruned_int8_convrot.safetensors https://huggingface.co/Comfy-Org/MiniMax-H3/resolve/main/diffusion_models/minimax_h3_ref2va_pruned_int8_convrot.safetensors ComfyUI-Cleaner 1.0.0-beta
Important: Deletion is permanent. Files do not go to the Recycle Bin.
A few things worth knowing
This is not a foolproof way to determine whether a model is genuinely unnecessary. A model might be loaded dynamically by a custom node, referenced by workflows stored elsewhere, or used by ComfyUI's built-in templates.
The utility accounts for several of these situations, but the final decision remains yours. Models used only by templates can still be suggested for deletion, so reviewing the proposed list is essential.
The utility was developed for my Windows Easy Install setup, although it also supports other ComfyUI installations. It requires Python 3.9 or later, with no additional Python packages.
Available on GitHub
ComfyUI-Cleaner is free to use and currently at version 1.0.0-beta.
It is a vibe-coded project created to solve a practical problem I encountered. I am sharing it in case others find it useful.
The repository includes the source code, configuration details and instructions for using both utilities.
GitHub: https://github.com/chribonn/ComfyUI-Cleaner
As with any utility that deletes files, review the results carefully and keep backups of anything important.
This is still a beta release, and I am sure there are scenarios I have not considered. If you have ideas for improvements or encounter bugs, please document your particular case and share your findings through the GitHub repository. Include enough detail to explain what happened, what you expected to happen, and, where possible, how the issue can be reproduced.
The more real-world situations we can identify, the more useful and reliable the utility can become.



Top comments (0)