ComfyUI DirectML AMD

PassAudited by ClawScan on May 7, 2026.

Overview

This appears to be a purpose-aligned ComfyUI setup helper, but it changes a local ComfyUI source file and uses external installs/downloads that users should review before running.

Before installing, confirm you are on Windows with an intended ComfyUI directory, review that apply-directml-fixes.py will patch comfy/model_patcher.py, and keep the generated backup. Treat the missing download-models.ps1 and benchmark.py references as documentation gaps, and do not run separately sourced helper scripts unless you inspect them first.

Findings (3)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

Running the script changes your ComfyUI installation and could break it if applied to an unexpected ComfyUI version or directory.

Why it was flagged

The reviewed helper rewrites ComfyUI's comfy/model_patcher.py after creating a backup. This is the documented DirectML fix, but it is still a local application-code mutation.

Skill content
backup_file(model_patcher) ... with open(model_patcher, 'w', encoding='utf-8') as f: f.write(content)
Recommendation

Run it only in the intended ComfyUI directory or with a trusted COMFYUI_PATH, keep the backup, and review the file diff if possible before continuing.

What this means

You will be downloading executable Python packages and large model files from external services.

Why it was flagged

The setup uses external package installation and model downloads. These are expected for a ComfyUI setup skill, but the shown commands do not include hashes or other verification steps.

Skill content
.venv\Scripts\python.exe -m pip install torch-directml --force-reinstall ... curl -L -o ... "https://huggingface.co/...safetensors"
Recommendation

Use official package/model sources, prefer pinned versions or checksums when available, and avoid running downloaded content from untrusted mirrors.

What this means

If you obtain similarly named scripts from somewhere else, they may not be the same as what this skill describes.

Why it was flagged

The documentation describes helper scripts for model downloads and benchmarking, but those files are not included in the provided manifest, so their contents were not reviewable here.

Skill content
### `download-models.ps1` ... ### `benchmark.py`
Recommendation

Only run the included apply-directml-fixes.py from this package, and inspect or verify any separately obtained downloader or benchmark scripts before using them.