AGENT SKILL

comfyui

Install and operate the MAGI-1 ComfyUI custom node and bundled workflows.

License 许可
Apache-2.0
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
comfyui
license
Apache 2.0
metadata
{ "disco-role": "operating" }
description
Install and operate the MAGI-1 ComfyUI custom node and bundled workflows.
disable-model-invocation
true

02 资源文件 6

03 SKILL.md 正文

MAGI-1 ComfyUI operating skill

Use this sub-skill when a user wants to install, recognize, run, adapt, or troubleshoot the MAGI-1 ComfyUI custom node and its example workflows.

Do not use this sub-skill for source-code CLI inference commands or direct MagiPipeline API usage; route those questions to the MAGI-1 inference sub-skill. ComfyUI itself is an external host application dependency; this skill captures MAGI node behavior and workflow adaptation, not a verified ComfyUI host import.

Fast operating path

  1. Install ComfyUI, then install the MAGI-1 node by one of these choices:
    • Registry install from inside the ComfyUI directory: comfy node registry-install MAGI-1.
    • Source install by placing the MAGI-1 source tree under ComfyUI/custom_nodes/MAGI-1 and installing MAGI-1 runtime dependencies into the same Python environment used by ComfyUI.
  2. After either install method, ensure ComfyUI sees a plugin-root __init__.py: move or copy the repository's comfyui/__init__.py to the MAGI-1 plugin root if it is not already there. The plugin root must still contain the comfyui/, inference/, and example/assets/ subtrees expected by that initializer.
  3. Download model weights and edit the selected MAGI JSON config so checkpoint paths are absolute for load, t5_pretrained, and vae_pretrained. Use absolute paths for the ComfyUI node's config_path, T5 path, input media, and save path fields.
  4. Launch ComfyUI with comfy launch or python main.py from the ComfyUI directory.
  5. Add nodes from Add Node → Magi or import one of the bundled workflows, then replace every placeholder path before queueing the graph:

Detailed references

Guardrails for future agents

  • State that imported workflow JSONs contain placeholders and must be reassigned inside the user's ComfyUI runtime.
  • State that MAGI's ComfyUI MagiProcess node sets single-node/single-GPU distributed environment variables internally before inference.
  • Do not claim ComfyUI host imports were verified unless the user has actually run ComfyUI with the MAGI node in their environment.
  • Keep all workflow links generated-skill-relative; do not link to source checkout files.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-triton-opt

Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-solve

END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-skill

PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-pattern-fusion

Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-orchestrate

Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-feedback

Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).

★ 351