A simple, clean way to generate images with the Anima model in ComfyUI — without touching node graphs. Write a prompt, hit Generate. It also includes a built-in library for managing your LoRAs and styles.

Install

Easiest — ComfyUI-Manager (any OS, any install): open the Manager, search PeroPixfy, click Install, then restart ComfyUI. It uses whatever Python your ComfyUI runs, so it works the same on Windows portable, venv, Linux, macOS, and the Desktop app.

Windows portable — one-click .bat: download peropixfy_install.bat (from the latest release), put it in your ComfyUI folder (the one that contains ComfyUI and python_embeded), and double-click it. (Needs Git for Windows.) Then restart ComfyUI.

Manual (any OS): see Manual install below.

First run

Open the PeroPixfy tab in the ComfyUI sidebar — it switches to fullscreen (← ComfyUI or ESC to go back).

No models yet? On first run PeroPixfy offers to download the Anima models for you with one click, so even a brand-new, empty install works.

What you can do

Single — make one image at a time: pick a model, write your prompt, fine-tune it, redo details, and upscale for sharper results.
Multi — generate a whole batch at once (a list of variations × how many each), then quickly keep the ones you like.
Library — browse and organize your LoRAs and styles.

Updating

Via ComfyUI-Manager: Update the node and restart. On Windows portable you can instead put peropixfy_update.bat in the same folder, double-click it, then restart. (Re-running the installer also updates.)

Building from source, project layout, and internals → Development below.

Development

PeroPixfy runs on the ComfyUI engine itself. Instead of the node graph you work in a web app served by ComfyUI's own server — normally opened from the sidebar tab (a fullscreen overlay; the address bar stays on the ComfyUI page), or reached directly at http://127.0.0.1:8188/peropixfy. Generation goes through the same server's queue and graph executor, so output is pixel-identical to ComfyUI (verified across t2i / LoRA chain + bypass / i2i).

Manual install

cd ComfyUI/custom_nodes
git clone https://github.com/mrm987/PeroPixfy.git

No third-party Python dependencies to install. The built web/ is bundled, so it runs without a build step. data/ (presets, gallery, settings) is created on first run.

For development you can link a working folder into custom_nodes with a junction: mklink /J ...\custom_nodes\PeroPixfy <work folder>.

Running

Default — launch ComfyUI with your usual bat; the launcher (web_extension/) adds the sidebar tab. http://127.0.0.1:8188/peropixfy also works directly.
Lean (optional) — scripts\run_peropix.bat: --disable-all-custom-nodes --whitelist-custom-nodes PeroPixfy (+ SageAttention). Spectrum is vendored, so it loads too. The gain is limited (most startup time is LoRA loading).

Build from source

cd ui
npm install
npm run dev     # Vite dev server (proxies to 8188) — ComfyUI must be running
npm run build   # type-check + build web/

Project structure

__init__.py   ComfyUI plugin entry (routes only — no graph nodes)
server/
  routes.py   serves the /peropixfy SPA + gallery/settings API (ComfyUI deps isolated here)
  library/    vendored Style-Manager — /peropixfy/api/library/*
  gallery.py  generations table (generation records)
  migrate.py  one-time import of existing Style-Manager data
ui/           React + TS + Vite source → npm run build → web/
web/          build output (served statically by routes.py)
data/         loras.db, thumbs/, styles/, settings.json (git-ignored)
scripts/      run_peropix.bat (lean launcher)
vendor/       third-party code vendored verbatim (see Credits)

Core design

The single source of truth for generation parameters is GenerationParams (ui/src/workflow/types.ts). The graph is always rebuilt via buildGraph() (ui/src/workflow/builder.ts); a bypassed LoRA is expressed by omitting the node and linking MODEL through directly. Records and reproduction store the params, not the graph.

Credits / third-party

vendor/spectrum/ — ComfyUI-Spectrum-KSampler by sorryhyun (MIT). Vendored verbatim; the upstream LICENSE is kept alongside and re-synced by scripts/update_spectrum.py.

한국어

ComfyUI에서 Anima 모델로 이미지를 만드는 가장 간단한 방법 — 복잡한 노드 그래프 없이, 프롬프트 쓰고 생성만 누르면 됩니다. 로라·스타일을 관리하는 라이브러리도 내장돼 있습니다.

설치

가장 쉬움 — ComfyUI-Manager (모든 OS·모든 설치): Manager에서 PeroPixfy 검색 → Install → ComfyUI 재시작. 유저의 ComfyUI가 쓰는 파이썬을 그대로 쓰므로 Windows 포터블·venv·Linux·macOS·Desktop 앱에서 동일하게 동작합니다.

Windows 포터블 — 원클릭 .bat: peropixfy_install.bat 를 받아 (최신 릴리즈) ComfyUI 폴더(ComfyUI와 python_embeded가 든 폴더)에 두고 더블클릭. (Git for Windows 필요) 그다음 ComfyUI 재시작.

수동 (모든 OS): 위 Manual install(영문) 참고.

첫 실행

ComfyUI 사이드바의 PeroPixfy 탭을 엽니다 — 전체화면으로 전환됩니다(← ComfyUI 또는 ESC로 복귀).

모델이 하나도 없어도 됩니다. 첫 실행 때 Anima 모델을 원클릭으로 받아줘서, 완전히 빈 새 설치에서도 바로 시작할 수 있어요.

할 수 있는 것

싱글 — 한 장씩 작업: 모델 고르고, 프롬프트 쓰고, 다듬고, 디테일 다시 뽑고, 더 선명하게 업스케일.
멀티 — 한 번에 여러 장 생성(변형 목록 × 장수), 마음에 드는 것만 골라 남기기.
라이브러리 — 내 로라·스타일을 둘러보고 정리.

업데이트

ComfyUI-Manager에서 Update 후 재시작. Windows 포터블이면 같은 폴더에 peropixfy_update.bat 을 두고 더블클릭한 뒤 재시작해도 됩니다. (설치 bat 재실행도 업데이트됨.)

소스 빌드·프로젝트 구조·내부 동작 등 기술 내용은 위 Development 섹션(영문) 참고.