ComfyUI_Spectrumify

A custom ComfyUI node that converts any input image into pixel-art using the classic ZX Spectrum palette. The node is a clean-room reimplementation of the conversion pipeline from spectrumify-public by JGF, adapted to the ComfyUI tensor convention.

Three widths (SMALL 80px, MEDIUM 153px, BIG 256px), three colour modes (16 Colors, Grayscale, Black & White) and a pixel_scale slider give you the same nine presentation modes that spectrumify ships with, plus an inline preview and optional GIF export.

Screenshot

Installation

ComfyUI Manager

1. Open ComfyUI Manager.
2. Search for Spectrumify and click Install.
3. Restart ComfyUI.

Manual

cd ComfyUI/custom_nodes
git clone https://github.com/JoseGomezFreelance/ComfyUI_Spectrumify_A-public.git
cd ComfyUI_Spectrumify_A-public
pip install -r requirements.txt

Restart ComfyUI. The node appears under image/transform → Spectrumify (ZX).

Usage

Connect any IMAGE output to the image input and configure:

Widget	Values	Description
size	SMALL (80px), MEDIUM (153px), BIG (256px)	Downscaled width before quantisation. Height is computed preserving aspect.
color_mode	16 Colors, Grayscale, Black & White	Palette used for nearest-neighbour quantisation.
pixel_scale	1..16	Integer upscale factor (nearest-neighbour) applied after quantisation.
save_gif*	skip / save	When save, writes an animated GIF (one frame per image in the batch) to the ComfyUI output/ folder.
filename_prefix*	arbitrary text	Prefix for preview and GIF filenames. Non-alphanumeric characters are stripped.

*optional inputs — accessible via the node's widget panel.

The output spectrumified is a standard IMAGE tensor (B, H, W, 3 float32 in [0, 1]), ready for the rest of your workflow (e.g. saving, VAE encoding, upscalers). The node is an OUTPUT_NODE so ComfyUI displays the result inline.

Development

python -m venv .venv
source .venv/bin/activate
pip install -r requirements-dev.txt
pytest --cov=spectrumify --cov=nodes --cov-report=term-missing

The test suite does not depend on ComfyUI or torch: when torch is missing, the tensor helpers transparently fall back to numpy arrays, so the pipeline can be exercised in isolation.

Credits

• Original conversion logic, palette and algorithm design: JGF (José Gómez Freelance) — spectrumify-public.
• ZX Spectrum hardware palette © Sinclair Research, 1982 (public reference).

Speedrun

This node was written in speedrun mode. Total wall-clock time: 20 min 40 s, in a single session of Claude Code (Opus 1M Max) piloted by JGF.

Milestone	Time
Full package generation (code + tests + README + LICENSE)	~10 min 43 s
Dependency security & reputation analysis (docs/)	~between 10:43 and 12:00
Dependency install into .venv + first pytest run	included
Debugging 14 failures (BW shape, palette expectations, importlib)	included
144 tests passing, 98.18% coverage	20 min 40 s

Internal technical report estimate: 2-4 hours. Actual result: ~1/6 of that budget.

By the way, writing the original internal report took 1 hour.

License

Apache 2.0 — same as the upstream spectrumify-public project. See LICENSE for the full text.