I couldn't find a decent image viewer for ComfyUI outputs that showed the metadata readably — so I built one. Then built a wedge/sweep node to go with it.

MBQ Viewer reads the prompt data embedded in every ComfyUI PNG and displays it in a clean, colour-coded panel right alongside the image — models, prompts, sampler params, all readable without digging into JSON. Works on any PNG saved by ComfyUI's SaveImage node. ComfyUI doesn't need to be running.

MBQ Wedge is a companion ComfyUI node that sweeps any numeric parameter across a range — steps, CFG, denoise, anything — queuing one image per value from a single click. Each PNG gets the swept value embedded, so the viewer labels every image automatically. Lock your zoom to a crop and flip through the whole sweep at pixel level.

The loop: get a result you like in ComfyUI → freeze the seed → add MBQ Wedge and sweep the parameter you're uncertain about → open the output folder in MBQ Viewer → flip through results with the swept value labelled on every image, zoomed in to the exact detail you care about.

Say you like a scene but aren't sure how many steps you need: sweep steps 4 → 12. Want to tune an inpaint blend? Sweep denoise 0.4 → 1.0 at 0.05 increments. Works with any float or int input on any node.

MBQ Wedge — ComfyUI node

Install

Copy (or symlink) the ComfyUI-MBQWedge/ folder into your ComfyUI custom nodes directory and restart ComfyUI:

ComfyUI/custom_nodes/ComfyUI-MBQWedge/

The node appears under MBQ → MBQ Wedge.

How it works

Add MBQ Wedge to any workflow and wire one output to a numeric input:

Output	Connect to
`int_value`	Any INT input (steps, width, …)
`float_value`	Any FLOAT input (CFG, guidance, denoise, …)

Set start, stop, and increment. Click Queue once — the MBQ JS extension intercepts the submission and expands it into N separate jobs, one per sweep value.

Each output PNG has the exact swept value embedded in its ComfyUI prompt chunk. MBQ Viewer reads it directly — no guessing, no positional inference. Works correctly across multiple batches and renamed files.

Seeds advance correctly across sweep jobs: randomize, increment, decrement, and fixed modes all work as configured.

Node display

will produce → N iterations — live count updates as you adjust the range
current → X — shows the starting value; switches to integer display when wired to an INT input

Example — sweep steps 4 → 12

parameter_name = "steps"   (auto-filled when you connect the output)
start          = 4
stop           = 12
increment      = 1

Wire int_value → BasicScheduler steps. Click Queue → 9 jobs, 9 PNGs, each labelled steps: 4 through steps: 12 in MBQ Viewer.

MBQ Viewer — desktop viewer

Install

Option A — standalone exe (Windows)

Download MBQViewer.exe from the Releases page. No install, no Python required — just run it.

Windows SmartScreen: MBQ Viewer is not yet code-signed, so Windows will show a "Windows protected your PC" dialog on first launch. Click More info → Run anyway. This is normal for unsigned indie software.

Option B — run from source (Windows / Linux / Mac)

Requirements: Python 3.11+, PySide6

pip install pyside6
python mbq_viewer.py

python mbq_viewer.py path/to/image.png   # open image + its folder
python mbq_viewer.py path/to/folder/     # open folder

Drag images in from Explorer or ComfyUI. Drag thumbnails out to Explorer or ComfyUI.

Canvas

OpenGL-accelerated — smooth pan (left-drag), scroll-wheel zoom anchored at cursor
Middle-click to reset zoom to 100%
Zoom Lock (Z) — preserves the exact pan and zoom position when flipping between images, so you can inspect the same crop across an entire sweep
Fit Lock (F) — auto-fits every image to the window
Reset (R) — 100% zoom, clears both locks

Workflow panel

Reads the prompt JSON embedded by ComfyUI's SaveImage node and displays it in a colour-coded panel:

Tier 1 — models, prompts, sampler params (the things you usually care about)
Tier 2 — other scalar values
Tier 3 — plumbing nodes (collapsible, hidden by default)

Colour coding: cyan node headers · yellow keys · near-white values

Copy Summary — plain-text summary for notes or sharing
Copy Workflow — raw LiteGraph JSON for direct re-import into ComfyUI

Scroll Freeze (S) — locks the workflow panel's scroll position when flipping between images, so the same section of the metadata stays in view across an entire sweep.

MBQ Wedge integration

When a swept image is loaded, MBQ Viewer:

Lights the Wedge bulb (blue) in the status bar
Overlays the swept parameter and value on the canvas corner: steps: 6
Highlights the swept value in the Workflow panel's MBQWedge entry

The overlay is fixed to the canvas corner — unaffected by zoom or pan.

Keyboard shortcuts

Key	Action
`Ctrl+O`	Open image
`← / →`	Previous / next image
`F5`	Refresh folder
`Z`	Toggle zoom lock
`F`	Toggle fit lock
`R`	Reset zoom to 100%, clear locks
`S`	Toggle scroll freeze
`Ctrl+Q`	Quit

Status bulbs

Bulb	Colour	Meaning
Wedge	Blue	MBQWedge sweep detected in this image
Zoom	Green	Zoom Lock active
Fit	Teal	Fit Lock active
Scroll	Green	Scroll Freeze active
UnComfy	Amber	File may be a preview save, not a final output

File map

File	Role
`mbq_viewer.py`	Main window — layout, menus, navigation, metadata display, filmstrip
`mbq_functions.py`	`ImageCanvas` (OpenGL view), `ImageFolder`, `WorkflowCache`
`mbq_parser.py`	PNG chunk extraction, ComfyUI prompt graph parsing, three-tier node classification
`ComfyUI-MBQWedge/mbq_wedge.py`	MBQWedge ComfyUI node
`ComfyUI-MBQWedge/js/mbq_wedge.js`	JS extension: widgets, queue intercept, seed handling

Notes

ComfyUI only. MBQ reads the prompt chunk written by ComfyUI's SaveImage node. A1111, NovelAI, InvokeAI, and other tools are not supported.
Soft-fail. If a PNG has no prompt chunk, MBQ shows the image and leaves the workflow panel blank — it never crashes or shows garbage.
No ComfyUI runtime dependency. MBQ Viewer reads PNG files directly; ComfyUI does not need to be running.
Cross-platform. Developed on Windows; should work on Linux. Mac support is untested.

github.com/Beakfx/mbq