2026 Data Visualization Delivery Decision Matrix: Matplotlib vs Plotly PNG Export — DPI, sRGB & Remote M4 Batch Acceptance

① Decision matrix: Matplotlib vs Plotly for PNG delivery

Pick the library that matches your governance model. Matplotlib rewards teams that want deterministic, code-reviewed styling in Python only. Plotly fits exploratory dashboards that already live as Figure objects and need identical PNGs from the same JSON without re-implementing theme tokens twice.

② Executable export parameters (copy-paste)

These snippets assume a Linux or macOS worker; on a remote Mac M4 node, run them under the same user as your LaunchAgent or CI shell so font caches stay warm.

Matplotlib — Agg backend, dpi, transparency

# Shell: force non-interactive backend for batch jobs
export MPLBACKEND=Agg

# Python
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt

fig, ax = plt.subplots(figsize=(10, 5.625), layout="constrained")  # ~16:9 canvas in inches
ax.plot([0, 1], [0, 1])

fig.savefig(
    "out/matplotlib_chart.png",
    dpi=300,
    transparent=True,
    facecolor="none",
    edgecolor="none",
    bbox_inches="tight",
    pad_inches=0.05,
    metadata={"Software": "None"},  # optional: strip default generator strings per policy
)

• dpi: 144–300 for deck graphics; 300+ when text must survive print PDF downscaling.
• figsize: treat inches as a contract—pair with one documented dpi tier per brand guideline.
• transparent + facecolor="none": avoids accidental white matte behind legends.
• Alternative backend: mplcairo where sub-pixel text rendering matters; still validate on target viewers.

Plotly — Kaleido write_image (PNG)

Plotly’s Figure uses write_image for raster output; there is no separate write_png on the figure object—pass format="png" (or omit it when the filename ends in .png) so the Kaleido path matches your CI expectations.

import plotly.graph_objects as go

fig = go.Figure(data=go.Scatter(x=[0, 1], y=[0, 1]))
fig.update_layout(
    template="plotly_white",
    paper_bgcolor="rgba(0,0,0,0)",
    plot_bgcolor="rgba(0,0,0,0)",
)

fig.write_image(
    "out/plotly_chart.png",
    format="png",
    width=1920,
    height=1080,
    scale=2,          # logical 1920×1080 → file pixels 3840×2160 unless engine defaults differ
    engine="kaleido",
)

• width / height: set explicit pixel boxes for slide masters (1920×1080, 3840×2160, etc.).
• scale: use 1 for true-1x web assets; 2 for Retina when your downstream expects doubled raster.
• engine="kaleido": pin kaleido and plotly versions in requirements.lock—upgrades have changed raster bounds.

Optional Pillow bridge (Image.save / PNG codec)

When you composite Matplotlib output with logos, use Pillow for the final save(..., format="PNG", compress_level=6) only after you have validated RGBA straight-alpha behavior. Pillow’s low-level PngImagePlugin path is rarely needed directly; prefer img.save("out.png") with explicit mode RGBA when overlays demand it.

③ sRGB, ICC, and acceptance alignment

Neither Matplotlib nor Plotly magically embeds a “design system ICC contract” unless your pipeline adds it. For web and deck delivery, standardize on sRGB as the spoken language between data science and design ops, then prove it the same way you prove Figma exports: preview plates, profile policy, and manifests. Deep dives live in the Mac PNG sRGB vs Display P3 checklist and ICC metadata & lossless recompress article; here is the data-viz-specific minimum.

④ Remote Mac mini M4: batch path & acceptance checklist

Offload long-running chart batches to a dedicated Apple Silicon host so laptop sleep, VPN jitter, and local font drift do not corrupt nightly exports. Mirror the folder discipline you already use for design PNGs:

• ~/png_jobs/viz_export/inbox/ — writer process drops raw PNGs.
• ~/png_jobs/viz_export/work/ — optional compositing / watermark.
• ~/png_jobs/viz_export/out/YYYY-MM-DD/ — promoted assets + manifest.
• ~/png_jobs/viz_export/log/export.jsonl — one JSON object per file with tool, version, dpi or scale, sha256.

For very large canvases or 4K marketing frames, reuse the throughput ideas in batch 4K PNG on remote Mac M4. For automated QA hooks (dimensions, byte caps, retries), cross-read OpenClaw PNG QA batch check patterns even if you stay in plain shell first.

⑤ Summary

Matplotlib’s Agg + savefig path and Plotly’s Kaleido + write_image path both deliver excellent PNGs when you freeze backends, pin versions, and express size as either inch × dpi or pixel × scale—never both ambiguously. Pair exports with a small sRGB-aligned QA plate, transparent-alpha checks, and a dated out/ tree on a remote Mac M4 so data viz handoffs inherit the same rigor as marketing PNG pipelines.

Next steps: rent a remote Mac M4 for batch chart PNGs

Browse the MacPng homepage for scenario context, then open rental and purchase options and nodes & pricing—no login is required to compare packages. After you pick a node, follow the SSH / VNC setup guide, copy the folder layout above, and schedule Matplotlib or Plotly jobs overnight while your laptop stays closed. More design-delivery matrices live in the Tech Insights index.