Lab audit · Prior art + stolen vectors · 3 Nov 2026 · Index-level research

MIC vs
The World

You asked: is what we built already a thing? Then strip a decade of Fortune-500-grade compression R&D down to its vectors — and fire those vectors back at MIC. This is the whole answer, with receipts.

No single prior system combines all six MIC properties. The package looks novel. Almost every piece, on its own, does not.

The full combination — human-readable plain-text encoding, progressive semantic / edge / palette / pointer layers, decoder-free reconstruction, stroke-coherence filtering, a queryable scene-graph layer, and 50–80:1 vs JPEG — was not found anywhere else. Individual properties have strong precedents, and a handful of papers combine two or three of them. All findings below are index-level (search results + fetched page text, read 2026-11-03), not live-verified in an interactive browser.

RESEARCH DATE: 2026-11-03·28 VECTORS SURVEYED·12 CANDIDATES MAPPED·10 OPEN OPPORTUNITIES
01

What MIC actually is

L0 — Semanticscene description in text
L1 — Wireframe1-bit edges, base36 RLE
L2 — Color fillsmedian-cut palette, RLE
L3 — Pointerlink to the original file

Plus L0G, the structured scene-graph variant: objects with bounding boxes, colors with coverage %, exact text transcriptions, and face counts — queryable without decoding a single pixel. Decoding needs no neural network and no special software: parse text, draw pixels.

02

The six-property scorecard

(a)

Image as human-readable plain text — no binary

Not anticipated

No prior art encodes photos as human-readable, semantically structured plain text intended as a compression codec. SVG tracing (ImageTracer.js, Potrace, Vector Magic, StarVector) outputs readable text, but traced photos typically compress worse than JPEG and aren't photo codecs. Base85 containers (AEON SAPI, ZipText/ZTAR) are text but opaque alphabet soup — and still need a full image decoder.

Closest: StarVector (arXiv:2312.11556, 2024) — a CodeLLM that writes SVG code as text tokens, but generative, not a codec. Codewars ASCII-art codec encodes ASCII art, not photos.
(b)

Progressive layers: L0 / L1 / L2 / L3

Partially anticipated

Text + edge two-layer semantic compression exists. Edge + low-res + residual layering is classic (1988–2012). But the specific four-layer text stack — semantic text, RLE text edge map, RLE text palette fills, original-file pointer — was not found.

Closest: “Text + Sketch: Image Compression at Ultra Low Rates” / PICS (arXiv:2307.01944, Jul 2023) — text + HED edge map, decoded via ControlNet / Stable Diffusion. Canon patent US 8,218,908 (2012). AGDL (Zhang & Wu, 2021). JPEG XL DC frames; SlimCAE (CVPR 2021).
(c)

Decoder-free — no NN, no special software

Not anticipated*

Not anticipated for photo-quality reconstruction. Trivially-decodable text image codecs exist only for toy / ASCII-art domains. All semantic-compression academic work (textual transform coding, PICS, PerCo, Cross-Modal Compression) requires large generative models as decoders. SVG still needs a renderer; base85 still needs decoders plus the image codec.

*The Codewars ASCII quadtree codec needs only a trivial parser — decoder-free in spirit, but for ASCII art, not photos.
(d)

Coherence: keep long strokes, drop speckle

Half anticipated

The negative half — drop speckle — is standard and 20+ years old: Potrace and ImageTracer.js despeckle via a “turd size” threshold, deleting small connected components as noise. JPEG XL patches and noise synthesis make the same move: don't spend bits on noise. The positive half — explicitly keeping long connected strokes because they carry text / outline semantics — was not found as a named mechanism outside MIC.

Also: PICS found HED edge maps beat Canny on rate-distortion — HED suppresses texture noise, i.e. an edge-coherence selection.
(e)

L0G scene graph — queryable without decoding

Partially anticipated

Textual queryability is an active research theme and per-object attribute graphs exist — but a schema-fixed scene-graph layer (bboxes + color coverage % + transcriptions + face counts) inside an image codec was not found.

Closest: LOST-3DSG (arXiv:2601.02905) — per-object label, 3D bbox, color, material, description; ~157 bytes/object, ~3 KB/scene, queryable — but for 3D robotics, built with neural encoders. Weissman textual transform coding (2023; Phil. Trans. R. Soc. A 2025). TextBoost (arXiv:2603.04115) validates exact transcriptions as side info.
(f)

50–80:1 vs JPEG, text-legible wireframe

Not anticipated as claimed

No source found claims a classical edge + palette codec reaching 50–80:1 over JPEG on photos. Generative codecs exceed the ratio by orders of magnitude — PerCo at 0.003 bpp (≈8000:1 vs raw), PICS text+sketch at ≈0.013 bpp — but with hallucinated pixels and neural decoders, not wireframe-legible reconstructions. For reference, JPEG XL and AVIF achieve only ~2–2.5:1 over JPEG at equal quality.

The report flags this plainly: MIC's claimed ratio sits in generative-compression territory, achieved here by semantic rather than generative means — see Open Questions below.
03

Head-to-head: every candidate, every property

✓ = does it  ·  ◐ = partial / analog  ·  ✗ = does not do it

Candidate(a) Text(b) Layers(c) No decoder(d) Coherence(e) Scene graph(f) Ratio
Text+Sketch / PICS (arXiv:2307.01944)✗◐✗◐✗✗
Weissman textual transform coding (2023/2025)◐✗✗✗◐✗
Cross-Modal Compression ITI (2022)◐✗✗✗✗✗
AEON SAPI / ZipText (2026)◐✗✗✗✗✗
Codewars ASCII quadtree codec (1980s alg.)✓◐✓✗✗✗
Canon US 8,218,908 / sketch coding 1988✗◐✗✗✗✗
AGDL saliency sketch (2021)✗◐✗◐✗✗
Potrace / ImageTracer (turd-size despeckle)◐✗◐◐✗✗
StarVector (2024, LLM→SVG)✓✗✗✗✗✗
LOST-3DSG scene graph (2026)◐✗✗✗◐n/a
TextBoost OCR side info (2026)◐✗✗✗◐supports f
Sixel / Kitty terminal graphics✗✗n/a✗✗✗

Read the table the way the lab should: nobody gets past three partials. MIC's novelty isn't any single trick — it's stacking all six in one text file.

04

The vectors — 28 innovations, stripped to the principle

This is the reverse-engineered R&D pipeline: every major compression breakthrough surveyed (roughly 2015–2026, plus the foundational mechanisms they build on), reduced to the one mechanism that makes it work. Search it, filter it, steal from it.

05

The lab playbook — 10 vectors MIC hasn't fired yet

Established, high-leverage mechanisms from the survey that MIC's current design does not exploit. The report's highest-expected-value picks are marked — each attacks a dominant cost center (letterforms, speckle, RLE inefficiency) without requiring a neural decoder, so property (c) survives.

06

Open questions — read before you brag

What the research could not verify

  • No live testing of MIC itself was possible — no code, dataset, or encoded samples were provided to the researcher. The 50–80:1 claim and wireframe legibility are unverified vendor claims. The ratio sits in the regime currently occupied only by neural / generative codecs, which is a yellow flag for a classical method — but edge + palette representations of text-heavy or graphic photos can plausibly reach it; photographic texture is the risk case.
  • “AEON SAPI” (Zenodo, Jan 2026) has the trappings of a low-credibility self-published paper (grandiose claims, “emergent AI capability” framing); treated only as evidence that base85 + zlib text containers exist, not as technical prior art.
  • WebP2: Google's experimental successor (VP8-intra lineage) — confirmed it exists and its technical basis, but its standardization / deployment status remains murky; not load-bearing for the report.
  • Median-cut advances: searched, but no significant post-2015 advance over Heckbert (1982) was found as the standard fast palette method; noted as baseline.
  • Cheng et al. 2020 attention codec: corroborated via the NeurIPS 2023 supplement comparing ELIC against “Cheng's hyperprior” (+28.85% BD-rate gain for ELIC); not independently deep-dived.
  • All findings are index-level. A production prior-art opinion — especially patent clearance around US 8,218,908 and JBIG2 symbol-coding patents — would need a proper patent search, which was out of scope.
07

Sources

Text + Sketch / PICS — arxiv.org/html/2307.01944
Weissman, “Toward Textual Transform Coding” — arxiv.org/pdf/2305.01857
Weissman et al., Phil. Trans. R. Soc. A — ar5iv.labs.arxiv.org/html/2506.15948
Cross Modal Compression (ITI) — arxiv.org/pdf/2209.02574
PerCo (ICLR 2024) — storage.googleapis.com (PerCo PDF) · PerCo (SD) — arxiv.org/pdf/2409.20255
TextBoost — s.bingheai.cn/pdf/2603.04115
AGDL saliency sketch — arxiv.myfastools.com/pdf/2103.15368.pdf
Canon US 8,218,908 — patents.justia.com/patent/8218908
AEON SAPI — zenodo.org/records/18360858
ZipText — pypi.org/project/ziptext · Codewars ASCII codec — images.codewars.com
LOST-3DSG — arxiv.org/html/2601.02905 · StarVector — arxiv.org/html/2312.11556v1
SlimCAE — arxiv.org/pdf/2103.15726 · ELIC — alphaxiv.org/abs/2203.10886v2
ANS / rANS — en.wikipedia.org/wiki/Asymmetric_numeral_systems
Full source list (30+ entries, all read 2026-11-03, index-level) is carried in the research report behind this page.