Voynich Text and Image: How the Page Was Planned

Some manuscripts put text in one place and pictures in another.

The Voynich Manuscript often refuses that separation.

A plant stem rises through the page.

The words bend around it.

A branch leaves a narrow gap.

A token fits inside the gap.

A human figure stands beside a short isolated string.

A circular diagram pulls writing into a ring.

A radial structure makes the text point toward or away from a centre.

The right margin remains irregular because illustration geometry decides how much line width is available.

This is not a manuscript in which the image can be deleted cleanly and the text reflowed into an ordinary column without changing the original design.

The image and text share one surface.

They negotiate space.

That physical fact survives before we know what any sentence says.

It also creates one of the most important interpretive disputes in Voynich research.

If text fits the spaces left by drawings unusually well, perhaps the writer simply generated Voynich-looking strings to fill whatever room remained.

Or perhaps the page was planned in advance.

Perhaps a meaningful text was copied from a draft whose image and writing were already coordinated.

Perhaps a flexible shorthand system allowed the scribe to compress or expand meaningful material.

Perhaps the writer selected among equivalent forms according to available width.

The physical observation is real.

The mechanism is not yet unique.

Spatial fit proves coordination. It does not, by itself, tell us whether the coordinated text is meaningful, encoded, abbreviated or generated.

That is the page-planning problem.

Quick Read

One-sentence answer: Voynich pages were deliberately composed as shared text-image spaces: illustration outlines generally preceded writing, prose was fitted around pre-existing drawings, labels were anchored spatially to visual elements, and circular or radial diagrams imposed their own text geometry; these facts demonstrate page planning but do not establish whether the writing is literal description, meaningful shorthand, encoded text or generated filler.

Illustration outlines generally precede the principal text on illustrated pages.
Text repeatedly avoids stems, leaves, containers, bodies and diagram structures.
Some words fit closely into narrow spaces left by branches or drawing contours.
This means available writing width was often determined by image geometry.
Irregular right margins are therefore partly a consequence of the visual layout rather than free prose columns.
There is little evidence for a rigid ruled text grid governing the whole page before illustration.
Short labels occupy a different spatial job from running text and are often anchored near specific visual elements.
Image adjacency indicates a relation, but not necessarily literal naming.
Circular and radial writing should be treated as geometry-specific text roles rather than flattened automatically into normal prose lines.
Foldout pages extend the same planning logic across larger surfaces; their detailed physical operation belongs to the separate Foldouts article.
A generated-filler model can explain tight spatial fit efficiently.
A preplanned draft, copied exemplar, flexible orthography or abbreviation system can explain it too.
Meaningful text can be spatially adaptive.
Meaningless text can be spatially adaptive.
The decisive evidence must come from whether the same spatial behaviour predicts independent linguistic, statistical or visual relationships under fixed rules.

The page is therefore not merely a container for two information types.

It is itself part of the information system.

The Drawing Was Usually There First

The Order of Making article established the local production sequence.

On many illustrated pages, the principal outline precedes the text.

The evidence is behavioural.

Text stops before a drawn element.

Resumes after it.

Fits into open areas.

The simplest explanation is that the scribe could already see the drawing.

This changes the modern instinct to imagine:

write the text, then illustrate it.

Voynich often looks more like:

prepare the visual field, then place text inside the remaining geometry.

That is a strong production observation.

It still does not tell us whether the drawing was invented first, copied from an exemplar, or transferred from a draft that already coordinated text and image conceptually.

The Text Behaves as Though Available Width Matters

A normal paragraph in a modern book flows inside a rectangular column.

The text engine knows the width.

Lines wrap automatically.

Voynich is handwritten, but many illustrated pages create a comparable local constraint.

The available width changes from line to line because the drawing intrudes.

A plant leaf narrows one line.

A stem splits another.

The next line opens wider.

This makes line length partly a geometric variable.

That matters because Voynich also has line-position effects.

If line endings change token form, and drawings change where line endings occur, then image geometry can indirectly change text statistics.

The page-planning problem therefore connects visual layout to entropy, morphology and line architecture.

A purely textual corpus can lose that causal variable.

Some Words Fit Suspiciously Well

The most provocative page-planning observations involve tokens that seem to fit the exact width available between visual obstacles.

A small gap appears.

A short token appears.

A larger gap appears.

A longer token appears.

This is the strongest intuitive support for filler-generation theories.

If the writer can manufacture arbitrary valid-looking Voynich words of different lengths, any gap can be filled elegantly.

But the same observation is compatible with several meaningful mechanisms.

Choose among synonymous or equivalent technical forms.
Use abbreviation when space is tight.
Use fuller spelling when space is generous.
Copy a preplanned draft in which word and gap already correspond.
Adjust spacing between glyphs or tokens.
Continue the sentence on the next line where necessary.

Therefore a visual fit should not be used as a verdict.

It should become a measurable variable.

Does available geometric width predict token length more strongly than linguistic context does?

That is the kind of question that can separate mechanism families.

Generated Filler Is a Real Hypothesis, Not a Default Explanation

A local-generation process can fit pages beautifully.

Choose a legal token family.

Select a length compatible with the available gap.

Modify a nearby token if needed.

Continue until the page is full.

This naturally creates tight visual integration.

It can also contribute to local word families and line-bounded repetition.

The model therefore deserves serious consideration.

But once it claims the entire page-planning phenomenon, it inherits several burdens.

Why do labels form a separate textual register?
Why do Currier regimes exist?
Why do long-range vocabulary domains align partly with visual sections?
Why do the same token families recur beyond one local gap?
Why does the manuscript preserve systematic edge-to-edge dependencies?
Why do physical production units correlate with textual state?

Generated filler is strongest at explaining local spatial fit.

It still has to explain the rest of the information architecture.

A Prepared Draft Can Produce the Same Fit

Suppose the surviving parchment was not where composition first happened.

A draft or exemplar may already have coordinated image and text.

The Voynich maker copies the drawing outline.

Then copies the text into the corresponding spaces.

Now word length fits the drawing because the fit was solved earlier.

The surviving page tells us nothing about the invisible draft directly.

But the hypothesis is historically ordinary.

Manuscripts are copied from exemplars.

Illustrations can be transmitted.

Text and image can be planned together before final execution.

This makes “prepared draft” an important negative control for filler theories.

Tight fit cannot distinguish the two without additional evidence.

Flexible Orthography Can Make Meaningful Text Fit Variable Gaps

Meaningful writing is not always rigid in visible length.

Historical scribes can abbreviate.

Choose alternate spellings.

Use ligatures.

Compress common endings.

Write more tightly.

Expand a flourish when space remains.

The Abbreviation or Alphabet article established that such flexibility is historically plausible but must remain rule-governed.

This creates a direct page-planning prediction.

If abbreviation is being used to fit space, abbreviated variants should become more common in constrained geometric regions.

If fuller variants appear where more room is available, the spatial relationship strengthens.

Meaningful compression can therefore mimic a generator’s length control.

The difference is whether the variants preserve consistent underlying linguistic relationships.

The Rough Right Margin Is Partly Designed by the Picture

Voynich pages do not generally present a formal rectangular text block with a uniform right margin.

On illustrated pages, the right edge can be defined locally by:

a leaf;
a plant stem;
a vessel;
a human figure;
a circular diagram;
the edge of the page itself.

This means “line length” is not one stable page-wide constant.

Any study of line-final behaviour should therefore know the amount of usable space rather than merely the character count of the line.

A short line caused by a plant branch is physically different from a short line chosen voluntarily in an open text area.

If line-final variants are purely margin-management devices, the two should behave similarly when remaining width is equally constrained.

If the variants are semantic, geometry alone should not explain them.

Page layout can therefore help adjudicate the line-effect debate.

Labels Occupy a Different Spatial Contract

Running text fills regions.

Labels attach.

That spatial difference is important before semantics.

A short isolated Voynich string beside one figure, star or plant fragment is not behaving like a paragraph.

Its placement establishes a local relation.

But relation type remains open.

Name.
Category.
Number.
Property.
Direction.
Cross-reference.
Code.

This is why the Labels Versus Running Text article found a distinct textual population.

Page planning predicts such a distinction naturally.

The writer is doing a different spatial job.

The writing system changes its distribution accordingly.

Spatial role and textual register reinforce one another.

A Label Does Not Automatically Name the Thing It Touches

Spatial anchoring is powerful.

It is also dangerous.

Place one unknown word beside a star and the reader wants the word to mean the star’s name.

Place another beside a root and the reader wants a plant name.

But diagrams routinely label relationships other than identity.

A label can indicate:

sequence;
class;
quantity;
ownership;
state;
connection;
instruction.

Therefore page planning gives us one safe inference:

this text was positioned in relation to this visual element.

The semantic verb—names, counts, describes, classifies, points to—remains to be discovered.

Circular Text Has a Different Geometry of Reading

Linear prose has a left edge and a right edge.

Circular text does not.

A ring has no obvious first token unless the manuscript marks one.

This means page planning can define sequence through geometry rather than ordinary line flow.

A modern transcription has to cut the circle somewhere.

That cut creates a first token in the dataset.

The manuscript may not have had the same beginning.

Therefore circular text should preserve:

centre;
radius;
direction;
sector;
physical start markers if any.

The geometry is not decorative metadata.

It may be part of the ordering system.

Radial Text Can Encode Direction Without Ordinary Syntax

Radial text introduces another possibility.

A string points inward.

Or outward.

It may connect a centre to a rim.

Its neighbouring strings may be defined by angle rather than by line adjacency.

This creates a diagrammatic syntax.

Not grammar in the ordinary linguistic sense, but rules about where text belongs relative to visual structure.

If a radial label is flattened into a normal horizontal line, its relationship to centre, sector and neighbour can disappear.

This is another reason a text-only corpus is an incomplete representation of the manuscript.

The page layout contains relational variables the transcription must preserve separately.

Image Geometry Can Create Apparent Line Effects

The Line as a Unit article established that token distributions change at line beginnings and endings.

Page planning adds a confounder.

Illustrations determine where many lines begin and end.

Suppose gallows-rich openings occur after a plant stem forces a new line.

Is the gallows responding to a linguistic boundary?

A scribal new-line convention?

Or a visual-layout reset?

These can coincide.

A page-aware analysis should therefore ask whether line effects differ between:

lines ending naturally at the page margin;
lines shortened by illustrations;
lines beginning after visual intrusion;
text-only pages.

If the effect follows geometry, layout contributes strongly.

If it remains stable across these conditions, the writing system carries a deeper positional rule.

Image Geometry Can Also Create Apparent Word-Length Effects

Suppose short words cluster in narrow spaces.

That can be explained by selection pressure from page geometry.

But one must distinguish several mechanisms.

The writer chooses a naturally short word with suitable meaning.
The writer abbreviates a longer word.
The writer selects among encoded variants.
The writer generates a short meaningless token.
The scribe adjusts character spacing to fit.

All predict some relationship between width and token length.

They differ in what happens to lexical identity, morphology and neighbouring syntax.

This is why the spatial variable should be combined with text structure rather than used alone.

Geometry can constrain the candidate mechanism.

It cannot identify meaning by itself.

A Meaningful Text Model Makes Page-Planning Predictions

If the text carries language or technical information, spatial fit should not erase semantic consistency.

Words selected for narrow gaps should still behave like members of meaningful families.

Abbreviated forms should expand consistently.

Labels should correlate with visual or categorical roles under fixed rules.

Circular and radial text should preserve coherent syntax or field order appropriate to the geometry.

Line-end variants should have a principled relationship to the underlying language or notation.

The spatial adaptation becomes another layer of orthography or document design.

It should not destroy the recovered source system.

This gives a meaningful-text theory a demanding but fair test.

Can one stable underlying information system survive the page’s changing geometry?

A Generated-Text Model Makes Different Predictions

If local geometry directly drives token generation, available space should be a powerful predictor of visible form.

Token length should respond strongly.

Perhaps family selection should too.

Tokens in narrow gaps may be generated from a constrained short-form inventory.

Line-final families may reflect the amount of remaining room.

In the strongest geometric-generation model, much of the local text can be predicted from:

available width;
previous local token;
line position;
page regime.

That is a testable claim.

If geometry explains token choices extremely well after other variables are controlled, generation strengthens.

If lexical/topic variables remain strong and geometry only adjusts surface length, meaningful text becomes more plausible.

The page can help discriminate without needing to be translated first.

An Encoding Model Can Use Geometry Too

Ciphertext is often imagined as one fixed output length for one plaintext unit.

Historical hand ciphers need not behave that way.

Homophones can offer several visible alternatives.

Verbose groups can vary in length.

Nulls or fillers can be inserted or omitted.

A scribe could therefore choose among equivalent encoded forms according to available space.

This creates another route from meaningful underlying text to visually adaptive surface writing.

But a cipher model must specify the choice rule.

If one plaintext word has three possible ciphertext forms, why was this one chosen here?

Does width predict the choice?

Can the same rule be reproduced on unseen text?

Geometry can become part of an encoding hypothesis only when the selection process is explicit.

A Copied Exemplar Can Preserve Spatial Planning

There is one more important alternative.

Perhaps the Voynich page is not being composed from scratch.

The maker may copy a source whose images and text already have established relationships.

The outline is copied first.

The corresponding text is then copied into the same relative positions.

This naturally produces repeated page-planning conventions across a codex.

It also explains why the surviving manuscript can look highly deliberate even if the person producing it did not invent the underlying content.

Comparator manuscripts such as Masson and Sloane show that image-dominant knowledge can be copied, rearranged and re-hosted across manuscript traditions.

That establishes historical possibility.

It does not prove a specific Voynich exemplar ever existed.

The exemplar hypothesis remains a production alternative that tight spatial fit alone cannot eliminate.

Text-Only Pages Are a Critical Control

If images drive the unusual layout, pages with little or no illustration become valuable controls.

Quire 13 contains at least one text-dominant page.

Other manuscript regions provide long stretches of more ordinary horizontal writing.

Compare them with heavily illustrated pages.

Do line lengths become more regular?

Do line-final forms change?

Do token lengths shift?

Does the q-series frequency change only because of Currier regime, or also because of layout?

Controls without visual obstacles can help isolate geometry from language state.

This is stronger than assuming every unusual property belongs to Voynichese itself.

Page Planning Can Help Separate Production From Topic

Different visual sections also use different page geometries.

Herbal pages place large plants beside prose.

Zodiac pages organise labels around circles.

Quire 13 uses figures, pools and connectors.

Pharmaceutical pages combine fragments, vessels, labels and prose.

Quire 20 uses repeated star-marked records.

Text statistics differ across these regions.

Some differences may reflect topic.

Some Currier regime.

Some scribal hand.

Some document role.

And some simply page geometry.

A robust semantic analysis should therefore control for how the page was physically organised.

Otherwise layout can masquerade as vocabulary or grammar.

Current Binding Order Can Distort Page-Planning Interpretation

Page planning happens on physical sheets.

Reading order happens after sheets are stacked and bound.

If the current order is not the originally intended order, adjacent modern pages can falsely appear to form one designed sequence.

Conversely, halves of one integrated bifolium design can be separated by current collation.

This is why page planning should be reconstructed first within:

the individual surface;
the bifolium;
the foldout sheet.

Only afterward should current neighbouring pages be treated as a larger sequence.

The physical maker’s relationships are safer than the later binder’s relationships.

Foldout Geometry Extends the Same Principle

On a foldout, page planning can exceed one ordinary folio entirely.

A diagram crosses multiple panels.

Text may cross a crease.

Some relationships appear only when the whole sheet is open.

The Foldouts article owns the physical technology.

For page planning, the important lesson is simply:

the planning canvas can be larger than the page the reader sees when the book is closed or partly folded.

Any multimodal analysis should therefore know the full planning surface.

A panel-only view can cut the design at a modern convenience boundary.

What Survives the Page-Planning Work

Voynich text and images were often planned as one shared spatial composition.
Illustration outlines generally preceded the principal text on illustrated pages.
Text repeatedly avoids and fits around pre-existing visual elements.
Available line width therefore depends partly on image geometry.
Some tokens fit closely into narrow visual spaces.
This supports deliberate spatial coordination.
Tight spatial fit does not by itself prove generated filler.
Prepared drafts, copied exemplars, abbreviation and flexible orthography remain serious alternatives.
Labels occupy a different spatial role from running prose and form a distinct textual register.
Image adjacency proves relation, not necessarily naming or literal description.
Circular and radial writing should preserve geometry in analysis.
Page layout can confound line-position, word-length and section-level text statistics.
Text-only pages provide useful controls for separating geometry from writing-system effects.
Current binding order should not override bifolium and foldout planning relationships.

What Does Not Survive as Established Knowledge

Tight text fitting proves the text is meaningless filler.
Tight text fitting proves the text was meaningful and prewritten.
Every label literally names the adjacent object.
Every paragraph describes the picture beside it.
Every line break is semantic rather than geometric.
Every short token was chosen only because a narrow gap required it.
Circular text has an obvious modern left-to-right starting point.
Radial text can be flattened without information loss.
Page layout alone identifies the manuscript’s function.
Current page adjacency preserves the original compositional plan.

The spatial coordination survives.

The semantic mechanism remains open.

A Better Text–Image Analysis

Record outline, text and paint as separate production layers.
Measure usable line width rather than only character count.
Mark lines shortened by visual obstacles.
Compare token length with available geometric width.
Separate labels from running text.
Preserve circular and radial geometry.
Use text-only pages as layout controls.
Control Currier regime, hand and document role before assigning semantic causes.
Compare generated-filler, preplanned-draft, abbreviation, encoding and exemplar models under the same spatial observations.
Require the winning model to predict unseen spatial-text relationships without changing its rules.

What Would Count as a Real Page-Planning Breakthrough?

Imagine every line on a large sample of illustrated and text-only Voynich pages is given a geometric profile.

Researchers measure:

available width;
distance to visual obstacles;
token lengths;
line-final forms;
abbreviation-like variants;
Currier regime;
hand;
document role.

A model derived from one set of pages predicts which token families should appear in narrow versus wide spaces on unseen sheets.

Now suppose the same model also predicts label morphology and line-final variants.

If spatial width alone explains most variation, local generation or mechanical layout rules strengthen.

If stable lexical and grammatical relationships survive geometry while surface forms adapt systematically, meaningful language or notation strengthens.

Either result would be valuable.

The breakthrough is not deciding in advance whether the text is filler.

It is discovering how much of the visible writing is caused by what the page physically allowed.

Primary School: Write Around the Tree

Draw a large tree in the middle of a sheet of paper.

Ask a child to copy one paragraph around it without writing through the branches.

The lines become different lengths.

Some words no longer fit.

The child must decide whether to move the word, abbreviate it or continue on the next line.

The meaning did not change.

The geometry changed the visible text.

This is the core Voynich page-planning problem.

Lower Secondary: Same Meaning, Different Space

Give students three boxes of different widths.

Ask them to fit the same information into each using approved abbreviations.

Then give another group permission to write meaningless but word-like strings instead.

Both groups can fit the boxes.

Now ask what evidence would distinguish them.

Only the meaningful group preserves stable relationships when the text is decoded or reused elsewhere.

The exercise shows why spatial fit alone is underdetermined.

Upper Secondary: Control the Geometry

Give students two sets of coded pages.

Set A contains large illustrations that shorten lines.

Set B contains only text.

Ask them to compare:

average line length;
final-token length;
special ending frequency;
first-token behaviour.

If the difference follows layout, geometry is a confounder.

Students learn why a good comparison removes alternative causes rather than merely finding a pattern.

JC and Adult Readers: Page Planning as a Constraint Optimisation Problem

At a higher level, page composition can be modelled as constrained optimisation.

The page provides a two-dimensional region with obstacles.

The scribe must place a sequence of textual units subject to:

available width;
illustration boundaries;
line-height conventions;
token segmentation;
possible abbreviation or variant choices.

Different hidden mechanisms optimise under different freedoms.

A meaningless generator has high semantic freedom.

A natural-language copyist has lower semantic freedom but may have orthographic flexibility.

A cipher clerk may have multiple equivalent ciphertext forms.

An exemplar-copying scribe inherits a pre-solved layout.

The observed page is the outcome.

The research problem is to determine which mechanism can produce the outcome with the least arbitrary freedom while also explaining the rest of the manuscript.

A Parent and Teacher Guide

Notice which came first: drawing or text.
Measure the space the picture leaves for writing.
Do not turn tight fit into proof of nonsense.
Remember that meaningful shorthand can fit space too.
Separate labels from prose.
Preserve circular and radial geometry.
Use text-only pages as controls.
Ask which mechanism predicts both layout and text structure under fixed rules.

The broader lesson is one of the most useful in the entire Voynich series:

when information and space are tightly coupled, do not confuse the constraints of the container with the meaning of what it contains.

Reader Checklist: Before You Explain Why Voynich Text Fits Its Pictures

Did the outline clearly precede the text?
How much usable width was available?
Was the line shortened by a visual obstacle?
Does token length correlate with available width?
Could abbreviation or alternate forms explain the fit?
Could the layout have been copied from a prepared exemplar?
Could local generation explain the fit?
Does the same token family retain stable behaviour elsewhere?
Is the text a label, prose, circular text or radial text?
Is adjacency being mistaken for naming?
Are line effects being measured without controlling page geometry?
Is current page order being mistaken for original planning order?
Does the proposed mechanism predict unseen text-image layouts?

Frequently Asked Questions

Were Voynich drawings made before the text?

On illustrated pages, the evidence strongly indicates that the principal outlines generally preceded the text because the writing repeatedly avoids and fits around those outlines.

Does text fitting around drawings prove it is meaningless filler?

No. Generated filler is one viable explanation, but prepared drafts, copied exemplars, abbreviation, flexible orthography and variable encoding can also create tight spatial fit.

Do labels name the objects beside them?

Not necessarily. Their placement proves a spatial relationship. They could be names, categories, numbers, properties, cross-references or other identifiers.

Why are Voynich right margins irregular?

On many illustrated pages, available text width changes because plants, bodies, containers and diagrams intrude into the writing region. The irregular margin is therefore partly a product of page geometry.

Does image geometry affect line statistics?

Potentially yes. If drawings shorten or reset lines, they can affect line-final and line-initial positions, so layout should be controlled in positional-text studies.

Why is circular text different?

A circle has no inherent left margin or obvious first token. Any linear transcription must choose a cut point and direction, which are analytical choices rather than guaranteed medieval ones.

Could page layout support a cipher?

Yes. If one plaintext unit has several encoded forms of different lengths, a scribe could choose among them according to available space. A specific cipher must demonstrate the selection rule.

Why are text-only pages useful?

They reduce visual obstacles and therefore help separate layout-driven effects from properties intrinsic to the writing system or Currier regime.

What is the strongest current conclusion?

Voynich page composition is deliberately multimodal: drawings, text, labels and diagram geometry share one planned surface. That coordination is real. Its semantic mechanism remains unresolved.

Research and Further Reading

The Final Idea

The Voynich page is not a picture with text added to it.

Nor is it a text page decorated with pictures.

Too often, those modern categories are too simple.

The drawing determines where the writing can go.

The writing turns the empty spaces into textual fields.

Labels attach words to visual nodes.

Circles bend text into rings.

Radial structures create directional relationships.

Foldouts enlarge the planning surface beyond the ordinary page.

The manuscript’s information is therefore partly spatial before it is semantic.

That does not solve the Voynich Manuscript.

It tells us what a solution must respect.

Whatever Voynichese ultimately is, it was written by someone who was not only choosing signs. They were choosing where those signs belonged in relation to the rest of the page.

Continue Through the Voynich Research Map

This article is one specialist node in eduKateSG’s larger Voynich research library. Return to the canonical master to see the physical, visual, textual and historical evidence in one continuous argument.

Structure is evidence; structure is not translation. The master keeps resemblance, statistical structure, historical possibility and decipherment claims at separate evidentiary levels.

Voynich | Everything eduKate Knows and Tested | When Text Meets Image: How the Page Was Planned