Store What Should Remain; Ask What Is New
Every image contains two kinds of decisions.
Some belong only to the present scene.
Others belong to the larger world, brand, character, campaign, or body of work surrounding it.
- A woman may be sitting alone at a desk tonight, but her face, jewellery, clothing language, workplace, and relationship to that environment may have been established months ago.
- A company may need a new recruitment image, but its palette, logo rules, workplace identity, audience standards, and representation principles should not be reinvented by each employee who opens an image generator.
- A fictional courtyard may appear at dawn in one scene and during a storm in another, but it should not acquire a different floor plan, architectural history, or material language every time the model changes.
The problem is that most image-generation workflows treat all decisions as though they are equally new.
The user begins with an empty prompt field and has to restate everything:
the character, the place, the style, the palette, the attire, the mood, the exclusions, the architecture, the proportions, the recurring objects, the intended tone.
Then they do it again for the next image.
And again in a new conversation.
And again when they change platform.
Eventually, repetition becomes so exhausting that important details are shortened, forgotten, contradicted, or left for the model to invent.
The result is drift.
The Visual Register by Mithaq Praxis is built around a different rule:
Store the decisions that should remain. Ask only for the decisions that are new.
This is not merely a convenience feature.
It is the principle that allows the product to reduce unnecessary labour without removing meaningful human direction.
Not every decision belongs in the prompt
A prompt is often treated as though it must contain the entire identity of a project.
That is an unstable way to work.
A prompt is temporary. It belongs to one request, one scene, one platform, and often one model version.
A visual identity is larger than that.
If a project already knows that:
- its architecture uses warm limestone and restrained arches;
- its technology should feel near-future but credible;
- its recurring character has long indigo hair and amber eyes;
- its palette centres on charcoal, paper, muted green, and warm stone;
- its spaces should remain clean, calm, and lightly inhabited;
- neon cyberpunk imagery does not belong;
- a specific necklace should remain present;
- one logo must never be distorted;
then those are not new prompt decisions.
They are established continuity.
Requiring the user to restate them every time does not make the process more human-led. It merely forces the human to perform clerical repetition.
The meaningful new question is not:
What colour are the walls again?
It is:
What is happening in this room today?
The Visual Bible stores the first kind of knowledge.
The brief composer asks for the second.
Persistent identity and present intention
The relationship can be expressed simply:
The Visual Bible holds persistent identity.
The human supplies present intention.
The visual brief combines them.
Persistent identity may include:
- who the characters are;
- what a place looks like;
- which materials belong to the world;
- how a company presents itself;
- which colours recur;
- what attire is typical;
- which objects carry meaning;
- what must remain accurate;
- what should be avoided.
Present intention includes:
- what is happening now;
- why the image is being made;
- who appears in this scene;
- what each person is doing;
- what emotional condition belongs to this moment;
- where the viewer’s attention should go;
- what format the image must serve;
- what differs from the usual visual state.
The Bible should not invent the event.
The scene should not have to reconstruct the entire world.
That division of labour is the spine of The Visual Register.
The principle came from practical frustration
I already work with recurring visual systems.
Bayt al-ʿAhd, Algorithm Atelier, and Mithaq Praxis each have their own visual identity.
When I create a new Bayt scene, I do not reconsider from nothing whether the world contains:
- arches;
- warm stone;
- lanterns;
- manuscripts;
- fig trees;
- indigo and amber;
- layered domestic spaces;
- recurring jewellery;
- particular emotional and architectural rhythms.
Those decisions already exist.
What is new may be:
Farah sits beneath the fig tree after rain, reading a letter while Zayd watches from the covered walkway.
The present scene changes.
The deeper world remains.
For Mithaq Praxis, the stored decisions may include:
- restrained Islamic modernism;
- Ottoman and Levantine architectural influence;
- clean institutional spaces;
- muted green, charcoal, stone, and paper;
- editorial sophistication;
- desert-facing calm;
- purposeful holographic or embedded technology;
- no cyberpunk neon;
- no cluttered generic laboratory styling;
- no accidental return to Algorithm Atelier’s honeycomb language.
Then the new scene might be:
Farah and Zayd review a systems diagram at the central table while a small service robot carries archive boxes through the background.
The system should not ask me to redefine Mithaq Praxis every time.
It should ask what is new about this particular moment.
That is not a reduction of intention.
It is the preservation of intention across time.
Repetition is not the same as authorship
There is a tendency in discussions about human-led work to assume that more manual input always means more human authorship.
That is not necessarily true.
- Typing the same character description fifty times does not deepen the creative act.
- Copying the same palette into every prompt does not produce more meaningful direction.
- Rewriting the same prohibition against neon does not make the user more thoughtful.
Those decisions were already made.
Once made, they can be stored.
The human contribution lies in making, refining, and governing the decision—not in repeatedly re-entering it into an empty box.
A good system should preserve human choice without demanding unnecessary duplication.
This distinction matters because automation is not always the enemy of deliberateness.
Automation becomes a problem when it replaces decisions that should remain human.
It becomes useful when it carries forward decisions the human has already made.
The Visual Register should automate memory, not intention.
What should remain?
Not every repeated detail deserves to become a permanent rule.
A Visual Bible should not absorb every incidental feature from every image.
The fact that one scene contains rain does not mean the world is always rainy.
- A character wearing blue once does not automatically make blue part of their identity.
- A company using a dramatic low-angle image for one campaign does not mean every future employee photograph should be filmed that way.
Something belongs in the Bible when it has value beyond the present scene.
Useful questions include:
- Will this matter in future images?
- Would changing it accidentally create continuity problems?
- Does it help distinguish this project from another?
- Has the user corrected this detail more than once?
- Does it carry cultural, narrative, professional, or symbolic meaning?
- Should another collaborator know this before producing new work?
- Is it a default, a fixed rule, or merely an example?
The Visual Register should not turn one successful image into permanent law.
It should help the user recognise which decisions have become part of the identity.
Four levels of persistence
Not all stored decisions should have equal force.
A useful Visual Bible should distinguish among different levels of persistence.
1. Locked continuity
These details should remain stable unless the Bible itself is deliberately revised.
Examples include:
- a recurring character’s identifying features;
- established height relationships;
- a company logo;
- a product’s proportions;
- the architecture of a known room;
- a cultural or religious boundary;
- a permanent accessibility requirement;
- the number and identity of recurring group members.
If a generator changes these, the result has drifted.
2. Strong defaults
These define the project’s ordinary visual behaviour but may be overridden for a clear reason.
Examples include:
- preferred palette;
- common materials;
- usual lighting posture;
- typical attire;
- recurring level of realism;
- preferred environmental density;
- normal photographic or illustrative character.
The user does not need to choose these again unless the new scene departs from them.
3. Contextual rules
These apply only under particular conditions.
Examples include:
- formal attire for institutional events;
- softer lighting for domestic scenes;
- a campaign-specific accent colour;
- particular architecture for one location;
- different logo placement for vertical and horizontal formats;
- seasonal wardrobe changes.
The system should apply them only when the relevant context is active.
4. Optional motifs and easter eggs
These may appear selectively.
Examples include:
- a cat in the background;
- a recurring flower;
- a familiar book;
- a hidden date;
- a small symbolic object;
- one piece of jewellery used in emotionally significant scenes.
These details create depth when used sparingly.
If inserted into every image, they stop feeling meaningful and become visual clutter.
The Register should remember them without treating them as mandatory decoration.
What is new?
Once persistent identity is carried forward, the composer can focus on the decisions that actually belong to the present image.
The user might be asked:
- What do you want to see?
- What is the image for?
- Who is present?
- What is each person doing?
- Where does the scene take place?
- What should the viewer notice first?
- What should the image feel like?
- What changes from the selected Visual Bible?
- Is anything normally present being omitted deliberately?
- What format or placement must the image serve?
These questions are not filler around a prompt.
They define the present act of direction.
Suppose the selected Bible already knows the four women in a summer friendship series.
It knows:
- their distinct faces;
- hair colours;
- relative heights;
- usual swimwear palette;
- Mediterranean resort environment;
- friendship dynamic;
- preference for natural anatomy;
- prohibition against duplicated characters;
- non-sexual editorial tone.
The user should not need to type all of that again.
The new direction may be:
All four try to balance on one oversized pool float. One is already falling, one grabs another’s arm, one laughs helplessly, and one remains intensely determined not to lose her composure.
That sentence contains the present scene.
The Bible provides the people and world.
The brief combines them.
A new decision may override an old default
Stored continuity must not become automatic obedience.
A user should be able to depart from a default deliberately.
- If the Bible normally uses warm evening light, the new scene may require cold morning light.
- If a character usually wears dark clothing, the present scene may involve ceremonial white.
- If a company usually favours restrained documentary photography, a particular campaign may intentionally use bold flat illustration.
The important distinction is between:
- accidental drift;
- deliberate variation.
The Visual Register should make overrides visible.
For example:
Inherited default
Restrained neutral palette with muted green accents.
Scene override
Use a temporary red accent for the emergency-preparedness campaign.
Or:
Inherited environment
Research library with warm stone, walnut, paper, and integrated near-future displays.
Scene override
The library is shown during a power outage, lit only by emergency strips and handheld lamps.
The world has not been forgotten.
It is being placed under a new condition.
That is creative direction.
An override should not silently rewrite the Bible
A one-time variation belongs to the brief.
A lasting correction belongs to the Bible.
These should not be confused.
- Suppose a user creates one scene where a recurring character wears their hair short for a disguise.
That should not automatically alter the permanent character profile. - Suppose a company runs one monochrome campaign.
That should not erase the main brand palette. - Suppose a building is shown damaged after a fictional event.
That does not redefine its ordinary architecture.
The system should ask:
Is this change for this brief only, or should it update the Visual Bible?
This small distinction prevents temporary ideas from contaminating persistent continuity.
It also allows the Bible to evolve deliberately.
Corrections reveal what should be stored
Repeated failure is often evidence that a decision belongs in the Visual Bible.
If the user keeps correcting:
- The room should be cleaner.
- The technology should be more purposeful.
- Do not use blue neon.
- Keep all four women visually distinct.
- Zayd should be taller, but not by a full head.
- The logo must use the actual wordmark.
then those are no longer isolated prompt edits.
They are accumulated visual knowledge.
The system should allow the user to promote a correction from one brief into the Bible.
For example:
Save as a continuity rule?
Height differences among the four men should remain subtle because they are only one to three inches apart.
Or:
Save to the Mithaq Praxis avoid list?
Do not use decorative holograms that have no visible relationship to the work being performed.
The Bible becomes more useful because it learns from real production problems—not because a model autonomously rewrites it, but because the human recognises which corrections should persist.
The inheritance should remain visible
A user should be able to tell which parts of a brief came from:
- the Visual Bible;
- the new scene;
- a temporary override;
- a system-generated technical translation.
This matters for clarity and trust.
A finished brief might show:
Inherited from the Visual Bible
- four established adult women;
- Mediterranean resort environment;
- coordinated but distinct swimwear;
- candid non-sexual friendship;
- realistic anatomy;
- no duplicated characters;
- bright seasonal palette.
New scene direction
- the group attempts to balance on one oversized pool float;
- one woman falls;
- one grips another’s forearm;
- one laughs;
- one maintains exaggerated seriousness.
Technical choices
- vertical 2:3;
- full-body action composition;
- bright midday light;
- water-level camera position;
- crisp editorial photography.
Scene-specific exclusions
- no text overlay;
- no collage;
- no additional people;
- no glamour posing.
The user can inspect the structure rather than receiving an opaque block of polished prose.
That visibility keeps the tool from claiming authorship over decisions it merely carried or formatted.
The same principle works without a Visual Bible
Not every user will begin with a persistent world.
Someone may need one image for one purpose.
They should still be able to write:
A young engineer demonstrates a water-filtration prototype to a small rural community.
The composer can then ask for the new decisions necessary to make that scene usable.
Because there is no Bible attached, more information may need to be established:
- location;
- tone;
- audience;
- representation;
- clothing;
- level of realism;
- visual medium;
- format;
- environmental detail;
- what should be avoided.
This does not mean the tool must force the user to create a Bible first.
The brief can stand alone.
But if the user later creates several related images, the system may notice that some decisions recur.
It might ask:
You have used the same organisation, prototype, palette, and community setting across several briefs. Would you like to preserve these as a Visual Bible?
A Bible can therefore be designed before work begins or emerge from repeated practice.
Both paths are valid.
A creative world may contain nested continuity
Visual identities are rarely flat.
Mithaq Praxis may have an overall institutional Bible, but different areas within it can have their own profiles:
- Editorial Room;
- Systems Laboratory;
- Research Library;
- Continuity Lounge;
- Threshold Gallery;
- Farah’s workstation;
- Zayd’s workstation.
The parent Bible may define:
- palette;
- architecture;
- technology language;
- institutional mood;
- materials;
- relationship to the Bayt compound.
A child environment profile may define what is unique to one room.
- The Research Library may emphasise paper, walnut shelving, annotated books, retrieval interfaces, and desert light.
- The Systems Laboratory may use larger interactive surfaces, robotics workstations, diagnostic displays, and controlled technical lighting.
Both belong to Mithaq Praxis.
Neither should be rebuilt from zero.
The same structure can apply to companies:
Company Bible
→ Department profile
→ Campaign profile
→ Present brief
Or to fiction:
Story-world Bible
→ Region
→ Location
→ Character group
→ Present scene
The principle remains the same at every level:
Inherit what is already known. Ask for what changes here.
Corporate consistency should not mean corporate sameness
A governed Company Visual Bible can reduce unnecessary reinvention.
It may store:
- brand identity;
- palette;
- approved logos;
- workplace depiction;
- product accuracy;
- audience standards;
- accessibility requirements;
- legal restrictions;
- preferred photographic language;
- prohibited stereotypes;
- review requirements.
An employee creating a recruitment visual should not need to memorise and manually paste every corporate rule.
Those should be inherited.
But the employee still needs to state what the new image should communicate.
For example:
A senior technician welcomes three apprentices into the workshop on their first morning.
The company Bible may establish the workplace, uniforms, equipment, representation principles, logo rules, and image tone.
The new request establishes:
- the event;
- the people involved;
- the emotional purpose;
- the present communication need.
A governed system therefore does not remove decision-making.
It places different decisions with the appropriate people.
- The brand team governs what should remain.
- The requester defines what is needed now.
- The designer interprets and refines the scene.
- The reviewer judges whether the result serves the organisation.
Governance is controlled inheritance
This gives a more precise way to understand corporate use.
Governance is not primarily a more complicated prompt form.
It is control over which decisions are inherited and who may change them.
A company may decide:
- staff can use the approved palette but cannot alter it;
- a designer may create campaign-level variations;
- only a brand administrator may replace official logos;
- product descriptions remain locked;
- certain content requires legal review;
- publication status requires approval.
The creative composer can remain largely the same.
What changes is the authority around persistent decisions.
This prevents every employee from creating a private interpretation of the company’s brand while still allowing scene-specific direction.
The Visual Bible becomes organisational memory.
The principle protects designers from vague requests
When persistent rules are stored, it becomes easier to see what the requester still has not decided.
A manager may say:
Make us something for the annual report.
The company Bible can supply:
- the visual identity;
- approved colours;
- photography treatment;
- logo rules;
- typography space;
- accessibility standards.
But it cannot supply the missing present intention.
The composer must still ask:
- Which section of the annual report?
- What should the image communicate?
- Who is the audience?
- Is the focus growth, responsibility, research, staff, or customers?
- What subject belongs in the frame?
- What action should be shown?
The Bible does not conceal the absence of a brief.
It makes that absence more visible.
That is useful to designers because it separates:
- established brand decisions;
- unresolved communication decisions.
The requester cannot assume that “use the brand” is a complete creative direction.
The principle protects users from forms
There is another reason to store persistent decisions: a good tool should not make the user fill in fields merely because the database contains them.
If the Bible already knows:
- the character;
- the setting;
- the palette;
- the style;
- the attire;
- the avoid list;
then those sections should appear as inherited summaries, not mandatory repeated questions.
The composer should become shorter as the Bible becomes stronger.
This is where the product earns the right to ask users to be deliberate.
It does not waste their attention on things already decided.
It reserves attention for what actually needs thought.
A deliberate tool should not confuse deliberateness with bureaucracy.
Progressive disclosure follows the same law
The interface does not need separate creative modes to support beginners and advanced users.
It can ask essential new questions first:
- What do you want to see?
- What is it for?
- Who or what appears?
- What is happening?
- Where is it?
- What should it feel like?
- What format is needed?
Then allow deeper decisions to be opened when useful:
- blocking;
- visual hierarchy;
- shot size;
- camera angle;
- lighting direction;
- material detail;
- lens character;
- depth of field;
- negative space;
- treatment-specific controls.
The user should not have to choose between being “basic” and “professional.”
They should be able to decide how much depth the current image requires.
- A simple icon may need very little.
- A book cover, campaign hero, or four-person action scene may need much more.
The tool should ask only what is new and only what is relevant.
Technical parameters should follow intention
Aspect ratio, camera angle, and lighting may be new decisions, but they should not appear as disconnected technical settings.
They should follow the image’s purpose.
A user creating a website hero may need:
- a wide composition;
- clear negative space;
- a readable focal point;
- restrained background activity.
A vertical book promotion may need:
- a strong central figure;
- layered depth;
- space for title treatment;
- a composition that remains legible on a phone.
A company presentation image may need:
- clear subject separation;
- lower visual complexity;
- room for accompanying text;
- colours that survive projection.
The system can recommend technical choices based on the user’s stated purpose, but it should make the reasoning visible.
For example:
A wide landscape ratio may suit the website banner because it leaves room for a headline beside the subjects.
The user then accepts or changes the recommendation.
The system is helping translate intention into craft—not replacing intention with defaults.
Inheritance can reduce prompt length without reducing meaning
A common belief is that stronger image prompts must be longer.
Sometimes they are.
But once a Visual Bible exists, the prompt can become shorter because the underlying specification is stored elsewhere.
A human-readable brief might refer to:
Apply the Mithaq Praxis institutional environment profile and the established Farah and Zayd character cards.
The platform-specific prompt may still need selected details expanded, depending on whether the generator can access the references.
But the user does not need to manually reconstruct the whole identity.
This is especially valuable when moving between systems.
The Visual Register can select the relevant parts of the Bible and include only what the target workflow needs.
The canonical source remains complete.
The exported prompt remains situational.
Different platforms will need different amounts of inheritance
A model operating inside a long conversation may already know the established character references and world.
- A fresh platform may know nothing.
- A local workflow may support saved character adapters or reference images.
- Another may only accept text.
The Visual Register should not pretend these contexts are identical.
Instead, it can preserve one canonical brief and expand the inherited material as needed.
For example:
Context-rich export
Use the established Mithaq Praxis environment and Farah/Zayd character references.
Context-free export
The system may include a fuller summary of:
- their appearances;
- the room;
- the palette;
- the technological language;
- relevant exclusions.
The underlying decisions remain the same.
Only the amount of explanation changes.
This is another reason the product should centre the Bible rather than a single supposedly universal prompt.
The user should be able to see what changed
When a new brief is created from an existing one, the system should make the differences visible.
For example:
Previous scene
Four women share breakfast in a shallow infinity pool during soft morning light.
New scene
The same four women play a chaotic volleyball game at hard midday.
Changes:
- activity;
- light;
- composition;
- energy;
- water movement;
- subject blocking.
Unchanged:
- character identities;
- friendship tone;
- resort architecture;
- attire system;
- realism;
- non-sexual editorial treatment;
- body-proportion rules.
This can help users understand continuity as a set of deliberate relationships rather than a vague sense that the images “match.”
It also makes revision easier.
If the new image feels wrong, the user can ask whether the problem lies in:
- inherited identity;
- new direction;
- technical interpretation;
- generator failure.
Decision records become more meaningful
A lightweight decision record can show how persistent and new decisions interacted.
For example:
Inherited rule
Technology must appear purposeful and integrated into real workflows.
New scene request
Show a large holographic research map during a collaborative systems review.
Draft result
Rejected because the display dominated the room as decorative spectacle and obscured the people using it.
Revision
Reduced the display scale, placed it above the shared table, and made each participant interact with a distinct layer of the map.
This record reveals actual design judgment.
The problem was not that the model failed to include a hologram.
It failed to interpret the relationship between technology, people, and institutional purpose.
- The Bible provided the principle.
- The new brief provided the scene.
- The human review determined whether the result honoured both.
A growing Bible should make the system quieter
As the Visual Bible matures, the tool should not become louder, more complex, or more controlling.
It should become quieter.
The strongest Bible reduces the number of questions required.
It already knows:
- the people;
- the place;
- the materials;
- the visual lineage;
- the boundaries;
- the recurring problems.
The composer can focus on the present.
This is an important product test.
If the user builds a rich Visual Bible and still has to fill in the same long form for every image, the product has failed.
The value of accumulated continuity should be felt as reduced repetition and increased precision.
The system should carry more while deciding less.
The Bible itself must remain revisable
What should remain today may change tomorrow.
- A visual identity can mature.
- A company can rebrand.
- A character can age.
- A fictional building can be renovated.
- A creator can recognise that an old motif no longer belongs.
The rule is not:
Store everything permanently.
It is:
Store what should remain under the current version of the identity.
Changes should be deliberate and visible.
A future version of the system may allow:
- revisions;
- dated entries;
- change notes;
- parent and child profiles;
- campaign overrides;
- archived rules;
- restoration of previous versions.
The first prototype may not need all of this.
But the conceptual model should treat the Bible as living continuity, not immutable law.
A register should preserve reasons, not only facts
A rule becomes more portable when its reason is preserved.
Compare:
Avoid bright blue holograms.
with:
Avoid bright blue holograms because they pull Mithaq Praxis toward generic cyberpunk imagery and away from its restrained institutional identity. Interfaces should use muted, task-specific light and remain subordinate to the people and work.
The second entry is easier for:
- another designer to interpret;
- a manager to understand;
- an AI system to apply across different scenes;
- the creator to revise later.
The reason prevents the rule from becoming arbitrary.
It also allows intelligent variation.
- The problem is not the colour blue in isolation.
- The problem is the visual language it creates in that context.
The Visual Bible should preserve judgment, not merely prohibition.
This is how the system remains human-led
A human-led tool does not require a human to perform every repetitive action manually.
It keeps meaningful authority with the person.
The human decides:
- what belongs to the identity;
- what should be locked;
- what remains flexible;
- what the present image is meant to show;
- when a default should be overridden;
- whether a correction applies once or forever;
- which output succeeds;
- how the Bible evolves.
The system may:
- retrieve established rules;
- organise them;
- ask relevant questions;
- expose contradictions;
- format a brief;
- carry continuity forward;
- adapt the wording for another workflow.
The distinction is not between manual and automated.
It is between human-held decisions and machine-assumed decisions.
The Visual Register is designed to preserve the first and make the second visible when they occur.
Store the world. Ask for the moment.
The Visual Bible stores the world.
- The brief asks for the moment.
- The company stores its identity.
- The requester defines today’s communication need.
- The character card stores who the person is.
- The scene establishes what they are doing now.
- The environment profile stores the room.
- The brief determines whether it is dawn, midnight, crowded, empty, damaged, ceremonial, or ordinary.
- The motif library stores what may carry meaning.
- The human decides whether this is the scene where it should appear.
This is how continuity and intention can coexist.
One does not replace the other.
The Bible prevents the world from disappearing.
The scene prevents the world from becoming static.
The product law
The principle can be stated in a single line:
Store what should remain. Ask what is new.
But inside that line is the whole design philosophy of The Visual Register.
It means:
- do not make users repeatedly prove decisions they have already made;
- do not let a model reinvent established identity without permission;
- do not mistake administrative repetition for authorship;
- do not allow stored identity to replace present intention;
- do not freeze defaults into permanent creative limits;
- do not hide overrides;
- do not update continuity without the human knowing;
- do not ask questions whose answers already exist;
- do ask for the decisions that make this image different from every image before it.
The result should be a system that remembers without taking over.
- A system that carries continuity without manufacturing meaning.
- A system that gives the human less clerical work and more room for actual direction.
Because the goal is not to make the user repeat themselves more beautifully.
It is to preserve what they have already decided—and return their attention to the part that still requires them.
© 2026 • MITHAQ PRAXIS • CC BY-NC-ND 4.0 Unless Otherwise Stated.