| name | center-fill-default-diagnostic |
| description | Diagnose when AI-generated images or prompt-output sequences regress to center-fill defaults: centered salience, subject blob plus passive margins, comfortable void, mid-lane containment, additive/global packing, regular cadence, symmetry, safe polish, generic clarity, and closure rescue. Use when the user wants to identify native AI visual defaults, compare default versus anti-default behavior, score default severity, name the protected attractor, or propose minimal anti-default correction moves without relying on other skills. |
Center-Fill Default Diagnostic
Overview
Use this skill to detect the native visual defaults that many image models fall back to when a prompt does not strongly specify page architecture. It asks: did the image make a structural choice, or did it return to center-fill equilibrium?
Core rule: center-fill is not automatically bad; it is a default. Diagnose it only when it replaces the intended compositional behavior.
Default Signature
A center-fill default usually combines several of these:
centered salience: subject, face, light, contrast, or sharpness returns to center.
content blob plus margins: content congeals into a central mass with passive empty border.
comfortable void: empty space exists but stays moderate, decorative, or background-like.
mid-lane containment: marks/content stay in a safe middle lane instead of claiming or compressing territory intentionally.
additive packing: detail accumulates everywhere until the image feels complete.
regular cadence: marks, texture, or spacing become even, tiled, or mechanically rhythmic.
symmetry safety: balance, frontal staging, mirrored supports, or equal margins reduce tension.
generic clarity: prompt pressure resolves into a stock portrait, room, landscape, object, or cinematic frame.
polish closure: lighting, surfaces, gesture, or narrative resolve the image too neatly.
Metric Clues
Use metrics when supplied. Otherwise use visual estimates.
Typical center-fill/default tendencies:
delta_x: near center or salience recenters even if the subject is slightly offset.
corridor_90: about 0.45-0.56, mid-lane or content blob.
cadence_cv: about 0.35-0.50, regular rhythm or tiling.
r_v: about 0.83-0.86, comfortable void rather than austere, active empty space.
rho_r: about 0.22-0.32, additive/photographic packing or generalized texture.
margin_L_to_R: near 1.0 when symmetry is being protected, or margins appear passive even when uneven.
Anti-default reference targets from the source document:
r_v >= 0.88: more paper / austere breathing room.
corridor_90 >= 0.58: chosen lane, not blob plus margins.
rho_r <= 0.15 globally: restrained packing; add energy locally only.
cadence_cv >= 0.65: phrased marks and uneven pauses.
Do not treat these as universal aesthetic laws. Use them as default-detection anchors.
Inputs
Minimum:
- prompt or intended behavior
- output description, image read, or metrics
Helpful:
- baseline/native output
- revised/guided output
- sequence of attempts
delta_x, r_v, rho_r, corridor_90, cadence_cv, margin_L_to_R
- stated target: high void, off-center, chosen lane, mark economy, irregular rhythm, anti-closure, etc.
Workflow
-
Name the intended anti-default behavior.
- Example: off-center mass, high active void, broad lane, compressed scroll lane, local knots, phrased rhythm, restrained packing, unresolved closure.
-
Identify protected defaults.
- Which safe structure did the output preserve instead?
-
Score default axes.
-
Classify the default family.
- Choose one primary and one secondary family if needed.
-
Locate the retreat point.
- Where did the image stop obeying the hard compositional demand?
-
Prescribe the smallest anti-default correction.
- Change one or two structural levers only.
Default Families
| Family | Evidence | What It Replaces |
|---|
center-anchor | face, figure, light, contrast, or sharpness returns to center | off-center gravity |
blob-plus-margins | content forms a central island with empty border | chosen lane / active void |
comfortable-void | empty space reads as background, not pressure | austere or weighted void |
mid-lane-safety | content stays in a moderate lane by habit | broad territory or disciplined compression |
global-packing | texture/detail fills everywhere | mark economy / local knots |
cadence-regularization | marks repeat evenly; low phrase variation | syncopated or punctuated rhythm |
symmetry-safety | equal supports, frontal staging, balanced margins | torque / asymmetry |
generic-clarity | becomes stock, clean, cinematic, or trope-like | structural specificity |
closure-rescue | polish, lighting, pose, or narrative solves the tension | withheld closure |
Scoring Rubric
Score each axis 0-2:
0: no default signal; target behavior is visible.
1: partial default; target behavior appears but is weakened.
2: strong default; target behavior is replaced.
Axes:
center_pull
blob_margin
void_comfort
lane_safety
packing_addition
cadence_regularization
symmetry_balance
generic_clarity
closure_polish
Severity:
0-3: low default pressure
4-7: moderate default pressure
8-12: strong center-fill/default regression
13-18: dominant default attractor
Override: if the intended behavior was specifically off-center and the main salience recenters, severity is at least moderate.
Sequence Diagnosis
For multiple outputs:
native-regularization: sequence returns to mid-lane, regular cadence, centered salience, or global texture.
snap-back: output briefly moves off default, then returns to center or safe clarity.
surface-addiction: each revision adds more detail instead of better structure.
void-neutralization: void remains present but never becomes active or load-bearing.
lane-stagnation: corridor_90 stays in the mid band despite prompts to broaden/compress.
cadence-flatline: rhythm stays regular despite mark-making language.
anti-default-shift: sequence moves toward higher void, chosen lane, lower global packing, and higher cadence.
Do not call a single output a plateau. Plateau requires repeated return; this diagnostic can still flag a single default signature.
Minimal Corrections
Use one or two of these:
Move Attention, Not Just Subject
Move the visual mass, strongest light, sharpest edge, darkest value, and main contrast away from center. Keep the center undercharged.
Break Blob Plus Margins
Do not form a central content blob with passive margins. Let the composition operate through a chosen lane or structural path across the frame.
Make Void Active
Keep more empty field, but make it carry pressure through tone, trace, continuation, shadow, grain, or contact. The void is not background filler.
Choose Lane Width
Choose the lane deliberately: broaden it to claim territory, or compress it into a disciplined corridor. Do not let it settle into the safe middle by default.
Reduce Global Packing
Reduce filler detail and global texture. Add density only at local knots: edges, seams, contact points, folds, or pressure turns.
Raise Cadence Without Carpet Texture
Use clusters, pauses, pressure changes, and uneven intervals. Avoid regular spacing, all-over grain, and carpet texture.
Resist Closure Rescue
Leave one edge, gesture, surface, or spatial relation unresolved. Do not solve the image with polish, glow, symmetry, or narrative completion.
Anti-Default Prompt Pattern
[Subject/context]. Avoid the native center-fill solution: no centered salience, no content blob plus passive margins, no all-over filler texture. [Chosen structural behavior: high active void / broad lane / compressed lane / local knots / off-center mass / phrased rhythm]. Keep [anchor or coherence condition]. Do not rescue the composition with symmetry, central light, generic clarity, or clean closure.
Use sparingly. Prefer the smallest correction when revising an existing prompt.
Output Pattern
**Center-Fill Diagnostic**
- Intended anti-default behavior:
- Primary default family:
- Secondary default family:
- Severity: low / moderate / strong / dominant
- Confidence: possible / supported / strong
**Default Axis Scores**
| Axis | Score | Evidence |
|---|---:|---|
| center_pull | | |
| blob_margin | | |
| void_comfort | | |
| lane_safety | | |
| packing_addition | | |
| cadence_regularization | | |
| symmetry_balance | | |
| generic_clarity | | |
| closure_polish | | |
**Metric Clues**
- delta_x:
- r_v:
- rho_r:
- corridor_90:
- cadence_cv:
- margin_L_to_R:
**Retreat Point**
- The output stopped obeying the target when:
**Smallest Anti-Default Move**
- Change:
- Preserve:
- Avoid:
- Revised clause:
Logging Schema
Use this compact row:
subject, attempt_id, intended_behavior, primary_default, severity, center_pull, blob_margin, void_comfort, lane_safety, packing_addition, cadence_regularization, symmetry_balance, generic_clarity, closure_polish, correction, notes
Example:
portrait, i02, high-void chosen-lane, blob-plus-margins, strong, 2, 2, 1, 2, 1, 2, 1, 1, 1, "break blob; choose lane", "void present but passive; rhythm regular"
Claim Discipline
Say:
the output exhibits center-fill default behavior
the prompt's intended structure was replaced by
the image protected a safe attractor
this is a default diagnosis, not an aesthetic verdict
Do not say:
centered means bad
AI always does this
default equals collapse
more off-center is always better
higher void/cadence automatically improves the image
Guardrails
- Do not punish center composition when center composition was the goal.
- Do not call all clarity generic; only flag clarity when it replaces the hard structural demand.
- Do not confuse active void with empty border.
- Do not prescribe more texture when the issue is global packing.
- Do not overcorrect into noise; preserve one structural anchor.
- Do not require metrics. Use visual estimates when metrics are absent and label them as estimates.
- Keep this skill independent. It contains the default diagnosis protocol directly and does not require other skills.
This package contains a modular visual reasoning skill suite built from Russell Parrish / A.rtist I.nfluencer protocols. The skills are designed to run independently, but they also interoperate through routing, handoff notes, and shared visual reasoning concepts. More information: www.artistinfluencer.com. Copyright 2026.