Repair Mode: The 30-Second Fix for Almost-Good AI Images
Playbooks17 April 2026Published by Pen & Muse

Repair Mode: The 30-Second Fix for Almost-Good AI Images

Back to Dispatches
Part 13 of a 15-part series92 min read · 18,266 words total
Dispatch SeriesPart 13 of 15
The Physics of Better AI Images

Realism isn’t a look—it’s the consistent physics of causes across geometry, light, and camera capture.

Series PositionPart 13 of 15
Repair Mode: The 30-Second Fix for Almost-Good AI Images
Shot Design for AI Images: Framing, Blocking, and Camera Position Like a Director

Previous · Part 12

Shot Design for AI Images: Framing, Blocking, and Camera Position Like a Director

The Real-Camera Prompt Cookbook (Part 14): Shot Recipes That Don’t Look Like AI

Next · Part 14

The Real-Camera Prompt Cookbook (Part 14): Shot Recipes That Don’t Look Like AI

This builds on Part 12: Shot Design for AI Images: Framing, Blocking, and Camera Position Like a Director

Continue with Part 14: The Real-Camera Prompt Cookbook (Part 14): Shot Recipes That Don’t Look Like AI

Repair Mode: How to Fix an Almost-Good AI Image Without Rewriting Everything

Most people “fix” AI images by starting over.

But the better move is repair. Because almost-good images already contain the physics—your job is to restore the parts that drifted.

What “Almost-Good” Usually Means (and why it matters)

An almost-good image is typically coherent in structure:

  • framing is close
  • subject reads
  • lighting direction feels plausible
  • materials are mostly behaving

The failure is usually in one tight cluster: a mismatch between what the image implies (geometry, contact, scale, lens character) and what it actually delivers (edge contrast, micro-contrast, depth cues, or exposure behavior).


The Repair Stack: Change Order, Not Just Content

If you jump straight to “rewrite the prompt,” you waste the existing good signal. Instead, work in this order.

1
Identify the single most visible “physics lie.”
2
Test one small change that targets that lie.
3
Lock the fix, then only address secondary symptoms.
4
Stop once the image stops feeling like a composite.

Step 1 — Diagnose the Failure Channel (The part you can’t skip)

Use the “reader test.” Ask what feels wrong in one sentence, without blaming the whole image.

Common failure channels:

1) Edge & Contour Wrongness (the “cutout” feeling)

Symptoms:

  • subject edges look too clean or too obedient
  • silhouettes lack tactile tension
  • hair/leafage boundaries feel smeared or over-resolved
✦ Applied Series

Unlock the full applied series

Get all 15 parts of The Physics of Better AI Images — 92 min read · 18,266 words across the full sequence.

Inside the series

Parts 1-4: Constraints make it real

Parts 5-9: Camera causes, not vibes

Parts 10-12: Surfaces, framing, and truth priority

Parts 13-14: Repair and recipe workflows

Part 15: Master Prompt Sheet

Realism isn’t a look—it’s the consistent physics of causes across geometry, light, and camera capture.

Membership gives access to subscriber dispatches and ongoing Pen & Muse tools. This paid series is sold separately.

Explore membership plansLog in

If this resonates, see how to apply it to your own work with the interactive Dispatch agent.

Be first to like this dispatch

More in PlaybooksView all →
Keep Reading, Then Step Inside
Cover of The Deletion Code
Book + Immersive Experience£3.99

The Deletion Code

Pen & Muse

In the hyper-connected metropolis of Neo-Kobe, digital diver Bennett Walker enters Senator Halloway’s neural-link and discovers something impossible: a jagged deletion code where the senator’s formative memories should be. Before Walker can trace it, the link catastrophically fails—and a remote override uses his own credentials to frame him for the death.

Platform Access

Interested in building narratives using our proprietary architecture? Join the creator waitlist.