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


Previous · Part 12
Shot Design for AI Images: Framing, Blocking, and Camera Position Like a Director

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



