4.1Inpainting (Vary Region), Pan, Zoom Out
The essentials in 30 seconds
- Inpainting fixes without destroying: Vary (Region) regenerates only the selected area and blends it with the rest — you keep a successful image instead of rolling the dice again.
- Outpainting extends: Zoom Out widens the field, Pan extends in one direction (handy for clearing text space).
- Sonia's reflex: small successive touches, an area neither too small nor too large, a prompt adapted to the area — and retry, inpainting is probabilistic.
Running case — Sonia and the Algiers café poster
Sonia, freelance graphic designer in Algiers, has generated an almost perfect poster for a downtown café: warm composition, late-afternoon light, a steaming cup in the foreground — but the hand holding the cup is botched and the framing is too tight for the client's logo. Instead of regenerating everything and losing that atmosphere, she selects the hand with Vary (Region), describes what she wants in that exact spot, and retries twice until she gets correct fingers. Then she applies a Zoom Out to recover some margin and a Pan upward to clear space for the logo. First milestone of the running case: a successful image is a fragile asset — you fix it locally instead of gambling it away.
This same case returns in the lesson, the diagram, the script and the 48h challenge — and continues in 4.2, 4.3, 4.4 and 4.5.
1. Inpainting: fix one area, keep the rest
The beginner's costly reflex: regenerating everything because a single detail is off. Inpainting — in Midjourney, Vary (Region) — does the opposite: you select an area with a brush or lasso, you adjust the prompt to describe what you want in that spot, and the model regenerates only that area while blending it with the rest. Like the hand on Sonia's poster, you repair without rebuilding.
Typical uses: fixing a botched hand, changing clothing, removing or adding an object, changing an expression, cleaning up a cluttered background. The decision rule: if the essentials work, repair locally; only rerun if the problem is upstream (framing, mood, misinterpreted subject).
2. Extending the image: Zoom Out and Pan (outpainting)
Two tools extend the image instead of fixing it. Zoom Out (often offered at ×1.5 or ×2) zooms out: the model invents what lies around the existing frame — ideal for turning a close-up into a wider scene, or recovering margin when a subject is too tight. Pan extends the image in one direction (left, right, up, down): perfect for building a panorama or clearing space for text (section 3.2). These functions are also called outpainting.
Like Sonia clearing space for her client's logo: you create free space after the fact when you forgot it beforehand — but best practice is still to plan it at generation time (section 3.2).
3. The method: small touches, right area, retries
Three usage rules. Work in small successive touches rather than one big edit: several targeted inpaintings give a far better result than one huge regenerated area. Select an area that is neither too small (the model lacks context to blend cleanly) nor too large (you lose control). And do not hesitate to retry: like any generation, inpainting is probabilistic — the third attempt is often the right one.
Crucial prompt point: during inpainting, the prompt applies to the selected region — describe what you want in that exact spot (material, color, light), consistent with the image, instead of keeping the whole-scene prompt. These functions also exist elsewhere: Photoshop (Generative Fill), Firefly, and Stable Diffusion interfaces, with even greater control.
Did you know?
Fix the cherry, don't rebake the cake.
🔎 Concrete example — 🔎 Concrete example — Sonia's poster fixed locally
The café poster is superb except for the hand and the too-tight framing. Sonia's actions, in order:
1. Save the image (prompt, version, seed) before touching anything.
2. Vary (Region) on the whole hand: area prompt « natural hand holding cup, warm side light » → retry until the fingers look right.
3. Zoom Out to recover margin, then Pan upward for the logo space.Reading: save, fix the area, extend the frame — in that order.
Lesson: preserve everything that worked; never gamble an asset away.
Section vocabulary
- Inpainting
- Regeneration of a SELECTED area of the image, blended with the rest; in Midjourney: Vary (Region).
Simple example: Sonia redoes only the hand holding the cup, without touching her poster's atmosphere.
What it is not: regenerating the whole image because one detail is off.
Reflex: one botched detail on a successful image: local retouching first.
- Outpainting
- Extending the image beyond the existing frame: the model invents what lies around it.
Simple example: Sonia widens her too-tight poster instead of recomposing it.
What it is not: cropping an already too-tight image.
Reflex: framing too tight: extend before cropping.
- Zoom Out / Pan
- Zoom Out zooms out and widens the field; Pan extends the image in one precise direction (left, right, up, down).
Simple example: Sonia pans upward to clear space for the café logo.
What it is not: stretching the image, which distorts the subject.
Reflex: missing text space: Pan in the right direction.
- Small-touch retouching
- Method of chaining several targeted local fixes rather than one big regeneration.
Simple example: Sonia fixes the hand, then widens, then clears the logo: three actions, not one big job.
What it is not: selecting half the image and hoping it works.
Reflex: one fix at a time, checking between each.
In practice — In practice — Fix and extend without redoing everything
- Take an almost successful image and spot THE detail that is off (hand, stray object, clothing) — then save it (prompt, version, seed).
- Use Vary (Region): select the area with context around it, describe what you want IN THAT SPOT, and generate; retry if needed.
- On a too-tight image, apply a Zoom Out to widen the frame and recover margin.
- Use Pan in one direction to clear a space meant for text.
Ready-to-say script — explaining local retouching to a client (30 s)
“This is Sonia: your poster is almost perfect, so I won't redo it — I'll just repair the hand directly inside the image, then widen the frame for your logo. That keeps the atmosphere you approved instead of gambling it away.”
Say this when a client asks you to “start over” for one detail: show them the area before/after.
Copy-paste template — local retouching prompts
Inpainting (AREA prompt, not scene):
« natural hand holding ceramic cup, warm side light from the left, skin texture »
Object to remove:
« clean wooden table surface, same warm light, no object »
Area to clean up:
« simple blurred café background, warm tones, no people »Almost approved client poster
Like Sonia: save, repair the area, extend for the logo — the approved atmosphere stays put.
Portrait with an awkward detail
Vary (Region) on the stray element (deformed jewelry, odd lock of hair) with a sober area prompt, then retry.
Visual too tight for the format
Zoom Out to recover margin, then Pan toward the text area: the format is saved without recomposing.
Stubborn seam
After three retries with no clean blend, switch to Photoshop (clone stamp, Generative Fill) — see 4.4.
✓ Best practices
- Save every successful image (prompt, version, seed) before attempting any edit.
- Describe the AREA in the inpainting prompt (material, color, light), not the whole scene.
- Work in small touches and retry: the third attempt is often the right one.
✗ Common pitfalls
- Regenerating everything for a single detail: you risk losing a happy composition.
- Selecting an area that is too small (visible seam) or too large (inconsistency).
- Keeping the original prompt instead of describing what you want in that exact spot.
Key takeaways
- INPAINTING (Vary Region): regenerate ONE AREA without touching the rest — fix a hand, an object, clothing.
- OUTPAINTING: Zoom Out (widen the field) and Pan (extend in one direction, to clear text space).
- METHOD: small successive touches; area neither too small nor too large; area prompt; retry.
- EQUIVALENTS in Photoshop (Generative Fill), Firefly and Stable Diffusion.
48h challenge — apply tomorrow
Take an almost successful image: save it, fix one local flaw with Vary (Region) (area prompt + at least 1 retry), then try 1 Zoom Out or 1 Pan for text space.
- I can fix one area with Vary (Region) without touching the rest.
- I can extend an image (Zoom Out, Pan) for margin and text space.
- I save my successful images before any edit.
Frequently asked questions
Why does my inpainting retouch blend poorly with the rest of the image?
Four classic causes. The area is too small: the model needs context — include the whole hand, not three fingers. The area is too large: too much freedom brings inconsistency — chain two targeted retouches. The prompt is unsuited: it applies to the region, so describe what you want in that spot (material, color, light). The light or perspective is unspecified: mention the image's light direction. And retry — inpainting is probabilistic. If nothing works, finish by hand (section 4.4).
Is it better to fix by inpainting or to fully regenerate the image?
If the essentials work (composition, light, mood), inpainting wins: a successful image is a fragile asset that rerolling can lose for good. If several elements are botched (framing, mood, subject), fix the prompt and rerun — the problem is upstream. In between (good direction, average execution), variations (section 1.3) are often the best choice. Hierarchy: variations, then inpainting, then a new prompt.
What is inpainting (Vary Region) for?