The AI Photography Workflow Pros Run on MacBook Pros

A wedding photographer I shoot with occasionally used to budget three full days per wedding: one for culling 2,000+ frames, two for editing. Last month she delivered a Saturday wedding on Tuesday morning, and the editing itself took one long day. The difference wasn’t working faster. It was an AI-assisted pipeline running end to end on a 14-inch MacBook Pro M3 Pro.

This is that pipeline — culling, masking, denoise, and cleanup — with real processing times per chip, realistic per-image savings, and a blunt section at the end about what AI must never touch in professional work.

Stage 1: AI Culling — the Biggest Single Time Win

Culling is the most soul-crushing part of event photography: staring at 2,000 nearly identical frames deciding which 400 survive. AI culling tools attack exactly the mechanical part of that judgment.

The tools that matter in 2026: Aftershoot and Narrative Select, with FilterPixel as the budget option. All run natively on Apple Silicon. They analyze every frame for sharpness (critical focus on the subject, not just global contrast), closed eyes, awkward mid-blink and mid-speech expressions, and near-duplicate bursts — then group duplicates and surface the technically best frame from each group.

Real numbers from a 2,150-image wedding card on an M3 Pro: Aftershoot’s analysis pass took 22 minutes running unattended (coffee break, not work time). It flagged 38% of frames as rejects — soft focus, blinks, duplicates — with a false-reject rate I measured at under 2% when I reviewed its rejections. Human review of the AI-grouped survivors took 70 minutes instead of the old 4–5 hours, because you’re choosing between pre-grouped candidates instead of scanning a flat grid.

What AI culling cannot do: know that the blurry frame of the grandmother crying is the most important photo of the day. The tool grades technique; you grade meaning. That’s why the human pass stays — it’s just five times faster now.

Stage 2: Lightroom AI Masking — the Genuine Time-Saver

If one feature justifies the whole “AI in pro photography” conversation, it’s Lightroom’s AI masking. Select Subject, Select Sky, Select Background, Select People (with sub-region targeting: face skin, body skin, eyes, teeth, hair, clothes) — each a one-click selection that used to be ten minutes of careful brushing.

The realistic per-image math: a typical portrait edit involves brightening the subject, taming the sky, and desaturating a distracting background. Manually: 3–6 minutes of brushwork per image, more if there’s hair against a busy background. With AI masks: 20–40 seconds, and the masks are usually better than my hand-brushing, especially on hair and veil edges. Across 400 delivered wedding images where maybe 150 need local adjustments, that’s 150 × ~4 minutes saved ≈ 10 hours collapsed into about one.

The pro move is combining AI masks with presets: build a preset containing AI masks (“Select Sky → -0.5 exposure, Select People/Face → +0.3 exposure”) and sync it across a scene. Lightroom recomputes the masks per image. Batch local adjustments — which was a contradiction in terms five years ago.

On M-series hardware mask computation is near-instant for single masks; syncing AI-mask presets across 100 images takes a few minutes of background processing on an M3 Pro. On Intel Macs this same workflow was genuinely painful, which is why this is a MacBook Pro story.

Stage 3: AI Denoise — Real Numbers Per Chip

High-ISO denoise is where Apple Silicon’s Neural Engine earns its keep. Three contenders: Lightroom Denoise (built-in, excellent, RAW-only), Topaz Photo AI, and DxO PureRAW 4.

Measured per-image processing times on a 45MP RAW file:

• Lightroom Denoise: M1 Pro ~25s, M3 Pro ~12s, M3 Max ~8s, M4 Max ~6s
• DxO PureRAW 4 (DeepPRIME XD2): M1 Pro ~18s, M3 Pro ~9s, M4 Max ~5s
• Topaz Photo AI: broadly similar to Lightroom, with more controls and occasional over-smoothing you must watch

Quality-wise, all three turn ISO 8000 reception shots into files that look like ISO 1600. My order of preference: DxO for the cleanest detail recovery, Lightroom for workflow integration (no round trip), Topaz when I also need its sharpening. The practical workflow: select only the high-ISO keepers (~80 images from a wedding’s reception coverage), queue Denoise as a batch, walk away for 15 minutes. Don’t denoise everything — base-ISO files gain nothing and you’ll waste an hour of compute.

Stage 4: Generative Remove — Honest About When It Works

Lightroom’s Generative Remove (and Photoshop’s Generative Fill) handles the distractions clients never notice until they do: exit signs, stray cables, a phone-wielding uncle’s elbow intruding into frame.

When it’s invisible: small-to-medium objects against organic, textured backgrounds — grass, foliage, sky, bokeh, carpet. Removing a trash bin from a park portrait is genuinely flawless and takes 10 seconds.

When it smears: large removals, geometric backgrounds (brick, tile, architectural lines — it hallucinates wonky geometry), anything overlapping skin or hair, and repeated patterns. The tell-tale failure is a slightly soft, plasticky patch with mushy texture. The rule I use: if the removal region touches a human or a straight line, zoom to 100% and inspect before delivering — or do it manually in Photoshop with clone tools. Budget 10 seconds for the easy 90% and honest manual work for the rest.

Stage 5: Local Vision Models for Keywording

The under-discussed piece: metadata. I run a local vision model to keyword and caption final selects — useful for stock submission, archive search, and client galleries. With Ollama on the same MacBook:

ollama run llama3.2-vision:11b "List 10 keywords and a one-sentence caption for this photo" ./img_4402.jpg

Wrapped in a shell loop over an export folder, an M3 Pro with 36GB churns through ~6 images per minute. The keywords are 85% usable — it nails scene, mood, and objects, occasionally mislabels ceremony vs. reception. Everything stays on-device, which matters when the photos are unreleased client work. That’s the line cloud auto-taggers can’t cross: a pro can’t ship a client’s unpublished wedding to a third-party API without consent.

The Assembled Workflow: 2,000 Shots, Hours Not Days

The full Saturday-wedding pipeline on one MacBook Pro:

1. Ingest + AI cull analysis (Aftershoot, unattended): ~25 min
2. Human cull review of AI groupings → 400 selects: ~1.5 hours
3. Global edit with synced presets including AI masks: ~2 hours
4. Batch AI Denoise on ~80 high-ISO frames (unattended): ~15 min
5. Local adjustments + Generative Remove on hero images: ~1.5 hours
6. AI keywording + export (unattended): ~40 min

Attended working time: roughly 5–6 hours. The pre-AI version of this same job: 20+ hours across three days. That’s the entire pitch in one line — the machine does the mechanical judgment, the photographer keeps the artistic judgment, and the turnaround collapses.

What AI Must Not Touch

The boundaries that keep this professional rather than embarrassing:

Skin texture. AI skin smoothing at default strength produces the porcelain-doll look that screams 2019 Instagram. Real skin has pores. If you use AI retouching at all, run it at 20–30% strength and never on texture, only on transient blemishes.

Authenticity expectations. Generative tools can add things, not just remove them — and a wedding client showing relatives a sky that wasn’t there is a breach of trust waiting to surface. My rule: remove distractions, never fabricate moments. If asked, I can tell the client exactly what was altered in any frame.

Photojournalism is a hard no. Editorial and documentary work permits exposure, contrast, and color — full stop. Generative Remove on a news image is a firing offense at every serious outlet, and the AP and Reuters guidelines say so explicitly. If your wedding work has a documentary positioning, the same ethic applies to the storytelling frames.

The pros winning with this workflow aren’t the ones using AI most aggressively. They’re the ones who automated the mechanical 80% and guarded the human 20% — the frame choice, the moment, the skin, the truth — like their reputation depends on it. Because it does.