I had a deadline problem last March. A mid-sized e-commerce client sent over 340 apparel shots that needed background replacement, skin smoothing, and basic color grading, all due in 48 hours. My usual Photoshop action stack handled the color grading and export prep without complaint. But the background removal on fabric edges, lace, sheer sleeves, that part was going to eat the entire timeline. A colleague had been pushing me to try Luminar Neo’s background AI for months. I finally installed it at 11pm and ran my first test batch by midnight.

What followed was six months of genuinely mixed results, and that’s what I want to walk through here.

What Luminar Neo Is Actually Doing Under the Hood

Luminar Neo isn’t a filter. It’s a host application that also runs as a plugin inside Photoshop via the Extensions panel. When it processes a background removal or a sky replacement, it’s running a neural network trained on millions of segmented images. The difference between this and Photoshop’s native Remove Background (which uses Adobe Sensei) is architecture depth and edge-case training data.

Photoshop’s built-in tool is fast and works well on high-contrast edges. Hair against a white seamless, product shots on gray, it handles those cleanly. Where it starts breaking down is complex textures against visually similar backgrounds, fine fabric weave, flyaway hair against busy environments, reflective surfaces. Luminar Neo’s segmentation model appears to have been trained more heavily on exactly these cases, which is why fabric edges on my apparel test came out cleaner without manual refinement.

The tradeoff is processing overhead. Luminar Neo running as a Photoshop plugin introduces latency that native tools don’t. On my workstation (AMD Ryzen 9 5900X, 64GB RAM, RTX 3080), a single background removal through Luminar Neo takes about 8 to 12 seconds. Photoshop’s native version takes 2 to 4 seconds. That gap matters when you’re running batch operations.

The Batch Problem Nobody Talks About in Review Videos

Here’s where most plugin reviews fail you: they show you one image processed in real time and call it efficient. Nobody sits with 300 files and measures actual throughput.

Luminar Neo does not support Photoshop scripting or Actions in any meaningful way when running as a plugin. You cannot record an Action that triggers Luminar Neo’s AI tools and then batch that Action through Image Processor or a Droplet. The plugin is interactive by design. This means every single image requires a manual trigger inside the Luminar Neo panel.

For my 340-image apparel job, that cost me roughly 90 minutes of click-and-confirm time that my normal action-based workflow would have automated entirely. I’ve been tracking time savings from automated workflows for years now, and manual plugin triggering across a large file set is a real regression, not a minor inconvenience.

If your volume is under 30 images per session, this won’t sting much. If you’re running the kind of production load where batch automation is the reason you built your workflow in the first place, factor this in before you pay the $99 per year subscription.

Where It Genuinely Earns Its Place in the Stack

Luminar Neo’s Relight AI is the tool I’ve actually kept in regular rotation. It simulates 3D lighting adjustments by analyzing depth information in the image and applying luminosity changes that respect foreground and background separation. For product compositing where the client has changed the background after the shoot (which happens more than it should), Relight AI gets you 80% of the way to plausible lighting in under a minute.

The Portrait tools, specifically the Face AI and Skin AI modules, are also legitimately good on beauty and lifestyle work. The Skin AI texture smoothing avoids the plastic look that frequency separation shortcuts can produce when applied too aggressively. It reads skin texture and smooths selectively rather than globally. On a campaign I delivered for a Chicago-based wellness brand earlier this year, I used Skin AI for the initial pass and then did targeted dodge-and-burn over the top. Total retouching time per image dropped from about 25 minutes to roughly 14 minutes. On a 60-image shoot, that’s 11 minutes per image saved, or about 11 hours returned to my schedule.

How to Integrate It Without Letting It Break Your Existing Workflow

The setup that works for me is using Luminar Neo for a specific, bounded phase rather than running it throughout the full pipeline. I open my layered PSD, run whatever Luminar Neo processing I need via the plugin panel, flatten that result to a stamped layer (Ctrl+Alt+Shift+E), and then pass control back to my Photoshop action stack for the remaining steps. Color grading, sharpening, export sizing, watermarking, all of that still runs through Actions as it always has.

This keeps Luminar Neo’s output inside a container that my existing automation can still handle. It also means if Luminar Neo’s processing isn’t right on a particular image, I haven’t committed to it in the file structure. The stamped layer is non-destructive relative to everything below it.

One practical note: Luminar Neo outputs back to Photoshop as a Smart Object by default. Convert it to a pixel layer before stamping or you’ll add unnecessary file weight. A typical 24-megapixel file processed through Luminar Neo and returned as a Smart Object can sit at 180 to 220MB. Rasterize it first and you’re back to a manageable 40 to 60MB range.

The Honest Verdict After 180 Days

Luminar Neo is a capable tool for specific retouching tasks, and a poor fit for batch-heavy production pipelines where automation is the entire point. The Relight AI and Skin AI modules solve real problems that Photoshop’s native tools address clumsily or not at all. The background AI is slightly better than Remove Background on complex edges but not so dramatically better that it justifies the workflow friction for high-volume jobs.

At $99 per year, it’s not a budget question, it’s a use-case question. Know exactly which part of your work it’s solving before you commit, because the tool rewards specificity and punishes vague optimism.