Last month a candle maker sent me a phone photo of an amber jar candle sitting on her kitchen counter. She had no Photoshop, no design training, and fifteen minutes before she needed to publish an Etsy listing. We ran it through Pixlr AI together and she had a clean, lit-from-the-left product shot on a concrete surface by the time her coffee went cold.
That’s the honest pitch for Pixlr’s generative tools: they’re not as deep as Photoshop’s, but they’re fast, they run in a browser tab, and the free tier covers most small-business image work. Here’s the exact sequence I use, with the clicks that matter and the ones that waste your time.
Set Up Before You Touch Any AI Tool
Open Pixlr E rather than Pixlr X. The X version is the stripped-down quick editor and it hides most of the generative panels. E gives you layers, which you will need the moment an AI edit comes back slightly wrong.
Two habits at the start:
- Duplicate your base layer (Ctrl+J or Layer > Duplicate). Every AI operation in Pixlr flattens onto whatever layer is active. Keep the original untouched so you can always go back.
- Work at the largest size you have. Generative Fill samples surrounding pixels, so a 4000px original gives you far cleaner results than a 1200px one you’ll later upscale.
If your source image needs colour or exposure correction, do that first. AI tools inherit problems. Feeding in a blue-cast photo and asking for a warm background produces a fight between the two.
Step 1: Cut Out Your Subject
Go to Cutout (under the Cutout/AI tools menu) and let it run. It handles high-contrast subjects well: jars, bottles, shoes, mugs, electronics, anything with a clean edge against a plain-ish background.
Set feather between 1 and 2 pixels. Zero feather gives you a razor edge that looks pasted on; anything above 3 starts eating into the product itself.
Where Cutout struggles: hair, fur, sheer fabric, glass with visible background behind it, and anything the same colour as its backdrop. A white teapot on a white table will come back with holes. For those, shoot on a coloured surface or plan to clean up manually with the Erase brush at 200% zoom.
The edge check nobody does
Zoom to 200% and drag along the outline before you move on. Look for leftover background fringing in the corners of rectangular objects. Two minutes with a soft Erase brush here saves you from a very obvious halo once the new background goes in. This is the single biggest quality difference between an edit that reads as professional and one that reads as a weekend experiment.
Step 2: Generate a Background That Matches the Light
With your cutout on its own layer, open the AI image generation panel and describe the scene. The trick is that the prompt should describe photography conditions, not just objects.
Weak prompt: “nice background for a candle.”
Strong prompt: “weathered concrete surface, soft window light from the left, shallow shadow cast to the right of the object, muted warm neutrals, low camera angle.”
That formula works consistently: surface + light direction + colour mood + camera angle. If you skip the light direction, the AI invents its own and your subject will look like it’s sitting in a different room. For a deeper walkthrough of the generative panels and their quirks, this guide to using the browser editor’s generative tools like a pro is worth reading alongside this workflow.
Generate to a new layer every time. You’ll want to toggle it on and off to compare against the original, and you’ll often need two or three attempts before one lands.
Step 3: Expand the Frame Instead of Reshooting
Generative Expand is the most underrated tool in the set. It grows your canvas and invents plausible content to fill the gap.
Real uses:
- Turning a square Instagram crop into a 16:9 blog header
- Adding headroom above a product so you can drop in a text block
- Extending a plain surface so a wide banner doesn’t show obvious tiling
Expand in small increments, maybe 15% at a time, and check each result. Big single jumps tend to produce smeared textures and repeated patterns. Also keep the expansion on plain areas: extending a patterned tile floor or a bookshelf will expose the AI’s weakness for repeating motifs badly.
Step 4: Clean Up With Remove Object
Before finishing, hunt for the small stuff. Cords, a stray price sticker, dust on a dark surface, a distracting bright object in the corner. The Remove Object tool handles all of these in seconds, and each removal makes the composition read more deliberately.
On textured backgrounds, brush slightly larger than the object and let the tool blend. On flat backgrounds it’s almost invisible. This step takes about ninety seconds and is usually the difference between “a photo” and “a product shot.”
A Real 12-Minute Edit, Minute by Minute
Here’s how the candle jar actually went:
- 0:00 to 1:30 Uploaded the 3024px phone photo, duplicated the layer, nudged exposure +8.
- 1:30 to 4:00 Ran Cutout, feathered at 1.5px, then spent two minutes cleaning faint white fringing along the jar’s left edge at 200% zoom.
- 4:00 to 7:30 Generated three backgrounds with the concrete prompt above. Attempt one was too warm, attempt two had an odd shadow, attempt three worked. Kept it on its own layer.
- 7:30 to 9:00 Painted a soft black shadow under the jar on a new layer at 18% opacity to anchor it to the surface. This single step did more for realism than the AI background did.
- 9:00 to 10:30 Removed a stray reflection and a bit of label glare with Remove Object.
- 10:30 to 12:00 Expanded the canvas to 16:9, exported as JPG at quality 90.
Nothing here required a subscription. The whole thing ran in a browser on a four-year-old laptop.
Where Pixlr AI Still Falls Down
Be honest about the limits so you don’t burn an afternoon fighting them.
Text inside generated images is unreliable. If you need the word “Amber” on a label, add it in Pixlr’s text tool after generation, not in the prompt. The same goes for logos and any kind of signage.
Hands and faces in generated content go wrong. For anything with a person in the background, generate the scene empty and composite a real photo of a person in if you must. Or don’t include people at all.
Patterns repeat. Tiles, brick, fabric, foliage in the distance. Expand less and crop more.
Resolution ceilings exist. Free-tier exports and generated fills cap out lower than you’d want for print. For web and social, you’re fine. For a 24-inch poster, you’ll need a different toolchain.
There’s also a category mismatch worth knowing about. Pixlr AI is a photo editor with generative features bolted on, not a from-scratch image generator. If you actually want original illustration rather than photo manipulation, the approach is different enough that this practical guide to creating art with generative tools will steer you better than any retouching workflow.
Habits That Separate Good Results From Obvious AI Edits
Three things I check on every single edit:
Light direction on both layers. Squint at the image. Do the highlights on your subject agree with the highlights in the generated background? If your jar is lit from the right and the concrete is lit from the left, the composite will feel wrong even to people who can’t articulate why.
Contact shadows. Objects float when they don’t touch the ground visually. A soft brush, low opacity, one pass under the base. That’s it.
Colour temperature. Generated backgrounds often come back slightly cooler or warmer than your original photo. Nudge the background layer with Color Balance by 3 to 6 points toward the subject’s cast. It’s a tiny move that makes everything sit together.
Then export. PNG if you need transparency for a transparent-background product cutout you’ll reuse across mockups; JPG at quality 85 to 92 for everything else. Save the layered file too, because the next product in the range will almost certainly want the same background and the same shadow treatment.