How to put text behind an image in Photoshop, Canva and Premiere Pro
Six apps, the exact masking step each one needs, and what to do when the cutout is the part that goes wrong.
Free toolText Behind ImagePut a word behind the person in a photo, so they stand in front of it. The subject is cut out here, in your tab, and the picture is never uploaded.Open the editorThe short answer
Every method is the same three moves. Cut the subject out of the photograph, put the text underneath that cutout, and leave the original photograph at the bottom. What changes between apps is only how you make the cutout and what the layer is called.
| App | The step that makes the cutout | Cost |
|---|---|---|
| Photoshop | Select > Subject, then Layer Mask | Subscription |
| Canva | The Text Layer app, or Background Remover on Pro | Account, Pro for one route |
| Illustrator | Image Trace, or place a cutout made elsewhere | Subscription |
| Premiere Pro | A mask on a duplicate clip, tracked forward | Subscription |
| DaVinci Resolve | Magic Mask in the Color page | Free tier does it |
| A browser | A segmentation model that runs on your own machine | Free |
Photoshop, in five steps
This is the version worth learning, because the layer mask is non-destructive: the pixels the mask hides are still there, so you can paint the edge back afterwards without starting again.
Duplicate the photo layer
Cmd J, or Ctrl J. You now have two identical layers. The top one becomes the subject and the bottom one stays as the background.
Select the subject on the top layer
Select then Subject. On a person against a plain background this is usually right first time. Clean it up with Select and Mask if the hair needs it.
Turn the selection into a layer mask
The circle-in-a-rectangle button at the foot of the Layers panel. The top layer is now the subject with nothing around it.
Add the text between the two layers
Click the background layer first so the new type layer lands above it, then set your text. The order from top to bottom must be: subject, text, background.
Position it, then fix the legibility
Drag the text where you want it. If it disappears into the photo, add Layer Style then Drop Shadow with the distance at zero and the size high. That is a halo, not a shadow, and it is what separates the word from the picture.
The mistake almost everyone makes here is putting the text layer at the top and then wondering why it covers the person. Photoshop adds a new layer directly above whichever layer is selected, so select the background before you press T.
Canva
Canva has two routes and they are not equal. The cutout is the only hard part in either.
- The Text Layer app. Open the Apps panel, search for Text Layer, upload the photo and it does the cutout for you. This is what Canva's own page for this effect recommends and it does not need Pro.
- Background Remover. Duplicate the image, select the copy, then press Edit photo and Background Remover, then send the text layer backwards with Cmd [. This is a Pro feature.
Both need a Canva account, and both send your photograph to Canva's servers. That is the trade in this row of the table: a real type system and a licensed font library, in exchange for the picture leaving your machine.
Illustrator
Illustrator has no Select Subject, so there is no quick route. Two things actually work.
- Bring in a cutout you made elsewhere. Place a transparent PNG of the subject, put the type underneath it, and keep the untouched photo at the back. This is the fastest path and it is why the effect is usually built in Photoshop and finished in Illustrator.
- Draw the mask. Pen tool around the subject, select the path and the image, then Object, Clipping Mask, then Make. Accurate and slow.
Premiere Pro and DaVinci Resolve, where the object moves
A video needs the cutout to follow the subject from frame to frame, which is a different problem and the reason no browser tool offers it.
| App | The method | What it costs you |
|---|---|---|
| Premiere Pro | Duplicate the clip on a track above, draw a pen mask around the subject on the copy, then Track Selected Mask Forward. Put the text on a track between the two. | Tracking time, plus hand-correcting the frames where it slips. |
| DaVinci Resolve | Magic Mask in the Color page. Stroke the subject once and it follows them through the clip. Free tier included. | Much less. This is the best free route for video by a distance. |
| After Effects | Roto Brush 2, which is the same idea with a better edge on hair. | Subscription, and render time. |
If you meant Word, PowerPoint or Google Slides, it is a different job
Two questions share this phrasing and they have nothing to do with each other. Everything above is the depth effect: cutting a subject out of a photograph so type can pass behind it. The other one is putting a picture underneath a text box in a document, which is layer order and takes two clicks.
It is worth naming because the second question is asked far more often than the first. If that is the one you have, you do not need any of the masking above.
| App | What to do |
|---|---|
| Microsoft Word | Click the picture, open Layout Options, and choose Behind Text under With Text Wrapping. The text then flows over it. |
| PowerPoint | Right-click the picture, then Send to Back. Put the words in a text box with no fill on top of it. |
| Google Slides | Right-click the image, then Order and Send to back. There is no text wrap, so the words go in a separate text box. |
| Google Docs | Click the image, then the Behind text icon in the toolbar that appears under it. Docs only gained this in 2020; older advice tells you to use a drawing, and you no longer have to. |
None of these cut anything out. The picture goes behind the whole text box, so the words cross the entire image rather than passing behind one person in it. If that is what you wanted, go back to the top of this page.
The iPhone, and why it is not exportable
iOS 16 put the lock screen clock behind the subject of your wallpaper, and that is where most people first saw this effect. It is a system feature: the depth mask is applied to the lock screen and there is no file at the end of it.
What you can do on a phone is lift the subject out. Press and hold on the subject in Photos until it lights up, then Copy, and paste that cutout into any app that takes a transparent PNG. From there it is the same three-layer sandwich as everywhere else.
What the cutout actually gets right, measured
The masking step is where this effect succeeds or fails, and the claims made for it are mostly unmeasured. So here are two browser segmentation models run over the same five photographs, with the same word placed across each subject, and every result looked at rather than scored by a formula.
Run on 4 September 2026. MacBook Pro, Apple M1 Max, 32 GB, macOS 26.6.2, Chrome 152.0.7977.65. Every photo 1200x800 JPEG. The models are MODNet fp16 (a human matting network, listed in the tool as People) and ormbg fp16 (a general segmentation network, listed as Anything else).
| Photograph | People, 13 MB | Anything else, 84 MB |
|---|---|---|
| Close portrait, backlit, loose hair and a bunch of flowers | Correct. Held the loose strands and the flower stems | Correct. Slightly softer along the hairline |
| Person on a wall, seen from behind against sky and water | Correct | Correct |
| Small hiker with a backpack, hazy backlight, mountains | Correct, including the pack straps | Correct |
| Woman from behind in a busy street, several other people in shot | Correct on the near subject only. The two people further back stayed behind the text | The same. Only the near subject came forward |
| Leopard walking on a dirt track | Nothing found. The word passed straight over the animal | Correct, including the ears and the whiskers |
Two things in that table are worth more than the rest. The small model is not worse on a leopard, it is blank: it was trained to find a person, it found none, and it returned nothing at all. And on the street photograph both models agreed that one person was the subject and the other two were background, which is a judgement rather than an error, but it is a judgement you cannot overrule.
What each model costs to run
Byte counts read straight out of the browser's Cache Storage after a run, and download times measured with a forced re-fetch on the same connection.
| File | Bytes | Downloaded in | Cutout time, per photo |
|---|---|---|---|
| MODNet fp16, People | 12,984,781 | 2.57 s at 40.5 Mbps | 1.53 s median, 1.525 to 1.580 |
| ormbg fp16, Anything else | 88,117,930 | 10.66 s at 66.1 Mbps | 8.65 s median, 8.595 to 8.704 |
| ONNX Runtime WASM, shared | 23,567,050 | 0.32 s | Loaded once, whichever model you pick |
So the general model is 5.6 times slower per photo and 6.8 times bigger to fetch, for a result that was no better on any of the four photographs with a person in them. That is the whole reason a tool should ask which one you want rather than picking the bigger one and calling it better.
The same photo through four browser editors
Same machine, same day, same 1200x800 portrait, timed in the page from the moment the file was handed to the editor. Two of the seven editors we opened would not take a photo at all without a Google account, so they are not in this table.
| Editor | First photo | Second photo | Where the model runs |
|---|---|---|---|
| text-behind-image.org | 2.23 s | Not measured | Browser |
| floi, People model | 2.16 s in a fresh tab | 1.53 s | Browser |
| textbehindimageai.com | Over 59 s, after an 84.07 MB download | Not measured | Browser |
| textbehindimage.com | 71.96 s | 38.20 s | Browser |
The honest reading of that table is that the top two are level. text-behind-image.org finished its first photo in 2.23 seconds and floi finished its first in a fresh tab in 2.16, which is inside the noise. The gap that matters is the bottom two, and it is a factor of twenty-five.
One more number from the same session, because it explains where a lot of that time goes. The editor ranking first for this phrase requests 229 Google Fonts stylesheets on load, 11.02 MB of the 12.22 MB it pulls across 250 requests, before you have dropped anything in.
The three things that go wrong, and the fix for each
The word vanishes into the photo
Tone, not colour. Put a soft halo behind the letters in the opposite tone: a dark glow under a white word, a light one under a dark word. In Photoshop that is a drop shadow with distance 0 and size high.
A rim of old background came with the subject
Contract the mask by a pixel or two. Photoshop does it with Select and Mask and the Shift Edge slider. A browser tool that offers an Edge control is doing the same thing to the same matte.
The hair looks cut out with scissors
Feather the boundary by two or three pixels. Hair has no hard edge in the photograph, so a hard edge in the mask is the thing that reads as fake, not the shape.
Which route to take
| If you | Use |
|---|---|
| Want one image, now, and would rather not upload the photo | A browser editor that runs the model locally |
| Need to paint the mask by hand, or the subject is unusual | Photoshop |
| Are already working in a Canva document | Canva's Text Layer app |
| Have a video | DaVinci Resolve's Magic Mask, free tier |
| Want it on your lock screen only | The iPhone's own depth effect. No export |
Frequently asked questions
Which Photoshop version added Select Subject?
Select Subject arrived in Photoshop CC 2018 and the cloud-powered version that handles hair well arrived in 2021. Before that the job was the Quick Selection tool and a manual Refine Edge pass, which is the reason so many tutorials for this effect are ten minutes long. If your Photoshop has Select then Subject in the menu bar, you have the fast route.
Can I do this in Canva without a Pro subscription?
Not with the Background Remover, which is a Pro feature. The free route is the Text Layer app in Canva's app panel, which does its own cutout, and Canva's own feature page for this effect points at it rather than at Background Remover. Either way you need a Canva account and your photo goes to Canva's servers.
Why does my text look wrong even though the cutout is right?
Because the letters and the photograph behind them are too close in tone, which is a legibility problem rather than a masking one. It is the single most common reason this effect looks amateur, and the fix is not a different colour: it is separation. Add a soft dark halo behind a light word, or a thin outline in the opposite tone. Photoshop does this with a layer style, and most free browser editors cannot do it at all.
How do I put text behind a moving object in a video?
You need masking with tracking, not a single cutout. In Premiere Pro that is a mask on an adjustment or duplicate clip with Track Selected Mask Forward; in DaVinci Resolve it is the Magic Mask in the Color page, which follows a person or an object across the clip on its own. Both take minutes rather than seconds, and neither is something a browser tool can do.
Is the iPhone depth effect the same thing?
It is the same illusion built a different way. iOS finds the subject in your lock screen wallpaper and lets the clock sit behind it, but that is a system feature applied to the lock screen rather than an image you can export. To get a file you can post, you have to build the effect yourself in an editor.
How do I put an image behind text in Word or Google Slides?
That is layer order rather than masking, and it takes two clicks. In Word, click the picture, open Layout Options and choose Behind Text. In PowerPoint and Google Slides, right-click the picture and Send to Back, then put the words in a text box over it. In Google Docs, click the image and press the Behind text icon in the toolbar under it. None of these cut anything out, so the words cross the whole picture rather than passing behind one person in it.
Does the cutout have to be perfect?
Less than you would think, because most of the boundary is hidden. The text only crosses part of the subject, so only that part of the edge is ever visible, and a soft two-pixel feather hides a lot. Where it does matter is hair against a bright sky, which is the one case worth zooming in on before you export.