Image to editable PSD with text layers
Turn the text in a flat image into Photoshop type you can double-click and retype.
Layers
Resolution
Auto finds every element (up to 16 per pass). Billed per output: at most 17 credits a pass (17 images). A busy image runs up to 4 passes, and you can stop between them. Takes about 56–138 s.
New here?
11 images · 88 s · 11 credits
Background
Memory-foam cushions label
40mm dynamic driver label
Hybrid ANC mics label
60-hour battery label
Hybrid ANC microphone assembly
60-hour battery
40mm dynamic driver
Inner headphone housing
Upload an image on the left to split your own.
Layers
PSD · 11
NEW tagfx
BLOOM main title textfx
SPRING COLLECTION 2026 bottom textfx
Eau de Parfum subtitle textfx
Floating pink petal
BLOOM perfume bottle- +4
Background
Short answer
Most converters leave text baked into the picture. LayerPSD splits the image first, so every line of text is its own layer — then writes it into the PSD as real type, with its stroke, glow, shadow and gradient as layer styles. In our test on 30 designs, 99% of 206 lines of regular-size text came out this way with the words read right.
What carries over into the type layer
- Words
- Read from each text layer — Latin, Japanese, Korean and Chinese scripts are supported. Right-to-left text such as Arabic stays a pixel layer.
- Size and position
- Measured from the layer’s own pixels, so the type sits where the lettering was.
- Color and gradient
- Solid fill, or a top-to-bottom gradient as a Gradient Overlay style.
- Stroke, glow, shadow
- Written as Photoshop layer styles you can adjust or switch off.
- Font
- Matched by style (sans, serif, rounded, handwritten, display), not the exact typeface — pick your font after opening.
Tested on 30 designs
We run the same 30 designs (posters, flyers, social graphics, app screens, slides, menus, and three photos with no text) through the text pass 2 times and count how many of their 206 lines of regular-size text come out on their own layer with the words read right. Last run 2026-10-05.
99%
Lines on their own layer, with credits
409/412 · 2×206
99.9%
Characters read correctly
2×206
97%
Lines on their own layer, on the free try
400/412 · 2×206
| Kind of design | Designs | With credits | Free try |
|---|---|---|---|
| Posters and flyers | 8 | 66/66 | 63/66 |
| Social graphics and banners | 9 | 121/122 | 117/122 |
| App and game screens | 6 | 168/170 | 166/170 |
| Slides, menus and infographics | 4 | 54/54 | 54/54 |
| Photos and art without text | 3 | No text, nothing lifted | |
Headlines taller than 8% of the image (28 lines in this set) stay in the picture on purpose, so the image model paints the background behind them; “PSD with editable text” reads them when you export, and that step is not scored here. Per-design scores (JSON, CC BY 4.0)
Text-heavy designs: flyers, menus, slides and app screens
Every line is read in one pass before the split, so a dense layout is not merged into one block or left to the image model. In the test set, app and game screens came out 168 of 170 lines and slides, menus and infographics 54 of 54 (two runs, with credits). On a 120-line order-history screen made as a stress test, 115 to 120 lines were read exactly. What stays as pixels: lettering on a curve or in perspective, and text printed on a product or inside a photo.
Flash or 5.5? The same three images on both models
On 2026-11-13 we split a café flyer, a festival poster and a sales slide with Seedream 5.0 Flash and with Seedream 5.5, using the same settings and the same text pass before the split. Credits are what each run actually used.
| Image and settings | Model | Layers | Editable text lines | First pass | Credits |
|---|---|---|---|---|---|
| Café flyer1K · 4 layers | Seedream 5.0 Flash | 8 | 4 | 44 s | 5 credits |
| Café flyer1K · 4 layers | Seedream 5.5 | 8 | 5 | 43 s | 10.2 credits |
| Festival poster2K · Auto | Seedream 5.0 Flash | 36 | 7 | 106 s | 31 credits |
| Festival poster2K · Auto | Seedream 5.5 | 22 | 8 | 85 s | 33.9 credits |
| Sales slide2K · Auto | Seedream 5.0 Flash | 27 | 17 | 71 s | 11 credits |
| Sales slide2K · Auto | Seedream 5.5 | 18 | 13 | 55 s | 12.1 credits |
Which one to pick
- Pick 5.5 when a plain headline should open as type: on the flyer it turned “MORNING BREW” into an editable text layer. Flash returned that headline as a picture, which “More formats → PSD with editable text” can still read, one step later.
- Pick Flash for slides, screenshots and busy posters on Auto: a layer costs 1 credit instead of 1.9, its second pass split the poster into 36 layers where 5.5 stopped at 22, and slide text is read the same way on both.
- Neither model turns 3D or neon lettering into type: “NEON NIGHTS” stayed a picture on both, and 5.5 marked a single letter cut from that artwork as text.
Smaller lines of text are lifted by the same reader before the split on both models, so the slide’s line counts (17 and 13) vary from run to run, not by model. Times are the first pass; credits include every pass.
Opening the file
Type is rebuilt, not recovered: every line the model can read becomes a native text layer, and anything it cannot read stays as pixels. These four steps get you from upload to editable words.
- 1
Split the image into layers on the home page.
- 2
Choose “PSD with editable text” in the export menu.
- 3
Open the file. When Photoshop asks, choose Update so it re-renders the type.
- 4
Each text group holds the original lettering (pixels) and a hidden editable type layer. Turn the type layer on, hide the pixels, and edit.
File export itself has no fee. Supplementary OCR, when needed, typically uses about 0.1 credit; completed readings can be reused while the result is unchanged and its cache remains available. Splitting is billed per layer — see pricing.
Questions
Short answers on splitting, export and billing. Anything else, email us and a person replies.
How do I make the text in an image editable?
Split the image so each block of text is its own layer, then export “PSD with editable text”. Every text layer that can be read becomes a Photoshop type layer; the rest stay as pixels.
Will it use the exact same font?
No. It picks a font of the same kind (for example a bold sans or a serif) and keeps the size, color and effects. Swap in the original typeface in Photoshop if you have it.
Which text stays as pixels?
Lettering the model cannot read confidently — heavily stylised logos, text on a curve or in perspective, tiny print — is kept as the original pixel layer, so nothing disappears.
Does it work with Photopea or Affinity?
The PSD is a standard RGB file. Type layers and layer styles open in Photopea; support for editing PSD type varies in other apps, where the pixel copy is always there.
How accurate is the text recognition?
On our 30-design test set (2026-10-05), 99% of 206 lines of regular-size text came out on their own layer with the words read right, and 99.9% of characters were correct. The free try uses a lighter model and scored 97%. On about 6% of lifted lines a trace of the old lettering was still visible in the background and needs a quick touch-up. Per-design scores: layerpsd.com/data/text-benchmark.json.
Does it work on text-heavy designs like menus, slides and app screenshots?
Yes, they are part of the test set: app and game screens scored 168 of 170 lines, slides, menus and infographics 54 of 54 (two runs each). Every line becomes its own type layer, so a 20-line menu opens as 20 editable lines.