Use Dreamina to try a short wording change in a finished image: upload the original, name the exact old and new text, then inspect the downloaded result. If you still have the original design with a live text box, edit that box directly. If you need to keep updating a flattened picture, extract or rebuild its lettering as editable text.
The file you have determines the work. A JPG, PNG or screenshot usually contains pixels where the letters appear; it does not retain the original font settings or text box. Adding another text box on top leaves the old lettering underneath unless you remove or cover it.
This guide covers the complete change, from identifying the right route to saving the image and its reusable source.
- Step 1. Identify what you can actually edit
- Step 2. Prepare the exact change
- Step 3. Replace a short phrase in Dreamina
- Step 4. Create editable text from a flattened image
- Step 5. Remove and rebuild lettering manually
- Step 6. Handle the cases that need a different approach
- Step 7. Check the download and retain a useful master
- Troubleshooting
- Free access and practical costs
- FAQ
Step 1. Identify what you can actually edit
Open a copy of the file and try selecting the wording. In a design project, you may be able to place a cursor inside a text box. In a layered document, look for a Type layer, commonly marked with a T. If selecting the object only selects the whole picture, treat the lettering as part of a flattened image.
A PSD extension is not proof that every word remains live: some layers may already be rasterized. Similarly, uploading a PNG into a design editor does not recreate its original text boxes.
Finish this step with: a chosen route and an untouched original. If someone can supply the original editable project, obtain it before rebuilding a complicated poster.
Step 2. Prepare the exact change
Copy the current wording and replacement into a short note. Include punctuation, capitalization, dates, units and line breaks. Identify surrounding elements that matter: another heading, a logo, a face, a diagram or a precise grid.
Check whether the new wording will fit. Replacing “May” with “September” changes the layout even if every letter is correct. Decide whether to widen the text area, wrap the line or shorten approved wording. Shrinking all the type can make the corrected image harder to use.
For a recurring event graphic, confirm all dependent information together: date, weekday, time, venue and registration detail. Do not change only the largest date while leaving a smaller footer with the old one.
Finish this step with: an approved replacement string and a clear idea of what must remain visually unchanged.
Step 3. Replace a short phrase in Dreamina
Open Dreamina’s AI photo editor and choose an available image-editing workflow. Attach the highest-quality original as the reference. For a correction to one image, check that only the intended reference is attached.
Name the substitution explicitly. The recorded fictional exhibition example used this instruction:
Change only “NOV 14–16, 2026” to “NOV 21–23, 2026”. Preserve every other letter, the grid, paper sculpture, circle, background and spacing.
That is the instruction submitted for this particular example, not a universal guarantee of preservation. For your own image, replace both quoted strings and name the elements that matter. Choose the original aspect ratio and an available suitable output setting, then generate.
What the existing date example shows
The valid single-reference Seedream 5.0 Pro result changed the date to NOV 21–23, 2026, removed the old date and kept the other four text groups visually consistent. The input and downloaded image were both 1792 × 2240.
For a closer view, open the original poster and downloaded date edit.
The sculpture, grid and spacing remained visually similar, but pixels outside the date area also changed. This was a visual correction in a fictional design, not a pixel-identical patch or an actual exhibition announcement. It also did not create editable text layers. The result came from a paid-account workflow and does not establish Free access to that model.
Inspect before accepting
Read the replacement character by character, then inspect every other text group. Compare the full design and its small display version. Check for changed names, accidental extra words, altered logos, shifted alignment and remnants of the original date.
If the replacement is wrong or other content drifts, return to the original. Choose a simpler edit or one of the layer-based routes below. Repeatedly editing a changed copy can compound errors.
Finish this step with: a checked image for a one-time correction, or a decision to rebuild editable lettering when further updates matter.
Step 4. Create editable text from a flattened image
Text extraction can help when the lettering is clear and you need future revisions. It recognizes the visible characters, reconstructs the area underneath and creates new editable type. It does not recover an exact hidden original design.
Canva Grab Text
Upload the image, select it and open the image-editing tools. Choose Grab Text where available, select the wording to extract and apply the operation. Read the recognized copy before changing it, then adjust the replacement’s font, size and position.
Use straight, clear lettering where possible. Perspective, decorative lettering and small text make the task harder. An exact font match depends on the font being available, and some old lettering may remain in the background. Grab Text access is distinct from ordinary Free text boxes; Canva identifies the extraction route as a Pro feature.
Ideogram Layerize Text
Open the image in the supported Studio workflow and use Layerize Text. Select a detected text layer, correct its words and adjust the available typography controls. Inspect both the type and the reconstructed area underneath it.
The current feature is beta, requires sign-in and uses credits. It is available across plans, including Free, while some export sizes and formats require a paid plan. Photographs and other artwork remain an image; extracting lettering does not turn the entire composition into a fully editable PSD or vector document.
Finish this step with: a saved project whose replacement text you can select and edit. Keep the original beside it so you can identify background changes or missed words.
Step 5. Remove and rebuild lettering manually
A manual browser workflow is useful when you want to control the font and each line’s placement. Photopea provides a free route. Open the file with File → Open and keep a working copy.
Clean a plain background
Select the Brush tool, sample a clean area beside the lettering and paint over the old words in the working copy. If the background changes tone across the line, sample suitable colors rather than covering everything with one approximate rectangle.
A deliberate opaque text panel is another layout option when it suits the design. Make it look intentional; it should not resemble an accidental patch over a photograph.
Repair a textured background
For a surface such as paper, sand or fabric, use an appropriate healing or cloning workflow with a clean nearby source. Work in small areas and preserve the direction of texture, edges and shadows. Zoom out often: an area that looks smooth close up can become visibly repetitive at normal size.
Add the new Type layer
Choose Type (T), click for a short line or drag a paragraph box for wrapping text, then enter the replacement. Adjust font, size, color and alignment. Match the baseline and total line width as well as the typeface.
Confirm the edit before switching to another layer. Save a layered working file, then export a separate PNG or JPG. The exported picture is for sharing; the layered file is what allows the next correction.
Finish this step with: a cleaned background and independently editable wording. Check that no fragment of the old letters remains behind the new line.
Step 6. Handle the cases that need a different approach
A paragraph rather than a short date
Rebuild a paragraph as live text when exact wording and later revisions matter. Proofread recognized text before laying it out; OCR can confuse similar characters or miss punctuation. Set the paragraph width and line spacing deliberately, then check the smallest text at its intended viewing size.
For a dense information poster, preserving the artwork and rebuilding the copy can be more manageable than asking a model to redraw every line correctly in one result.
If you are rebuilding the layout too, follow the full workflow to make a poster with AI and keep its final information editable.
A photographed sign or curved label
The replacement must follow the photographed surface, not just use a similar font. A straight overlay on an angled sign can look wrong even when its wording is exact. Match perspective, surface curvature, lighting and softness, or return to the underlying artwork and rebuild the mockup if you have it.
If the image documents the actual appearance of a product, keep that purpose clear. A concept-label redesign is a different deliverable from an accurate photograph of existing packaging.
Match replacement lettering to a sign or curved path
Choose the text structure before matching the surface. In Photopea, point text suits a short line; paragraph text wraps inside a rectangle. Neither automatically makes a line follow a curve.
For an arched heading, first create or select the intended curve in the Paths panel, a shape layer or a vector mask. Choose Type and click the curve's contour. Enter the approved replacement. Photopea places the text between X and O markers; use Path Select to move those endpoints or change which side of the curve carries the letters.
If the phrase is too long, adjust the available path length, wording or type size deliberately. Inspect spacing and the normal-size view rather than compressing the letters until they barely fit. Confirm the Type edit before selecting another layer.
Text following an arc is a different task from a rectangular sign photographed at an angle. A path controls the baseline; it does not supply correct perspective, three-dimensional wrapping or surface lighting. For a complex bottle or folded label, return to the original label artwork and suitable mockup process rather than calling an arc a completed physical-surface match.
Save the live Type layer in the PSD master and export a separate delivery image. These are documented controls and a proposed editing method, not a new repair performed on the exhibition poster.
A caption with no existing words to remove
Use a text overlay. In Adobe Express, upload the picture, choose Text, add and style a box, then download the finished image. Its Free text-on-photo workflow is suitable for this task; premium fonts or assets have their own access conditions.
Place the caption over a quiet area or an intentional panel. Adding an overlay is not a replacement method when old words remain visible underneath.
Finish this step with: a route appropriate to the whole text area, not just a correct first word.
Step 7. Check the download and retain a useful master
Open the actual downloaded image outside the editor. Compare it with the approved replacement note and source. Verify:
- Every changed word, digit and punctuation mark is correct.
- Old wording is absent, including small repeated dates or footers.
- Other words, logos and important image features remain acceptable.
- Alignment, spacing and background repair look coherent at normal size.
- Exported dimensions meet the intended display or printing requirement.
Keep the original, the editable working project and the final sharing file separately. In Photopea, saving as PSD preserves the document structure; exporting as PNG or JPG produces a flattened delivery image. In an online design editor, keep the editable project as well as the download.
If someone else will update the design, name the text fields clearly and identify the actual font. Include the approved wording in the handoff. A flattened image alone forces the next person to repeat the extraction or reconstruction.
Troubleshooting
Free access and practical costs
A manual Photopea route and Adobe Express text overlays offer browser-based Free options for the supported tasks. Extraction tools and generative editing have separate access and usage conditions. Dreamina’s US free allowance is 120 daily credits, with other regional allocations; the model and settings determine how much an attempt consumes.
Before paying, identify the actual gap: recognition, a specific font, a larger export or a generation model. A paid subscription does not automatically make an extracted font exact or prevent other image details changing.
For new artwork rather than a correction, compare AI tools for putting text in images by lettering and editable-output needs.
FAQ
Can I directly edit text in a JPG or PNG?
Usually not as its original text box. The visible letters are pixels. You can replace them with AI, extract recognized wording into new editable type, or remove and rebuild them manually.
Is extracting the words the same as editing the image?
No. If you only need copyable text, OCR may be sufficient. To deliver a new picture with different wording, you must also handle the old letters, background and new typography.
Can I keep every other pixel unchanged?
Do not assume that from a generative result. Use the original layered design when available. For strict preservation in a flat file, use a controlled manual workflow confined to the intended area and compare the exported result with the source.
What if the replacement is much longer?
Treat it as a layout change. Widen the text area, change line breaks or revise the approved wording before shrinking the type. Check neighboring elements and the smallest display version.
What should I save for future changes?
Save the original, an editable project with live text and a final PNG or JPG. For a one-time visual correction, begin with the original in Dreamina, specify the exact substitution and approve the downloaded result against the source.