How to localize a flattened image when you no longer have source layers

You can localize a flattened image without the source layers, but you cannot edit it as if the text, background, and effects were still separate objects. Work from an untouched export, inventory every text area, translate the copy outside the image, rebuild each region, and compare the result against the original. If the source is too damaged or the text is too tightly woven into the artwork, redesigning the image may be safer.

Why flattened images are harder to localize

An editable design file can keep text, shapes, photos, masks, and effects on separate layers. A flattened JPG, PNG, or WebP combines them into pixels. The letters may look clear, but they no longer carry an editable font name, size, tracking value, or shadow setting.

That changes the work in three ways. First, removing the old text can expose a background that must be reconstructed. Second, the original font may be unknown or unavailable. Third, the translated phrase may not fit the same space even when the visual treatment is matched closely.

Do not begin by trying to erase letters. Begin by recording what the source contains:

• each text block and its reading order;
• text that should stay unchanged;
• alignment and approximate boundary;
• overlap with a subject, logo, or detailed texture;
• visible color, outline, shadow, and hierarchy.

This inventory becomes the reference when a rebuilt area looks plausible but is missing a small label or has moved a line into the wrong place.

Check image quality and text density

The best available export is not always the largest file. Inspect the edges of letters, small punctuation, and the background around the text. A file may have large pixel dimensions but still show heavy compression or repeated scaling.

Text density matters because every region adds a translation decision and a visual repair. A poster with one large headline is a different task from a chart with labels, legends, footnotes, and numbers. If the image contains dense factual text, confirm that image localization is the right format. Rebuilding the information in an editable layout may be more reliable.

Translate without editable layers

Use a workflow that keeps language decisions separate from pixel edits:

1. Duplicate the original and lock the untouched copy.
2. Transcribe every text area into a numbered list.
3. Mark names, numbers, acronyms, and words that should not change.
4. Translate the list and have the target-language copy reviewed.
5. Decide which areas can be reconstructed automatically and which need manual work.
6. Remove or cover the source text region by region.
7. Place the approved translation within the recorded boundaries.
8. Rebuild type treatment and effects, then compare the whole image.

Do not copy the original line breaks without checking the target language. They may split a phrase in an awkward place. Do not shrink every long translation until it fits. That can demote a headline or make supporting text unreadable. Try an approved shorter phrasing, another line break, or a wider text block before reducing the type too far.

Finish the source-image preparation checks before replacing any text. If the editing workflow does not offer masks, font controls, or position controls, move the affected regions to a manual editor instead of assuming the tool can recover the missing layers.

Compare reconstructed areas

Compare the result to the untouched source in two views. At normal size, check hierarchy and readability. At closer zoom, check edges, background repair, and small characters.

Use a region table during review:

Check the negative space around each block. A repair can preserve the letters yet change the composition by crowding a person or leaving a strange empty area. Also inspect the background after the old text was removed. Repeated patterns, smeared edges, or altered objects are reasons to revise.

Keep notes on compromises. If the exact font was unavailable, record that the result uses a visual approximation. If a phrase was shortened, record who approved the wording. Those notes help the next reviewer understand what was intentional.

Know when manual redesign is safer

Stop automated reconstruction and move to manual redesign when the source does not support a confident review. Common stop conditions include unreadable text, severe compression, cropped characters, dense labels, text woven through a subject, or a background that cannot be repaired without obvious damage.

Manual redesign is also safer when the target language needs a new hierarchy rather than a small fit adjustment. A longer headline may need a different crop, a moved subject, or a rewritten supporting line. Those are design decisions, not simple text replacement.

Use this decision rule:

• Continue when every source line is known and the rebuilt regions can be checked clearly.
• Revise when the translation is correct but one or more visual attributes need repair.
• Redesign when meaning, readability, or important image detail cannot be protected within the original composition.

Missing layers do not make localization impossible. They remove certainty and editing control. A careful inventory, separate language review, and honest stop condition keep that loss of control visible.