How to prepare an image before replacing translated text

Prepare an image for text translation by finding preventable source problems before any words are replaced. Check that the text is readable, the crop is complete, compression has not damaged letter edges, and every text area has been recorded. Keep an untouched original so the translated result can be compared against a stable reference.

This preflight cannot guarantee a good translation or a close visual match. It does prevent avoidable confusion, such as translating a cropped word, missing a small badge, or judging a rebuilt area against the wrong version of the image.

Inspect the source image

Open the image at normal viewing size first. Ask whether a person can read each line without guessing. Then inspect the text more closely for damaged edges, missing punctuation, and overlap with the subject.

Start with five questions:

Is every character inside the crop?
Can small text be read without interpreting blurred shapes?
Does text overlap a face, product, logo, or detailed illustration?
Are any words meant to stay unchanged?
Is this the cleanest available export?

If an editable source exists, keep it beside the flattened export. It can answer questions about fonts, effects, hidden text, and original dimensions. If only a flattened copy remains, record that constraint. Do not write a plan that assumes layers can be recovered later.

Also check whether the image contains text that should not be translated. Brand names, model numbers, interface labels, dates, and decorative lettering may need different treatment. The right decision depends on the job, so mark each region as translate, keep, remove, or review.

Check resolution and compression

There is no single pixel threshold that makes an image suitable for every translation workflow. A large headline on a plain background may survive in a modest export. A small label over a textured surface may be difficult even in a larger file.

Inspect practical signals instead:

Signal	What to look for	Action
Letter edges	Blocks, ringing, blur, or missing thin strokes	Find a cleaner source if possible
Small characters	Accents, punctuation, counters, and narrow verticals	Confirm each character before translation
Repeated scaling	Soft edges or uneven sharpness	Return to the earliest available export
Color transitions	Banding or color bleed around text	Preserve the original for later comparison
Final-size readability	Text that works only when zoomed in	Redesign or remove the small copy

Avoid resaving the source several times before translation. Some formats and export settings can add visible damage with each generation. Keep the original file untouched, and make working copies for crops or annotations.

Check the input rules for the workflow you will actually use. Passing a file size or dimension limit only means that the file can enter the workflow. It does not show that the text is readable or that the background can be rebuilt cleanly.

Identify every text area

Create a text-area inventory before translating. Number the regions on a copy of the image, then use the same numbers in a simple table.

Region	Source text	Decision	Position and role	Risk note
1	Record exactly	Translate or keep	Main headline, top left	Overlaps subject
2	Record exactly	Translate or review	Supporting line	Small type
3	Record exactly	Keep or remove	Badge or logo	Decorative lettering

Fill the table with the actual source wording and decisions for the image being reviewed.

Record reading order as well as position. A large headline usually needs to remain dominant. Supporting copy should not become more prominent simply because the translation is longer. Badges and small labels need enough room to stay legible without competing with the main message.

Transcribe text before translating it. That gives the language reviewer something clearer than a screenshot and makes missing lines easier to catch. Keep names and numbers visible beside the translated copy so they can be checked separately.

Remove avoidable ambiguity

Ambiguity enters when a reviewer cannot tell whether a mark is a letter, effect, decoration, or damage. Resolve what you can before the image goes through any reconstruction step.

Remove temporary guides, selection boxes, captions, or interface chrome from the working copy.
Use the original export instead of a screenshot when it exists.
Confirm whether stylized logos should be translated.
Mark text hidden behind objects or cut by the crop.
Note low-contrast words that are intentionally subtle.
Ask for a language decision when a phrase could be read in more than one way.

Do not clean the image by erasing meaningful detail. Preparation should clarify the source, not redesign it in advance. If text sits on a complex background, leave the original intact and annotate the risk on a separate copy.

Source preparation should happen before translation or in a separate editing workflow. Uploading an ambiguous image does not resolve whether a mark is text, decoration, or damage.

Save a clean original for comparison

Keep three clearly named files: the untouched original, an annotated review copy, and the working copy used for translation or reconstruction. This separation prevents an editor from accidentally comparing the result with an image that has already been cropped, compressed, or cleaned.

Before approving a translated result, compare it against the untouched original for:

complete and correct text coverage;
crop and subject position;
headline hierarchy and reading order;
font style, scale, color, outline, and shadow;
background details around each rebuilt region;
changes that only become visible at normal display size.

Record the source filename and the review decision. If a compromise is accepted, such as a similar font because the original was unavailable, write that down. A clean baseline makes later repair easier and keeps the review grounded in the image that was actually approved.