Flux ArtFirst ask whether the character or its visible mark is missing
When a small acute accent, diaeresis, or mark below a letter disappears from a foreign-language poster, do not treat every case as a typo. The copying process may have deleted the mark; the text may still be intact but display it in the wrong position; or the mark may be clipped only during export. Flux Art is a multi-model AI visual creation and production platform; this article covers editable text layers and export review in external tools and does not claim the platform has built-in character-encoding diagnostics. Text rendered into pixels in a generated image has no directly readable text sequence, so check it separately against the approved glyphs rather than applying text-layer assumptions.
Keep the approved source separate from working copies
Have someone familiar with the target language approve the final wording, then preserve the source text, language, copy version, and marks that must remain. Do not retype from a faulty screenshot and treat that as the approved original. Record where the copy came from, which text object it entered, and the font and size used; first locate where the mark became abnormal. You can isolate the problem word in a diagnostic copy, but retain its context and original layout. Designers must not alter meaning, brand spelling, or grammar to make the text fit. Mark characters from an unknown source as unknown and refer them to the language owner.
One visible letter may contain more than one code point
Some marked Latin letters can be represented as a precomposed character or as a base letter followed by a combining mark. For example, é can be U+00E9 or the sequence U+0065 followed by U+0301. These sequences may be canonically equivalent, but that does not make every similar-looking shape equivalent, nor does it mean the byte sequences have the same length. Unicode normalization guidance describes these relationships. For diagnosis, preserve the original sequence and use a text tool that can display code points to record the actual content; do not infer that a mark is missing from cursor movement, file size, or visual appearance alone.
| What to check | Evidence to preserve | Distinction to make |
|---|---|---|
| Approved text | Original sequence and language version | Approved wording versus text retyped from a faulty screenshot |
| Working text layer | Actual code points and copy steps | A mark missing on input versus a mark not rendered |
| Equivalent samples | A limited comparison using the same font and size | Canonical equivalence versus visual resemblance alone |
| Text-box boundary | A copy where only one layout condition changes | Mark misplacement versus clipping at a boundary |
| Normalization | Before-and-after sequences, rule, and approval record | Standardizing a form versus deleting an accent without approval |
| Actual export | Problem-word locations and language-review results | A correct editable layer versus a correct final deliverable |
Use a limited sample to separate input from rendering issues
Suppose the approved word contains é. In the same software, font, size, and generously sized text box, place both the preserved original sequence and a confirmed canonically equivalent diagnostic sequence. If the original input already lacks the accent, trace the copying or conversion step. If both strings contain the correct mark but only one displays abnormally, record differences in the font and layout environment rather than rushing to change the approved wording. This is a hypothetical troubleshooting example, not a field-tested case. The sample is for diagnosis only; it is not a batch-replacement rule and does not prove that all writing systems can be handled the same way.
Check mark placement and text-box boundaries separately
Once the text sequence is confirmed intact, check that each mark aligns with the correct base letter, does not collide with the line above, and is not clipped by the text box or crop boundary above or below it. In a copy, widen only the box boundary or adjust only the line spacing to see whether the issue changes; do not also change the font, letter spacing, and text input, or the causal clue will be lost. If changing one boundary restores the mark, that only suggests clipping may be involved in this version; review again at the actual size. A small dot that looks decorative may be part of the language and must not be erased just to make the layout cleaner.
Do not use normalization as an excuse to remove marks
NFC and NFD handle canonically equivalent forms; they do not mean removing diacritics. NFKC compatibility normalization can also change some meaningful form distinctions, so do not apply it blindly to approved copy. If the text form must be standardized, preserve the before-and-after sequences, the rule used, and the approval record, then recheck the visible glyphs. If the font or layout environment does not suit the language, choose an approved option that supports the required characters and check all affected words. Do not replace marked letters with plain letters to work around a display problem; fewer characters do not mean the issue is fixed.
Read the exported image back into the review
Save the actual exported image or document and check the mark position, completeness, and readability at normal size for every problem word. If the text layer is correct but the mark is missing in the image, continue tracing the export, scaling, or cropping step. If the issue appears only in one recipient environment, preserve an environment comparison for the same file rather than casually overwriting the original. OCR can help find some differences, but one successful recognition does not prove that a mark's position and glyph are correct. A language reviewer should read the final deliverable back, especially names, key claims, and words that are easy to confuse.
Deliver a mapping from character to glyph
Keep the problem location, approved sequence, working sequence, font and layout conditions, export observations, and decision in one record. After fixing one combining mark, check other marked words in the same font and layout, but do not rewrite all copy indiscriminately. Deliver the approved source text, minimal diagnostic sample, editable source, and final file; explicitly list unknowns and languages not yet reviewed. The record concerns visible text in the image, not asset filenames, so passing filename normalization checks cannot replace this review, and this review does not prove that filenames are safe across systems.
Reference and visual-draft entry point
For character equivalence and normalization, see Unicode UAX #15: https://www.unicode.org/reports/tr15/ . This article applies only a limited set of rules for diagnosis; the actual language and font still need individual review. For a poster visual draft, start from Flux Art's English entry point at https://flux-art.net/en. For approved copy that must be character-accurate, retain editable text and language-review records. Visual styles can continue to be explored, but approved diacritics, combining marks, and names must not be treated as optional decoration during that exploration.