ComfyUI-Lora-Manager/docs/i18n-translation-guidelines.md

i18n Translation Guidelines

This document is the canonical set of conventions for translating LoRA Manager UI strings. It applies to human translators and AI agents alike. Read it before editing anything in locales/.

Source of truth: locales/en.json (10 locales, 1982 leaf keys; all locales share the exact same key structure).

Locales: en, zh-CN, zh-TW, ja, ko, fr, de, es, ru, he (RTL).

Status (2026-08 sweep): a full audit was executed and the terminology, placeholder, stale-text, and untranslated-block fixes described in §2–§6 were applied across all locales (commits 3c3ac49f … fd1227d3). The tables below are now the normative target state, not a to-do list — future edits should preserve these renderings and only add what is new.

Status (2026-09, Other Models): the other model type (VAE / Upscaler / Text Encoder / CLIP Vision / ControlNet) and the Other Models opt-in toggles added 36 new keys; all of them are now translated in all 9 locales (terminology in §2 "Other Models feature"). There are no remaining [TODO: Translate] placeholders in any locale.

Status (2026-09, revision): other.disabled.description, banners.otherModels.content and settings.folderSettings.enableOtherModelsHelp were refreshed in en.json to name all five sub_types (they had listed four, which read as "these are what enabling manages") and re-translated in all 9 locales in the same pass. clip_vision and controlnet are now both opt-in, so the first two describe capability and the third the master switch, not the default set — keep all three enumerating the full five (VAE / upscaler / text encoder / CLIP vision / ControlNet in en; locale slash-list casing follows each file's existing VAE / Upscaler / Text Encoder / … style, de compounds as CLIP-Vision- und ControlNet-Ordner).

Status (2026-09, "no folders found" state): the Other Models page gained an enabled but nothing to scan empty state with 6 new keys (other.noPaths.*); translated in all 9 locales in the same pass. The folder_paths JSON snippet shown in that state lives in templates/other.html, not in the locale files, so it is never translated — only the surrounding prose is. Terminology added in §2.

Status (2026-09, model sources): models can now be linked to ModelScope and TensorArt alongside Hugging Face, which added 15 keys (modelCard.actions.viewOnSource, loras.contextMenu.linkModelSource, modals.linkModelSource.*, modals.model.versions.sourceGroupInfo, toast.contextMenu.enrichNeedsSource, toast.contextMenu.enrichUnsupportedSource) and refreshed the two enrichHfAgent labels, which had hardcoded "HF" for a button that now also enriches ModelScope models. The modals.linkModelSource.urlPlaceholder value stays byte-identical to en.json (it is a URL, the §6 exception). Terminology in §2, "Model source feature".

---

1. Hard rules (do not violate)

R1 — Key structure is sacred

• Only locales/en.json may add/remove/rename keys. All other locales must keep the exact same nested key set. tests/i18n/test_i18n.py enforces this.
• When a new UI string is added to en.json, run python scripts/sync_translation_keys.py (adds the missing keys to all locales with [TODO: Translate] placeholder copies) — then stop. Do NOT translate proactively: placeholders are the expected end state during feature development, and translations are filled in only when the feature owner explicitly asks (workflow details in §7).
• Never reorder, re-indent, or reformat a locale file "for tidiness". The sync script preserves formatting; manual reformatting creates noisy diffs.

R2 — Placeholders and HTML must be preserved verbatim

• {name}-style placeholders must appear in the translation exactly as in en.json. Do not invent placeholders the source string does not have — the caller may not pass them (example bug: zh-CN recipes.controls.import.downloadLocationPreview added {path}; the template renders this key with no parameters, so the literal text {path} shows in the UI).
• {{...}} in a locale value is an escaped literal brace — keep it identical.
• Keep embedded HTML tags (e.g. <strong>...</strong>, <code>...</code>) intact. You may move the tag around the sentence if the target language needs different word order.

R3 — Never translate or transliterate these

• Model types: LoRA, Checkpoint, Embedding, Diffusion Model
• Products/brands: LoRA Manager, ComfyUI, CivitAI, CivArchive, HuggingFace, Ko-fi
• Ecosystem names: LyCORIS, DoRA, trigger-adjacent jargon Prompt, Workflow (these are used as-is in the target-language SD community; see §2 per-language policy)
• Theme names: Nord, Midnight, Monokai, Dracula, Solarized

R4 — The "Recipe" convention (the most important domain term)

Product intent: a Recipe records a LoRA combination + generation parameters (prompt, seed, sampler, …) that reproduces an image style. The metaphor is a cooking recipe — "follow it and you get a similar dish". It is not a menu, not a dish list, not a prescription.

Decision per language — translate only into a word whose everyday primary meaning is a cooking recipe; where that word would mislead users, keep the English "Recipe(s)":

Locale	Use	Never use
fr	Recipe / Recipes (keep English)	recette(s) — cooking reading is secondary and it was explicitly judged misleading
zh-CN / zh-TW	配方	食谱 (reads as "food cookbook")
ja	レシピ	— (leftover English "Recipe" in initialization.recipes.title / toast.recipes.recipeSaved → translate)
ko	레시피	—
de	Rezept / Rezepte	— (cooking meaning dominant; prescription reading acceptable)
es	receta / recetas	— (cooking meaning dominant)
ru	рецепт / рецепты	— (leftover English "Recipe" in initialization.recipes.title / toast.recipes.recipeSaved → translate)
he	מתכון / מתכונים	— (cooking meaning dominant)

Whatever the choice, one concept = one noun within a locale. Currently violated in:

• fr — "Recipe" (~97 keys, incl. nav) mixed with "recette" (~58 keys)
• zh-CN / zh-TW — 配方 (126/122 keys) mixed with 食谱 / 食譜 (14/17 keys, all in the rematch flow: globalContextMenu.rematchRecipes.*, toast.recipes.rematch*)
• de — "Rezept" (136 keys) mixed with leftover English "Recipe" (5 keys)
• ja / ru — leftover English "Recipe" in initialization.recipes.title ("Recipe Manager zu initialisieren" / «Инициализация Recipe Manager») and toast.recipes.recipeSaved

R5 — One term, one rendering (within each locale)

Same source word must not be translated several ways in one file. Known offender areas (see §5 for the full fix list): recipe, Checkpoint, Embedding, prompt, base model, preset, workflow, hash, metadata, tags, bulk. Every locale currently mixes variants of at least one of these — pick the preferred form in the §2 tables and normalize.

R6 — Register consistency

• zh-CN / zh-TW: pick 你 or 您 once. Do not mix (zh-CN has 44×你 + 5×您; zh-TW has 27×您 + 18×你).
• de: pick "du" or "Sie" once (currently 143×Sie + ~7×du).
• es: pick "tú" or "usted" once.

R7 — Punctuation per script

• Full-width punctuation ：（） is correct only in CJK locales (zh-CN, zh-TW, ja, ko).
• Latin/Cyrillic/Hebrew locales must use ASCII : () — full-width colons leaked in there are machine-translation artifacts. Known: fr toast.recipes.createError/createFailed, es toast.recipes.createError/createFailed (e.g. "…de la receta：" should be "…de la receta:").
• fr apostrophes must be U+2019 ' / ASCII ', never a straight double quote: fr header.filter.allowSellingGeneratedContentTooltip currently reads vendre d"images → fix to d'images. Do not mix ' and ' in one file (fr has 299 vs 15).
• Ellipsis: use ASCII ... (project style). Don't introduce ….
• Keep the sentence-ending period/omission consistent with the source string where the language allows it.
• he is RTL: mix of Hebrew and Latin scripts is normal; keep Latin term ordering natural.

R8 — No untranslated English leftovers

Full sentences left byte-identical to en.json are bugs (brand names and URL placeholders are the exception). Every locale has them; see §6 for the per-locale checklist. [TODO: Translate] placeholders are the sanctioned intermediate state during feature development (see §7) — do not "fix" them unless the feature owner asked for translations.

R9 — Mirror the source even when the source is wrong

If en.json itself contains an inconsistency (e.g. the Civitai vs CivitAI casing split, or the CivitArchive typo in modals.relinkCivitai.helpText.format4), translate/transcribe it as-is in your locale and instead fix the source in en.json (then propagate by re-syncing and re-translating affected keys). Do not silently diverge in one locale only.

---

2. Per-language term maps

Preferred rendering per term. "Fix" means the locale currently contains the wrong variant and must be normalized. en = keep the English word as-is.

fr

Term	Use	Fix
recipe	Recipe(s)	Replace all "recette(s)" (58 keys, e.g. recipes.actions.deleteRecipeWithShortcut, toast.recipes.rematchComplete) with "Recipe(s)"
Checkpoint	Checkpoint	statistics.modelTypes.checkpoint = "Point de contrôle" → "Checkpoint"
trigger words	mot(s)-clé(s)	unify: modals.model.triggerWords.editWord uses "mot déclencheur" — pick one
prompt / negative prompt	Prompt / prompt négatif	—
base model	modèle(s) de base	—
preset	préréglage	unify: modals.model.usageTips.addPresetParameter "prédéfini", toast.presets.restored "par défaut"
hash	hash	conflictConfirm.message "hachage" → "hash"
tags	tags	settings.sections.priorityTags "Étiquettes" → "Tags"
metadata	métadonnées	loras.controls.refresh.fullTooltip keeps English "metadata"
duplicates	doublon(s)	unify with "dupliqué(e)s"
bulk	groupé(e)	unify with "par lot / mode lot" variants

de

Term	Use	Fix
recipe	Rezept/Rezepte	leftover English "Recipe" keys → Rezept (e.g. toast.recipes.recipeSaved)
base model	pick Basis-Modell or Basismodell	currently 27× hyphenated vs 15× closed
metadata	Metadaten	4 keys use "Modelldaten" (onboarding.steps.fetch.title/content) → Metadaten
bulk	pick Massen- or Sammelmodus	loras.controls.bulk.action = "Massen" reads as "crowds" — use "Massenbearbeitung"/"Mehrfachauswahl"
register	Sie (formal)	7 keys use "du/dein" (settings.backup.managementHelp, modals.checkUpdates.message/tip, doctor.footer, …)

es

Term	Use	Fix
recipe	receta(s)	—
Checkpoint	Checkpoint	5 statistics keys "Punto(s) de control" → "Checkpoints" (statistics.metrics.checkpoints, statistics.insights.unusedCheckpoints.*, statistics.modelTypes.checkpoint)
trigger words	palabra(s) de activación	2 keys already use it; ~15 keys "palabra(s) clave" (reads as search keyword) → unify
base model	modelo base	—
preset	preajuste	3 keys keep English "preset", 1 "preestablecido" → preajuste
workflow	pick flujo de trabajo or workflow	currently 21× "flujo de trabajo" vs 10× "workflow"
bulk	masivo / por lotes	unify; "Batch Import" → traducción
tags	etiquetas	—

ru

Term	Use	Fix
recipe	рецепт(ы)	English leftovers: initialization.recipes.title, recipes.batchImport.*, toast.recipes.recipeSaved → translate
Checkpoint	Checkpoint (recommended)	3 variants today: "Checkpoint" (17 keys), «Чекпойнт», «Контрольная точка» (statistics, 6 keys) — statistics MUST drop «Контрольная точка»
Embedding	Embedding	«Эмбеддинг» variant exists in settings.priorityTags.modelTypes.embedding — unify
prompt	промпт	8 keys use «запрос» (reads as "database/HTTP request") → «промпт»
base model	базовая модель	—
preset	пресет	header.theme.presets "Предустановки" → пресеты
workflow	Workflow (recommended)	«рабочий процесс» used in 4 keys — unify
hash	pick хеш or хэш	both spellings co-occur
tag(s)	тег(и)	—
typos	—	settings.misc.loraSyntaxFormatHelp: «безпотерьного» → «беспотерьного»

he

Term	Use	Fix
recipe	מתכון / מתכונים	—
Checkpoint	Checkpoint	5 statistics keys «נקודת/נקודות ביקורת» (road/security checkpoint) → "Checkpoint(s)" (statistics.metrics.checkpoints, statistics.modelTypes.checkpoint, statistics.insights.unusedCheckpoints.*)
Embedding	Embedding	statistics keys use הטמעות → Embedding
prompt	pick הנחיה or פרומפט	9 keys הנחיה vs 3 פרומפט — unify (recommend פרומפט, SD-community loanword)
preset	קביעה מראש	header.filter.presetOverwriteConfirm uses פריסט → unify
hash	pick one of האש / גיבוב / hash	3 variants co-occur — unify (recommend hash or גיבוב)
metadata	pick מטא-דאטה or מטא-נתונים	38 vs 17 keys — unify
model	מודל	13 keys use דגם/דגמים — unify
bulk	pick one of 5 variants	5 different renderings ("כמות גדולה", "המוני", "קבוצתי", "אצווה", …) — unify; loras.controls.bulk.action "כמות גדולה" reads as "large quantity"

ja

Term	Use	Fix
recipe	レシピ	initialization.recipes.title keeps English "Recipe Manager" — translate to レシピマネージャー
Checkpoint	Checkpoint or チェックポイント (pick one)	3 variants: Checkpoint (~14), checkpoint lowercase (4), チェックポイント (4, e.g. settings.priorityTags.modelTypes.checkpoint)
Embedding	Embedding	4 keys lowercase "embedding" mid-sentence
bulk	一括	modals.checkUpdates.tip "バルクモード" → 一括モード
recipe counter	件 or 個	globalContextMenu.rematchRecipes.success uses 件, .cancelled uses 個 — unify

ko

Term	Use	Fix
recipe	레시피	—
Checkpoint	Checkpoint (recommended)	4 keys transliterate 체크포인트 (settings.priorityTags.modelTypes.checkpoint, toast.recipes.missingCheckpointPath/missingCheckpointInfo/downloadCheckpointFailed)
Embedding	Embedding	3 keys 임베딩 (settings.priorityTags.modelTypes.embedding, uiHelpers.nodeSelector.embedding)
base model	베이스 모델	6 keys «기본 모델» read as "default model" → 베이스 모델 (settings.downloadSkipBaseModels.*, toast.loras.downloadSkippedByBaseModel)
workflow	pick 워크플로 or 워크플로우	26 vs 6 keys — unify
bulk	일괄	modals.checkUpdates.tip "벌크 모드" → 일괄 모드
tag logic	—	header.filter.tagLogicAny = "모든 태그 일치 (OR)" is inverted (should be "하나 이상의 태그 일치") and identical to tagLogicAll
particle	—	modelCard.sendToWorkflow.checkpointNotImplemented: "Checkpoint을" → "Checkpoint를"

zh-CN / zh-TW

Term	zh-CN	zh-TW
recipe	配方 (fix 食谱 → 配方, 14 keys in rematch flow)	配方 (fix 食譜 → 配方, 17 keys in rematch flow)
Checkpoint	Checkpoint (fix 检查点 → Checkpoint, 5 keys: toast.recipes.missingCheckpointPath/missingCheckpointInfo/downloadCheckpointFailed, modelCard.actions.checkpointNameCopied, modelCard.sendToWorkflow.checkpointNotImplemented)	Checkpoint (fix 檢查點 → Checkpoint, 4 keys: modelCard.actions.copyCheckpointName, toast.recipes.missing*×2, toast.recipes.downloadCheckpointFailed)
base model	基础模型 (fix 基模型 → 基础模型, 3 keys in modals.model.versions.filters.*)	基礎模型 ✓ consistent
prompt	提示词 ✓	提示詞 ✓
preset	预设 ✓	預設 ✓
workflow	工作流 ✓	工作流 ✓
trigger words	触发词 ✓	觸發詞 ✓
hash	哈希 (哈希值 variant OK)	雜湊 ✓
register	你 (fix 5×您 → 你)	您 (fix 18×你 → 您)

Other Models feature (VAE / Upscaler / Text Encoder / CLIP Vision / ControlNet)

The other model type exposes five sub_types. They are model-type names, so they follow R3 and stay in Latin in every locale. The settings.folderSettings.subType* values are therefore intentionally byte-identical to en.json (same precedent as settings.priorityTags.modelTypes / checkpoints.modelTypes.checkpoint) — a §6 sweep must not "fix" them.

Term	Rendering	Note
VAE	VAE everywhere	acronym, always upper-case
Upscaler	Upscaler everywhere	CivitAI ModelType name
Text Encoder	Text Encoder everywhere	de compounds as Text-Encoder-Stammordner
CLIP Vision	CLIP Vision everywhere	de compounds as CLIP-Vision-Stammordner
ControlNet	ControlNet everywhere	brand casing, capital N

In prose these names sit next to localized nouns the same way Diffusion Model does (zh VAE 根目录, ja VAEルート, ko VAE 루트, ru Корневая папка VAE).

"Other Models" is the page/feature name, not a model type — translate it:

Locale	other.title	header.navigation.other
fr	Autres modèles	Autres
zh-CN	其他模型	其他
zh-TW	其他模型	其他
ja	その他のモデル	その他
ko	기타 모델	기타
de	Weitere Modelle	Andere
es	Otros modelos	Otros
ru	Другие модели	Другое
he	מודלים אחרים	אחרים

settings.folderSettings.otherSubTypes ("Managed Types") must name model types, matching each locale's header.filter.modelTypes rendering (zh 管理的模型类型, ja 管理するモデルタイプ, de Verwaltete Modelltypen, …).

The "no folders found" empty state (other.noPaths.*) uses two phrases that must stay consistent whenever that copy is edited. folder key means the folder_paths key name (vae, upscale_models, … — Latin per the table above); on disk means the folder must physically exist:

Phrase	Rendering
folder key	zh-CN 文件夹键 · zh-TW 資料夾鍵 · ja フォルダーキー · ko 폴더 키 · fr clé de dossier · de Ordnerschlüssel · es clave de carpeta · ru ключ папки · he מפתח תיקייה
on disk	zh-CN 在磁盘上 · zh-TW 在磁碟上 · ja ディスク上 · ko 디스크에 · fr sur le disque · de auf dem Datenträger · es en el disco · ru на диске · he בדיסק

settings.json and ComfyUI stay verbatim in every locale; "reload this page" / "restart LoRA Manager" reuse each locale's existing restart wording (settings.extraFolderPaths.*).

Model source feature (Hugging Face / ModelScope / TensorArt)

A model file can be linked to the page of an external model site. Hugging Face, ModelScope and TensorArt are brand names and stay Latin in every locale (R3); the generic nouns around them are translated:

Term	Rendering
model source	zh-CN 模型来源 · zh-TW 模型來源 · ja モデルソース · ko 모델 소스 · fr source de modèle · de Modellquelle · es fuente de modelo · ru источник модели · he מקור מודל
model page	zh-CN 模型页面 · zh-TW 模型頁面 · ja モデルページ · ko 모델 페이지 · fr page du modèle · de Modellseite · es página del modelo · ru страница модели · he עמוד המודל
model card	zh-CN 模型卡 · zh-TW 模型卡 · ja モデルカード · ko 모델 카드 · fr fiche de modèle · de Modellkarte · es ficha de modelo · ru карточка модели · he כרטיס מודל
AI enrichment (noun)	reuse the existing pair per locale: zh-CN 增强 · zh-TW 增強 · ja 補完 · ko 보강 · fr enrichissement (par IA) · de Anreicherung (KI-) · es enriquecimiento (con IA) · ru обогащение (с помощью ИИ) · he העשרה (AI)

modelCard.actions.viewOnSource ("View on {source}") follows each locale's existing viewOnHuggingFace pattern — de Auf … ansehen, ru Открыть …, he צפייה ב-…, ja … で見る, ko …에서 보기, zh 在 … 查看, fr Voir sur …, es Ver en …. {source} is replaced at runtime with the untranslated platform name, so the brand never appears inside the translated text.

modals.linkModelSource.enrichNote states the rule that only sites exposing a readable model card can be enriched and names TensorArt as the current exception. Keep the parenthetical exception in sync if another link-only source is ever added — the sentence is deliberately phrased as a rule, not as an apology for one site.

The context-menu and bulk-operation enrichment entry points read "Enrich Metadata with AI" in en, not "Enrich HF Metadata": they cover ModelScope as well, so no locale may reintroduce an HF qualifier in loras.contextMenu.enrichHfAgent / loras.bulkOperations.enrichHfAgent (the key names keep the historical Hf; only the values changed).

---

3. Cross-cutting confusion hot-spots (must-fix list)

All items below were resolved in the 2026-08 sweep — treat them as a regression watch-list: do not reintroduce these renderings.

1. Checkpoint rendered as a literal security/road checkpoint — fr, es, ru, he, zh-CN, zh-TW all had 4–6 keys in the statistics.* domain reading as "control point"; reverted to "Checkpoint".
2. "recipe" variants that break the one-noun rule — fr "recette" → "Recipe", zh 食谱/食譜 → 配方, de/ja/ru leftover English "Recipe" translated.
3. ko header.filter.tagLogicAny — was inverted ("모든 태그 일치 (OR)") and identical to tagLogicAll; now "어느 하나의 태그와 일치 (OR)".
4. ja modals.model.versions.actions.viewLocalTooltip — was the stale "近日対応予定" ("coming soon"); all 9 locales now describe the actual action.
5. Stale help texts — settings.downloadSkipBaseModels.help, settings.aiProvider.apiBaseHelp, settings.hideEarlyAccessUpdates.help retranslated in all locales to the current en.json wording.
6. en.json source bugs (fixed in source, then mirrored):
   • "Civitai" → "CivitAI" brand casing (values only; key names relinkCivitai etc. keep their lowercase form and must not be renamed)
   • modals.relinkCivitai.helpText.format4 "CivitArchive" typo → "CivArchive"
   • zh-CN recipes.controls.import.downloadLocationPreview invented {path} removed

---

4. Placeholder contract deviations (current)

{...} token sets must match en.json per key. All deviations found in the 2026-08 sweep were fixed, with one intentional exception:

toast.settings.mappingsUpdated — the caller passes a hardcoded English inflection (plural: count !== 1 ? 's' : ''). Languages that cannot build a plural by appending that s (zh-CN/zh-TW, ja, ko, de, ru, he) drop {plural} and render a count-friendly form (({count}) or a measure word); fr and es keep it (mappage{plural}, mapeo{plural}).

# keep a copy of this rule next to the key if it ever moves:
#   fr/es:  "... ({count} mappage{plural})"
#   de/ru/he: "... ({count})"
#   zh-CN: "（{count} 条映射）" / zh-TW: "（{count} 個對應）" / ja: "（{count} マッピング）"

Do NOT add {...} tokens the source lacks (the caller will not supply them, and the literal text renders in the UI), and do NOT rename source tokens ({typePlural} stays {typePlural}).

---

5. One term, one rendering — offender matrix

Cross-locale summary of §2 inconsistencies. "✓" = already consistent. All ✗ cells were resolved in the 2026-08 sweep; the row shows the single rendering now in force per locale.

Term	fr	de	es	ru	he	ja	ko	zh-CN	zh-TW
recipe	Recipe	Rezept	receta	рецепт	מתכון	レシピ	레시피	配方	配方
Checkpoint	Checkpoint	Checkpoint	Checkpoint	Checkpoint	Checkpoint	Checkpoint	Checkpoint	Checkpoint	Checkpoint
Embedding	Embedding	Embedding	Embedding	Embedding	Embedding	Embedding	Embedding	Embedding	Embedding
prompt	Prompt	Prompt	prompt	промпт	פרומפט	プロンプト	프롬프트	提示词	提示詞
base model	modèle de base	Basismodell	modelo base	базовая модель	מודל בסיס	ベースモデル	베이스 모델	基础模型	基礎模型
preset	préréglage	Voreinstellung	preajuste	пресет	קביעה מראש	プリセット	프리셋	预设	預設
workflow	Workflow	Workflow	workflow	Workflow	workflow	ワークフロー	워크플로	工作流	工作流
hash	hash	Hash	hash	хеш	hash	ハッシュ	해시	哈希	雜湊
metadata	métadonnées	Metadaten	metadatos	метаданные	מטא-נתונים	メタデータ	메타데이터	元数据	中繼資料
tags	Tags	Tags	etiquetas	теги	תגיות	タグ	태그	标签	標籤
duplicates	en double	Duplikate	duplicados	дубликаты	כפילויות	重複	중복	重复项	重複項
bulk	groupé	Massen-	por lotes	пакетный	בכמות גדולה	一括	일괄	批量	批量

Watch: ja/ko keep the model-type names Checkpoint/Embedding and Diffusion Model in Latin (consistent with their model-type sections) — do not transliterate them as チェックポイント/체크포인트.

---

6. Untranslated English leftovers (status)

Values byte-identical to en.json that are actual UI sentences are bugs (brand names and URL placeholders are the exception). As of the 2026-08 sweep, all previously untranslated blocks are translated in every locale: recipes.batchImport.* + toast.recipes.batchImport* (fr/de/es/ru/he/ja/ko), banners.communitySupport.*, modals.model.license.*, globalContextMenu.fetchMissingLicenses.*, the doctor.* issue/action/label subset, toast.settings.libraryLoadFailed / libraryActivateFailed, toast.api.moveFailed, settings.extraFolderPaths.restartRequired, toast.recipes.recipeSaved, sidebar.dragDrop.moveUnsupported, checkpoints.modelTypes.diffusion_model (ja/ko keep the English loanword), initialization.recipes.title.

The only values that remain intentionally identical to en.json are non-translatable: URL/path placeholders (https://…, C:/…), numeric presets (5 (1080p), 6 (2K), 8 (4K)), example token lists (character, concept, style(toon|toon_style)), service/provider names (CivitAI → CivArchive → Archive DB), model-type names (settings.priorityTags.modelTypes.*, settings.folderSettings.subTypeVae … subTypeControlnet — see §2), and the external playlist title (help.updateVlogs.playlistTitle, de: translated to "LoRA Manager-Update-Playlist").

Rule for uiHelpers.workflow.noPromptTargets: the second line (Mark as → Send Prompt Target) quotes literal ComfyUI context-menu items — keep those menu labels in English in every locale because that is what the user actually sees in ComfyUI.

License labels (modals.model.license.*): the restriction labels are now translated in all locales (the sibling creditRequired has always been translated).

---

7. Workflow for agents and translators

Adding a new UI string

1. Add the key to locales/en.json only.
2. Run python scripts/sync_translation_keys.py — it inserts the key into the other 9 locales (as a [TODO: Translate] placeholder) preserving formatting.
3. During feature development, stop here. While the UI copy is still in flux, leave the [TODO: Translate] placeholders as-is — translating churning strings into 9 locales is wasted work. Placeholders are a normal intermediate state, not a bug.
4. Once the wording is final and the feature owner explicitly asks for translations, translate all pending [TODO: Translate] keys in every locale (not just the latest feature's), applying §1–§3 (placeholders verbatim, Recipe rule, term maps, register). Find pending keys with: grep -c "TODO: Translate" locales/*.json
5. If the new string contains new terminology, extend §2 tables.

Fixing a translation bug

1. Locate the key (dotted path) in the relevant locale file.
2. Check the corresponding en.json value and the actual caller (grep static/js or web/comfyui for the key) to learn which placeholders are passed.
3. Fix trivially; for normalization sweeps (e.g. "recette" → "Recipe"), do it file-wide for the offending keys only — do not touch unrelated lines.
4. If the bug is in en.json itself (R9), fix the source first, then re-sync and update all locales.

Verification

pytest tests/i18n/test_i18n.py     # key parity + JSON validity + JS key references
python scripts/sync_translation_keys.py --dry-run   # shows which keys would change; add --verbose for per-key detail
npm test                           # frontend tests incl. i18n helpers

pytest tests/i18n only checks structure. Quality conventions in this document are not machine-enforced — a human/agent review pass is required.

Anti-patterns checklist

• Placeholders {x} / {{x}} differ from en.json
• Same source term translated 2+ ways in the same file (see §5)
• "Checkpoint" became a literal checkpoint; "recipe" became menu/prescription/food-cookbook
• Brand names translated or transliterated (LoRA, CivitAI, ComfyUI, …)
• Latin locale using full-width ：（）; fr using " as apostrophe
• Mixed 你/您, du/Sie, tú/usted
• Full English sentences left behind (see §6)
• Register/typos/mojibake; source string is stale vs en.json (compare semantics, not just words)