gpatti/MediaLore
Archived
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Compare commits

base: gpatti:236f168eebd4a71dac45a38acbf889691c5b03d3
Branches Tags
gpatti:main gpatti:scanning-updates
...
compare: gpatti:9d73459f48af31e757c3e644c51dfacd133ac678
Branches Tags
gpatti:main gpatti:scanning-updates

4 Commits
236f168eeb ... 9d73459f48

Author SHA1 Message Date
gpatti
9d73459f48 Merge pull request 'customize-context-length' (#25) from customize-context-length into main
All checks were successful
Build and Push Docker Image / build (push) Successful in 1m7s
Details 
Reviewed-on: http://gitea.lan/gpatti/MediaLore/pulls/25
 2026-04-13 23:41:09 +00:00
Garret Patti
9b2690f639 add tesseract ocr 2026-04-13 19:40:25 -04:00
Garret Patti
1350a6f94b separate text extraction and translation 2026-04-13 17:45:00 -04:00
Garret Patti
2fc9a34626 add configurable max_tokens per AI activity 
Allows users to configure the max_tokens sent to the AI endpoint for
each activity (tagging, description, extraction, translation) individually,
with per-library overrides following the same pattern as model overrides.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
 2026-04-13 13:57:07 -04:00
11 changed files with 577 additions and 97 deletions
Expand all files Collapse all files

117
package-lock.json generated
View File

@@ -17,7 +17,8 @@
"node-cron": "^4.2.1",
"react": "^19.2.4",
"react-dom": "^19.2.4",
"sharp": "^0.34.5"
"sharp": "^0.34.5",
"tesseract.js": "^7.0.0"
},
"devDependencies": {
"@tailwindcss/postcss": "^4.2.2",
@@ -2950,6 +2951,12 @@
"readable-stream": "^3.4.0"
}
},
"node_modules/bmp-js": {
"version": "0.1.0",
"resolved": "https://registry.npmjs.org/bmp-js/-/bmp-js-0.1.0.tgz",
"integrity": "sha512-vHdS19CnY3hwiNdkaqk93DvjVLfbEcI8mys4UjuWrlX1haDmroo8o4xCzh4wD6DGV6HxRCyauwhHRqMTfERtjw==",
"license": "MIT"
},
"node_modules/brace-expansion": {
"version": "1.1.12",
"resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-1.1.12.tgz",
@@ -4803,6 +4810,12 @@
"hermes-estree": "0.25.1"
}
},
"node_modules/idb-keyval": {
"version": "6.2.2",
"resolved": "https://registry.npmjs.org/idb-keyval/-/idb-keyval-6.2.2.tgz",
"integrity": "sha512-yjD9nARJ/jb1g+CvD0tlhUHOrJ9Sy0P8T9MF3YaLlHnSRpwPfpTX0XIvpmw3gAJUmEu3FiICLBDPXVwyEvrleg==",
"license": "Apache-2.0"
},
"node_modules/ieee754": {
"version": "1.2.1",
"resolved": "https://registry.npmjs.org/ieee754/-/ieee754-1.2.1.tgz",
@@ -5288,6 +5301,12 @@
"url": "https://github.com/sponsors/ljharb"
}
},
"node_modules/is-url": {
"version": "1.2.4",
"resolved": "https://registry.npmjs.org/is-url/-/is-url-1.2.4.tgz",
"integrity": "sha512-ITvGim8FhRiYe4IQ5uHSkj7pVaPDrCTkNd3yq3cV7iZAcJdHTUMPMEHcqSOy9xZ9qFenQCvi+2wjH9a1nXqHww==",
"license": "MIT"
},
"node_modules/is-weakmap": {
"version": "2.0.2",
"resolved": "https://registry.npmjs.org/is-weakmap/-/is-weakmap-2.0.2.tgz",
@@ -6167,6 +6186,26 @@
"semver": "bin/semver.js"
}
},
"node_modules/node-fetch": {
"version": "2.7.0",
"resolved": "https://registry.npmjs.org/node-fetch/-/node-fetch-2.7.0.tgz",
"integrity": "sha512-c4FRfUm/dbcWZ7U+1Wq0AwCyFL+3nt2bEw05wfxSz+DWpWsitgmSgYmy2dQdWyKC1694ELPqMs/YzUSNozLt8A==",
"license": "MIT",
"dependencies": {
"whatwg-url": "^5.0.0"
},
"engines": {
"node": "4.x || >=6.0.0"
},
"peerDependencies": {
"encoding": "^0.1.0"
},
"peerDependenciesMeta": {
"encoding": {
"optional": true
}
}
},
"node_modules/node-releases": {
"version": "2.0.36",
"resolved": "https://registry.npmjs.org/node-releases/-/node-releases-2.0.36.tgz",
@@ -6315,6 +6354,15 @@
"wrappy": "1"
}
},
"node_modules/opencollective-postinstall": {
"version": "2.0.3",
"resolved": "https://registry.npmjs.org/opencollective-postinstall/-/opencollective-postinstall-2.0.3.tgz",
"integrity": "sha512-8AV/sCtuzUeTo8gQK5qDZzARrulB3egtLzFgteqB2tcT4Mw7B8Kt7JcDHmltjz6FOAHsvTevk70gZEbhM4ZS9Q==",
"license": "MIT",
"bin": {
"opencollective-postinstall": "index.js"
}
},
"node_modules/optionator": {
"version": "0.9.4",
"resolved": "https://registry.npmjs.org/optionator/-/optionator-0.9.4.tgz",
@@ -6747,6 +6795,12 @@
"url": "https://github.com/sponsors/ljharb"
}
},
"node_modules/regenerator-runtime": {
"version": "0.13.11",
"resolved": "https://registry.npmjs.org/regenerator-runtime/-/regenerator-runtime-0.13.11.tgz",
"integrity": "sha512-kY1AZVr2Ra+t+piVaJ4gxaFaReZVH40AKNo7UCX6W+dEwBo/2oZJzqfuN1qLq1oL45o56cPaTXELwrTh8Fpggg==",
"license": "MIT"
},
"node_modules/regexp.prototype.flags": {
"version": "1.5.4",
"resolved": "https://registry.npmjs.org/regexp.prototype.flags/-/regexp.prototype.flags-1.5.4.tgz",
@@ -7585,6 +7639,30 @@
"streamx": "^2.12.5"
}
},
"node_modules/tesseract.js": {
"version": "7.0.0",
"resolved": "https://registry.npmjs.org/tesseract.js/-/tesseract.js-7.0.0.tgz",
"integrity": "sha512-exPBkd+z+wM1BuMkx/Bjv43OeLBxhL5kKWsz/9JY+DXcXdiBjiAch0V49QR3oAJqCaL5qURE0vx9Eo+G5YE7mA==",
"hasInstallScript": true,
"license": "Apache-2.0",
"dependencies": {
"bmp-js": "^0.1.0",
"idb-keyval": "^6.2.0",
"is-url": "^1.2.4",
"node-fetch": "^2.6.9",
"opencollective-postinstall": "^2.0.3",
"regenerator-runtime": "^0.13.3",
"tesseract.js-core": "^7.0.0",
"wasm-feature-detect": "^1.8.0",
"zlibjs": "^0.3.1"
}
},
"node_modules/tesseract.js-core": {
"version": "7.0.0",
"resolved": "https://registry.npmjs.org/tesseract.js-core/-/tesseract.js-core-7.0.0.tgz",
"integrity": "sha512-WnNH518NzmbSq9zgTPeoF8c+xmilS8rFIl1YKbk/ptuuc7p6cLNELNuPAzcmsYw450ca6bLa8j3t0VAtq435Vw==",
"license": "Apache-2.0"
},
"node_modules/text-decoder": {
"version": "1.2.7",
"resolved": "https://registry.npmjs.org/text-decoder/-/text-decoder-1.2.7.tgz",
@@ -7655,6 +7733,12 @@
"node": ">=8.0"
}
},
"node_modules/tr46": {
"version": "0.0.3",
"resolved": "https://registry.npmjs.org/tr46/-/tr46-0.0.3.tgz",
"integrity": "sha512-N3WMsuqV66lT30CrXNbEjx4GEwlow3v6rr4mCcv6prnfwhS01rkgyFdjPNBYd9br7LpXV1+Emh01fHnq2Gdgrw==",
"license": "MIT"
},
"node_modules/ts-api-utils": {
"version": "2.5.0",
"resolved": "https://registry.npmjs.org/ts-api-utils/-/ts-api-utils-2.5.0.tgz",
@@ -7955,6 +8039,28 @@
"integrity": "sha512-EPD5q1uXyFxJpCrLnCc1nHnq3gOa6DZBocAIiI2TaSCA7VCJ1UJDMagCzIkXNsUYfD1daK//LTEQ8xiIbrHtcw==",
"license": "MIT"
},
"node_modules/wasm-feature-detect": {
"version": "1.8.0",
"resolved": "https://registry.npmjs.org/wasm-feature-detect/-/wasm-feature-detect-1.8.0.tgz",
"integrity": "sha512-zksaLKM2fVlnB5jQQDqKXXwYHLQUVH9es+5TOOHwGOVJOCeRBCiPjwSg+3tN2AdTCzjgli4jijCH290kXb/zWQ==",
"license": "Apache-2.0"
},
"node_modules/webidl-conversions": {
"version": "3.0.1",
"resolved": "https://registry.npmjs.org/webidl-conversions/-/webidl-conversions-3.0.1.tgz",
"integrity": "sha512-2JAn3z8AR6rjK8Sm8orRC0h/bcl/DqL7tRPdGZ4I1CjdF+EaMLmYxBHyXuKL849eucPFhvBoxMsflfOb8kxaeQ==",
"license": "BSD-2-Clause"
},
"node_modules/whatwg-url": {
"version": "5.0.0",
"resolved": "https://registry.npmjs.org/whatwg-url/-/whatwg-url-5.0.0.tgz",
"integrity": "sha512-saE57nupxk6v3HY35+jzBwYa0rKSy0XR8JSxZPwgLr7ys0IBzhGviA1/TUGJLmSVqs8pb9AnvICXEuOHLprYTw==",
"license": "MIT",
"dependencies": {
"tr46": "~0.0.3",
"webidl-conversions": "^3.0.0"
}
},
"node_modules/which": {
"version": "2.0.2",
"resolved": "https://registry.npmjs.org/which/-/which-2.0.2.tgz",
@@ -8237,6 +8343,15 @@
"node": "^12.22.0 || ^14.17.0 || >=16.0.0"
}
},
"node_modules/zlibjs": {
"version": "0.3.1",
"resolved": "https://registry.npmjs.org/zlibjs/-/zlibjs-0.3.1.tgz",
"integrity": "sha512-+J9RrgTKOmlxFSDHo0pI1xM6BLVUv+o0ZT9ANtCxGkjIVCCUdx9alUF8Gm+dGLKbkkkidWIHFDZHDMpfITt4+w==",
"license": "MIT",
"engines": {
"node": "*"
}
},
"node_modules/zod": {
"version": "4.3.6",
"resolved": "https://registry.npmjs.org/zod/-/zod-4.3.6.tgz",

3
package.json
View File

@@ -20,7 +20,8 @@
"node-cron": "^4.2.1",
"react": "^19.2.4",
"react-dom": "^19.2.4",
"sharp": "^0.34.5"
"sharp": "^0.34.5",
"tesseract.js": "^7.0.0"
},
"devDependencies": {
"@tailwindcss/postcss": "^4.2.2",

4
src/app/api/ai-settings/library/[id]/route.ts
View File

@@ -38,6 +38,10 @@ export async function PUT(
promptTagger: typeof body.promptTagger === 'string' ? body.promptTagger : undefined,
promptExtract: typeof body.promptExtract === 'string' ? body.promptExtract : undefined,
promptTranslate: typeof body.promptTranslate === 'string' ? body.promptTranslate : undefined,
maxTokensTag: typeof body.maxTokensTag === 'number' ? body.maxTokensTag : (body.maxTokensTag === null ? null : undefined),
maxTokensDescribe: typeof body.maxTokensDescribe === 'number' ? body.maxTokensDescribe : (body.maxTokensDescribe === null ? null : undefined),
maxTokensExtract: typeof body.maxTokensExtract === 'number' ? body.maxTokensExtract : (body.maxTokensExtract === null ? null : undefined),
maxTokensTranslate: typeof body.maxTokensTranslate === 'number' ? body.maxTokensTranslate : (body.maxTokensTranslate === null ? null : undefined),
})
return NextResponse.json(getLibraryAiOverrides(id))

18
src/app/api/ai-settings/route.ts
View File

@@ -1,6 +1,6 @@
import { NextRequest, NextResponse } from 'next/server'
import { requireAdmin } from '@/lib/auth'
import { getAiConfig, updateAiConfig, getPreferredLanguage, setPreferredLanguage, getAiMaxRetries, setAiMaxRetries } from '@/lib/app-settings'
import { getAiConfig, updateAiConfig, getPreferredLanguage, setPreferredLanguage, getAiMaxRetries, setAiMaxRetries, type OcrMode } from '@/lib/app-settings'
export async function GET(request: NextRequest) {
const auth = await requireAdmin(request)
@@ -30,6 +30,13 @@ export async function PUT(request: NextRequest) {
promptExtract?: string
promptTranslate?: string
maxRetries?: number
maxTokensTag?: number
maxTokensDescribe?: number
maxTokensExtract?: number
maxTokensTranslate?: number
ocrMode?: string
ocrLanguages?: string
ocrConfidenceThreshold?: number
}
try {
body = await request.json()
@@ -42,6 +49,8 @@ export async function PUT(request: NextRequest) {
modelTagging, modelDescribe, modelExtract, modelTranslate,
promptDescribe, promptTagger, promptExtract, promptTranslate,
maxRetries,
maxTokensTag, maxTokensDescribe, maxTokensExtract, maxTokensTranslate,
ocrMode, ocrLanguages, ocrConfidenceThreshold,
} = body
if (typeof endpoint !== 'string') {
@@ -66,6 +75,13 @@ export async function PUT(request: NextRequest) {
typeof promptTagger === 'string' ? promptTagger : undefined,
typeof promptExtract === 'string' ? promptExtract : undefined,
typeof promptTranslate === 'string' ? promptTranslate : undefined,
typeof maxTokensTag === 'number' ? maxTokensTag : undefined,
typeof maxTokensDescribe === 'number' ? maxTokensDescribe : undefined,
typeof maxTokensExtract === 'number' ? maxTokensExtract : undefined,
typeof maxTokensTranslate === 'number' ? maxTokensTranslate : undefined,
(ocrMode === 'hybrid' || ocrMode === 'tesseract' || ocrMode === 'llm') ? (ocrMode as OcrMode) : undefined,
typeof ocrLanguages === 'string' ? ocrLanguages : undefined,
typeof ocrConfidenceThreshold === 'number' ? ocrConfidenceThreshold : undefined,
)
if (typeof preferredLanguage === 'string' && preferredLanguage.trim()) {

195
src/app/manage/ai-tagging/page.tsx
View File

@@ -16,6 +16,13 @@ interface AiSettings {
promptExtract: string
promptTranslate: string
maxRetries: number
maxTokensTag: number
maxTokensDescribe: number
maxTokensExtract: number
maxTokensTranslate: number
ocrMode: 'hybrid' | 'tesseract' | 'llm'
ocrLanguages: string
ocrConfidenceThreshold: number
}
interface AiJob {
@@ -47,6 +54,10 @@ interface LibraryOverride {
promptTagger: string
promptExtract: string
promptTranslate: string
maxTokensTag: number | null
maxTokensDescribe: number | null
maxTokensExtract: number | null
maxTokensTranslate: number | null
}
function formatElapsed(startedAt: number): string {
@@ -67,6 +78,8 @@ export default function AiTaggingPage() {
enabled: false, preferredLanguage: 'English',
promptDescribe: '', promptTagger: '', promptExtract: '', promptTranslate: '',
maxRetries: 3,
maxTokensTag: 8192, maxTokensDescribe: 8192, maxTokensExtract: 8192, maxTokensTranslate: 8192,
ocrMode: 'hybrid', ocrLanguages: 'eng', ocrConfidenceThreshold: 70,
})
const [loading, setLoading] = useState(true)
const [saving, setSaving] = useState(false)
@@ -296,7 +309,7 @@ export default function AiTaggingPage() {
}
}
const updateLibraryOverride = (libraryId: string, field: keyof LibraryOverride, value: string) => {
const updateLibraryOverride = (libraryId: string, field: keyof LibraryOverride, value: string | number | null) => {
setLibraryOverrides((prev) => ({
...prev,
[libraryId]: { ...(prev[libraryId] ?? emptyOverride()), [field]: value },
@@ -544,6 +557,25 @@ export default function AiTaggingPage() {
/>
</Field>
<Field label="Tagging Max Tokens">
<input
type="number"
min={1}
value={settings.maxTokensTag}
onChange={(e) =>
setSettings((s) => ({ ...s, maxTokensTag: Math.max(1, parseInt(e.target.value) || 8192) }))
}
className="w-32 rounded-lg px-3 py-2 text-sm outline-none focus:ring-2"
style={{
backgroundColor: 'var(--background)',
border: '1px solid var(--border)',
color: 'var(--text-primary)',
}}
onFocus={(e) => ((e.currentTarget as HTMLElement).style.borderColor = 'var(--accent)')}
onBlur={(e) => ((e.currentTarget as HTMLElement).style.borderColor = 'var(--border)')}
/>
</Field>
<Field label="Description Model">
<input
type="text"
@@ -561,6 +593,25 @@ export default function AiTaggingPage() {
/>
</Field>
<Field label="Description Max Tokens">
<input
type="number"
min={1}
value={settings.maxTokensDescribe}
onChange={(e) =>
setSettings((s) => ({ ...s, maxTokensDescribe: Math.max(1, parseInt(e.target.value) || 8192) }))
}
className="w-32 rounded-lg px-3 py-2 text-sm outline-none focus:ring-2"
style={{
backgroundColor: 'var(--background)',
border: '1px solid var(--border)',
color: 'var(--text-primary)',
}}
onFocus={(e) => ((e.currentTarget as HTMLElement).style.borderColor = 'var(--accent)')}
onBlur={(e) => ((e.currentTarget as HTMLElement).style.borderColor = 'var(--border)')}
/>
</Field>
<Field label="Text Extraction Model">
<input
type="text"
@@ -578,6 +629,91 @@ export default function AiTaggingPage() {
/>
</Field>
<Field label="Text Extraction Max Tokens">
<input
type="number"
min={1}
value={settings.maxTokensExtract}
onChange={(e) =>
setSettings((s) => ({ ...s, maxTokensExtract: Math.max(1, parseInt(e.target.value) || 8192) }))
}
className="w-32 rounded-lg px-3 py-2 text-sm outline-none focus:ring-2"
style={{
backgroundColor: 'var(--background)',
border: '1px solid var(--border)',
color: 'var(--text-primary)',
}}
onFocus={(e) => ((e.currentTarget as HTMLElement).style.borderColor = 'var(--accent)')}
onBlur={(e) => ((e.currentTarget as HTMLElement).style.borderColor = 'var(--border)')}
/>
</Field>
<Field label="OCR Mode">
<div className="flex gap-2">
{(['hybrid', 'tesseract', 'llm'] as const).map((mode) => (
<button
key={mode}
type="button"
onClick={() => setSettings((s) => ({ ...s, ocrMode: mode }))}
className="px-3 py-1.5 rounded-lg text-sm transition-colors"
style={{
backgroundColor: settings.ocrMode === mode ? 'var(--accent)' : 'var(--surface)',
color: settings.ocrMode === mode ? '#fff' : 'var(--text-secondary)',
border: '1px solid var(--border)',
}}
>
{mode === 'hybrid' ? 'Hybrid' : mode === 'tesseract' ? 'Tesseract only' : 'LLM only'}
</button>
))}
</div>
<p className="mt-1 text-xs" style={{ color: 'var(--text-secondary)' }}>
Hybrid runs local OCR first and falls back to the LLM when confidence is low. Tesseract only never calls the LLM. LLM only uses the original behaviour.
</p>
</Field>
<Field label="OCR Languages">
<input
type="text"
value={settings.ocrLanguages}
onChange={(e) => setSettings((s) => ({ ...s, ocrLanguages: e.target.value }))}
placeholder="eng"
className="w-full rounded-lg px-3 py-2 text-sm font-mono outline-none focus:ring-2"
style={{
backgroundColor: 'var(--background)',
border: '1px solid var(--border)',
color: 'var(--text-primary)',
}}
onFocus={(e) => ((e.currentTarget as HTMLElement).style.borderColor = 'var(--accent)')}
onBlur={(e) => ((e.currentTarget as HTMLElement).style.borderColor = 'var(--border)')}
/>
<p className="mt-1 text-xs" style={{ color: 'var(--text-secondary)' }}>
{`Tesseract language packs to use, joined with '+'. For Japanese manga use jpn+jpn_vert. Language data is downloaded automatically on first use.`}
</p>
</Field>
<Field label="OCR Confidence Threshold">
<input
type="number"
min={0}
max={100}
value={settings.ocrConfidenceThreshold}
onChange={(e) =>
setSettings((s) => ({ ...s, ocrConfidenceThreshold: Math.max(0, Math.min(100, parseInt(e.target.value) || 70)) }))
}
className="w-24 rounded-lg px-3 py-2 text-sm outline-none focus:ring-2"
style={{
backgroundColor: 'var(--background)',
border: '1px solid var(--border)',
color: 'var(--text-primary)',
}}
onFocus={(e) => ((e.currentTarget as HTMLElement).style.borderColor = 'var(--accent)')}
onBlur={(e) => ((e.currentTarget as HTMLElement).style.borderColor = 'var(--border)')}
/>
<p className="mt-1 text-xs" style={{ color: 'var(--text-secondary)' }}>
In hybrid mode, Tesseract results below this confidence score (0–100) fall back to the LLM. Default is 70.
</p>
</Field>
<Field label="Translation Model">
<input
type="text"
@@ -595,6 +731,25 @@ export default function AiTaggingPage() {
/>
</Field>
<Field label="Translation Max Tokens">
<input
type="number"
min={1}
value={settings.maxTokensTranslate}
onChange={(e) =>
setSettings((s) => ({ ...s, maxTokensTranslate: Math.max(1, parseInt(e.target.value) || 8192) }))
}
className="w-32 rounded-lg px-3 py-2 text-sm outline-none focus:ring-2"
style={{
backgroundColor: 'var(--background)',
border: '1px solid var(--border)',
color: 'var(--text-primary)',
}}
onFocus={(e) => ((e.currentTarget as HTMLElement).style.borderColor = 'var(--accent)')}
onBlur={(e) => ((e.currentTarget as HTMLElement).style.borderColor = 'var(--border)')}
/>
</Field>
<Field label="Automatic Tagging">
<label className="flex items-center gap-3 cursor-pointer select-none">
<div
@@ -890,7 +1045,7 @@ export default function AiTaggingPage() {
<Field key={field} label={label}>
<input
type="text"
value={overrides[field]}
value={overrides[field] as string}
onChange={(e) => updateLibraryOverride(lib.id, field, e.target.value)}
placeholder={`Leave blank to use global default${settings[field as keyof AiSettings] ? ` (${settings[field as keyof AiSettings]})` : ''}`}
className="w-full rounded-lg px-3 py-2 text-sm font-mono outline-none focus:ring-2"
@@ -906,6 +1061,39 @@ export default function AiTaggingPage() {
))}
</div>
<div className="flex flex-col gap-3">
<p className="text-xs font-medium uppercase tracking-wide" style={{ color: 'var(--text-secondary)' }}>Max Tokens</p>
{(
[
['maxTokensTag', 'Tagging', 'maxTokensTag'] as const,
['maxTokensDescribe', 'Description', 'maxTokensDescribe'] as const,
['maxTokensExtract', 'Text Extraction', 'maxTokensExtract'] as const,
['maxTokensTranslate', 'Translation', 'maxTokensTranslate'] as const,
]
).map(([field, label, globalField]) => (
<Field key={field} label={label}>
<input
type="number"
min={1}
value={overrides[field] ?? ''}
placeholder={`Leave blank to use global default (${settings[globalField]})`}
onChange={(e) => {
const raw = e.target.value
updateLibraryOverride(lib.id, field, raw === '' ? null : Math.max(1, parseInt(raw) || 1))
}}
className="w-40 rounded-lg px-3 py-2 text-sm outline-none focus:ring-2"
style={{
backgroundColor: 'var(--background)',
border: '1px solid var(--border)',
color: 'var(--text-primary)',
}}
onFocus={(e) => ((e.currentTarget as HTMLElement).style.borderColor = 'var(--accent)')}
onBlur={(e) => ((e.currentTarget as HTMLElement).style.borderColor = 'var(--border)')}
/>
</Field>
))}
</div>
<div className="flex flex-col gap-3">
<p className="text-xs font-medium uppercase tracking-wide" style={{ color: 'var(--text-secondary)' }}>Prompts</p>
{(
@@ -919,7 +1107,7 @@ export default function AiTaggingPage() {
<Field key={field} label={label}>
<textarea
rows={3}
value={overrides[field]}
value={overrides[field] as string}
onChange={(e) => updateLibraryOverride(lib.id, field, e.target.value)}
placeholder={globalValue ? `Leave blank to use global default:\n${globalValue}` : 'Leave blank to use global default'}
className="w-full rounded-lg px-3 py-2 text-sm outline-none focus:ring-2 resize-y"
@@ -1010,6 +1198,7 @@ function emptyOverride(): LibraryOverride {
return {
modelTagging: '', modelDescribe: '', modelExtract: '', modelTranslate: '',
promptDescribe: '', promptTagger: '', promptExtract: '', promptTranslate: '',
maxTokensTag: null, maxTokensDescribe: null, maxTokensExtract: null, maxTokensTranslate: null,
}
}

2
src/components/mixed/ImageLightbox.tsx
View File

@@ -414,7 +414,7 @@ export default function ImageLightbox({ url, name, onClose, onPrev, onNext, item
;(e.currentTarget as HTMLElement).style.color = 'var(--text-secondary)'
}}
>
{retranslating ? '⟳ Translating…' : '🌐 Re-translate'}
{retranslating ? '⟳ Translating…' : translatedText ? '🌐 Re-translate' : '🌐 Translate'}
</button>
</div>
</div>

58
src/components/mixed/MixedView.tsx
View File

@@ -453,6 +453,18 @@ export default function MixedView({ libraryId, initialPath }: Props) {
}
}
}}
onTranslate={async (e) => {
const itemKey = itemKeyFor(e)
const res = await fetch('/api/ai-tagging/translate', {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ itemKey }),
})
if (!res.ok) {
const data = await res.json().catch(() => ({}))
throw new Error((data as { error?: string }).error ?? 'Translation failed')
}
}}
onDelete={(e) => {
const rel = filtersActive ? e.name : (currentPath ? `${currentPath}/${e.name}` : e.name)
fetch(`/api/browse?libraryId=${encodeURIComponent(libraryId)}&path=${encodeURIComponent(rel)}`, { method: 'DELETE' })
@@ -582,7 +594,7 @@ export default function MixedView({ libraryId, initialPath }: Props) {
)
}
function EntryTile({ entry, onOpen, onTag, onDelete, onRename, onAiTag, onExtractText, onDescribe }: { entry: FileEntry; onOpen: (e: FileEntry) => void; onTag: (e: FileEntry) => void; onDelete?: (e: FileEntry) => void; onRename?: (e: FileEntry, newName: string) => Promise<boolean>; onAiTag?: (e: FileEntry) => Promise<void>; onExtractText?: (e: FileEntry) => Promise<void>; onDescribe?: (e: FileEntry) => Promise<void> }) {
function EntryTile({ entry, onOpen, onTag, onDelete, onRename, onAiTag, onExtractText, onDescribe, onTranslate }: { entry: FileEntry; onOpen: (e: FileEntry) => void; onTag: (e: FileEntry) => void; onDelete?: (e: FileEntry) => void; onRename?: (e: FileEntry, newName: string) => Promise<boolean>; onAiTag?: (e: FileEntry) => Promise<void>; onExtractText?: (e: FileEntry) => Promise<void>; onDescribe?: (e: FileEntry) => Promise<void>; onTranslate?: (e: FileEntry) => Promise<void> }) {
type ImgState = 'loading' | 'loaded' | 'error'
const [imgState, setImgState] = useState<ImgState>(
entry.thumbnailUrl ? 'loading' : 'error'
@@ -601,6 +613,8 @@ function EntryTile({ entry, onOpen, onTag, onDelete, onRename, onAiTag, onExtrac
const [textExtractError, setTextExtractError] = useState<string | null>(null)
const [describing, setDescribing] = useState(false)
const [describeError, setDescribeError] = useState<string | null>(null)
const [translating, setTranslating] = useState(false)
const [translateError, setTranslateError] = useState<string | null>(null)
useEffect(() => {
if (!menuOpen) return
@@ -830,6 +844,26 @@ function EntryTile({ entry, onOpen, onTag, onDelete, onRename, onAiTag, onExtrac
🔍 Extract Text for Folder
</button>
)}
{onTranslate && entry.mediaType === 'image' && (
<button
onClick={(e) => {
e.stopPropagation()
setMenuOpen(false)
setTranslating(true)
setTranslateError(null)
onTranslate(entry)
.catch((err) => setTranslateError(err instanceof Error ? err.message : 'Translation failed'))
.finally(() => setTranslating(false))
}}
disabled={translating}
className="flex items-center gap-2 w-full px-4 py-2 text-sm text-left transition-colors disabled:opacity-50"
style={{ color: 'var(--text-primary)' }}
onMouseEnter={(e) => ((e.currentTarget as HTMLElement).style.backgroundColor = 'var(--border)')}
onMouseLeave={(e) => ((e.currentTarget as HTMLElement).style.backgroundColor = 'transparent')}
>
🌐 Translate
</button>
)}
{onRename && (
<button
onClick={(e) => {
@@ -929,6 +963,28 @@ function EntryTile({ entry, onOpen, onTag, onDelete, onRename, onAiTag, onExtrac
</div>
)}
{/* Translation status overlay */}
{(translating || translateError) && (
<div
className="absolute inset-x-0 bottom-0 z-10 px-2 py-1.5 text-xs"
style={{ backgroundColor: translateError ? 'rgba(127,29,29,0.9)' : 'rgba(0,0,0,0.75)' }}
onClick={(e) => e.stopPropagation()}
>
<span style={{ color: translateError ? '#fca5a5' : 'var(--text-secondary)' }}>
{translateError ?? 'Translating…'}
</span>
{translateError && (
<button
onClick={() => setTranslateError(null)}
className="ml-2 underline text-xs"
style={{ color: '#fca5a5' }}
>
dismiss
</button>
)}
</div>
)}
{/* Delete confirmation overlay */}
{confirming && (
<div

157
src/lib/ai-tagger.ts
View File

@@ -4,7 +4,7 @@ import type { Library, Tag, TagCategory } from '@/types'
import { getDb } from './db'
import { getAiConfig, getEffectiveAiConfig, getPreferredLanguage } from './app-settings'
import { getTags, getCategories, addTagToItem, getActiveCategoryIdsForLibrary, getResolvedTagsForItem } from './tags'
import { getAiImagePath, getVideoFramePaths } from './thumbnails'
import { getAiImagePath, getOcrImagePath, getVideoFramePaths } from './thumbnails'
import { findFile } from './media-utils'
import { getLibrary, resolveLibraryRoot } from './libraries'
@@ -171,7 +171,8 @@ async function callVisionApi(
endpoint: string,
model: string,
base64Images: string[],
systemPrompt: string
systemPrompt: string,
maxTokens: number,
): Promise<string[]> {
const url = endpoint.replace(/\/+$/, '') + '/chat/completions'
@@ -195,7 +196,7 @@ async function callVisionApi(
})),
},
],
max_tokens: 8192,
max_tokens: maxTokens,
temperature: 0.1,
}),
})
@@ -338,7 +339,7 @@ export async function tagSingleItem(itemKey: string): Promise<string[]> {
customInstruction: config.promptTagger || undefined,
})
const suggestedIds = await callVisionApi(config.endpoint, taggingModel, base64Images, systemPromptWithContext)
const suggestedIds = await callVisionApi(config.endpoint, taggingModel, base64Images, systemPromptWithContext, config.maxTokensTag)
const validIds = suggestedIds.filter((id) => validTagIds.has(id))
for (const tagId of validIds) {
@@ -359,7 +360,8 @@ async function callVisionApiText(
endpoint: string,
model: string,
base64Images: string[],
systemPrompt: string
systemPrompt: string,
maxTokens: number,
): Promise<string> {
const url = endpoint.replace(/\/+$/, '') + '/chat/completions'
@@ -383,7 +385,7 @@ async function callVisionApiText(
})),
},
],
max_tokens: 8192,
max_tokens: maxTokens,
temperature: 0.1,
}),
})
@@ -410,7 +412,8 @@ async function callChatApiText(
endpoint: string,
model: string,
systemPrompt: string,
userMessage: string
userMessage: string,
maxTokens: number,
): Promise<string> {
const url = endpoint.replace(/\/+$/, '') + '/chat/completions'
@@ -428,7 +431,7 @@ async function callChatApiText(
{ role: 'system', content: systemPrompt },
{ role: 'user', content: userMessage },
],
max_tokens: 8192,
max_tokens: maxTokens,
temperature: 0.1,
}),
})
@@ -496,7 +499,7 @@ export async function generateItemDescription(itemKey: string): Promise<string>
: ''
const systemPrompt = `You are a media cataloging assistant. Describe the given image briefly and objectively in 1-3 sentences.${config.promptDescribe ? ' ' + config.promptDescribe : ''}${tagContext}`
const description = await callVisionApiText(config.endpoint, describeModel, base64Images, systemPrompt)
const description = await callVisionApiText(config.endpoint, describeModel, base64Images, systemPrompt, config.maxTokensDescribe)
db.prepare('UPDATE media_items SET ai_description = ? WHERE item_key = ?').run(description, itemKey)
@@ -506,36 +509,38 @@ export async function generateItemDescription(itemKey: string): Promise<string>
// ─── Text extraction ─────────────────────────────────────────────────────────
/**
* Extract text (OCR) from an image using the vision model.
* Only works for images in mixed libraries.
* If the extracted text is not in the user's preferred language, auto-translates it.
* Returns { extractedText, translatedText }.
* Run Tesseract OCR on a preprocessed image file.
* Returns the extracted text and a mean confidence score (0–100).
* A confidence of 0 with empty text means no recognisable text was found.
*/
/**
* Parse a structured extraction response from the AI.
* Returns null if the response cannot be parsed as valid JSON with the expected shape.
*/
function parseStructuredExtraction(raw: string): { text: string; needsTranslation: boolean } | null {
const jsonMatch = raw.match(/\{[\s\S]*\}/)
if (!jsonMatch) return null
async function extractWithTesseract(
imagePath: string,
languages: string,
): Promise<{ text: string; confidence: number }> {
const { createWorker } = await import('tesseract.js')
const workerPath = path.join(process.cwd(), 'node_modules/tesseract.js/src/worker-script/node/index.js')
const worker = await createWorker(languages, 1, { workerPath })
try {
const parsed = JSON.parse(jsonMatch[0])
if (typeof parsed.text === 'string' && typeof parsed.needsTranslation === 'boolean') {
return { text: parsed.text, needsTranslation: parsed.needsTranslation }
const { data } = await worker.recognize(imagePath)
return { text: data.text.trim(), confidence: data.confidence }
} finally {
await worker.terminate()
}
} catch {
// fall through
}
return null
}
/**
* Extract text (OCR) from an image using the configured OCR mode:
* - hybrid: try Tesseract first; fall back to LLM if confidence is below threshold
* - tesseract: local Tesseract only, no LLM call
* - llm: LLM vision API only (original behaviour)
*
* Only works for images in mixed libraries.
* Translation is not performed automatically — call translateItemText() separately.
* Returns { extractedText, translatedText } where translatedText is always null.
*/
export async function extractItemText(itemKey: string): Promise<{ extractedText: string; translatedText: string | null }> {
const libraryId = itemKey.split(':')[0]
const config = getEffectiveAiConfig(libraryId)
const extractModel = config.modelExtract || config.model
if (!config.endpoint || !extractModel) {
throw Object.assign(new Error('AI endpoint and model are not configured'), { code: 'NOT_CONFIGURED' })
}
const db = getDb()
const item = db
@@ -562,72 +567,49 @@ export async function extractItemText(itemKey: string): Promise<{ extractedText:
throw Object.assign(new Error('Text extraction is only available for images'), { code: 'NO_IMAGE' })
}
const { ocrMode, ocrLanguages, ocrConfidenceThreshold } = config
// ── Tesseract path ────────────────────────────────────────────────────────
if (ocrMode === 'tesseract' || ocrMode === 'hybrid') {
const ocrImagePath = await getOcrImagePath(resolvedMedia.path, libraryId)
const { text, confidence } = await extractWithTesseract(ocrImagePath, ocrLanguages)
const useTesseractResult = ocrMode === 'tesseract' || confidence >= ocrConfidenceThreshold
if (useTesseractResult) {
console.log(`[ocr] tesseract used for ${itemKey} (confidence=${confidence}, mode=${ocrMode})`)
if (!text) {
db.prepare('UPDATE media_items SET extracted_text = NULL, extracted_text_translated = NULL WHERE item_key = ?').run(itemKey)
return { extractedText: '', translatedText: null }
}
db.prepare('UPDATE media_items SET extracted_text = ?, extracted_text_translated = NULL WHERE item_key = ?').run(text, itemKey)
return { extractedText: text, translatedText: null }
}
console.log(`[ocr] tesseract confidence too low (${confidence} < ${ocrConfidenceThreshold}), falling back to LLM for ${itemKey}`)
}
// ── LLM vision path ───────────────────────────────────────────────────────
const extractModel = config.modelExtract || config.model
if (!config.endpoint || !extractModel) {
throw Object.assign(new Error('AI endpoint and model are not configured'), { code: 'NOT_CONFIGURED' })
}
const thumbnailPath = await getAiImagePath(resolvedMedia.path, libraryId)
const base64Images = [fs.readFileSync(thumbnailPath, 'base64')]
const preferredLanguage = getPreferredLanguage()
const customInstruction = config.promptExtract ? ' ' + config.promptExtract : ''
const systemPrompt = `You are an OCR assistant. Extract ALL text visible in the image exactly as it appears. Preserve line breaks and formatting.${customInstruction} If there is no text in the image, respond with exactly: [NO TEXT]`
// When a preferred language is configured, ask the AI to also flag whether translation is needed.
// This avoids a separate translation API call for text already in the target language.
let systemPrompt: string
if (preferredLanguage) {
systemPrompt = `You are an OCR assistant. Extract ALL text visible in the image exactly as it appears. Preserve line breaks and formatting.${customInstruction}
Respond ONLY with a valid JSON object — no markdown, no explanation:
{"needsTranslation": boolean, "text": "extracted text"}
Rules:
- Set needsTranslation to true if the text is NOT already written in ${preferredLanguage}.
- Set needsTranslation to false if the text IS in ${preferredLanguage}, or if there is no text.
- If there is no text in the image, use exactly: {"needsTranslation": false, "text": "[NO TEXT]"}`
} else {
systemPrompt = `You are an OCR assistant. Extract ALL text visible in the image exactly as it appears. Preserve line breaks and formatting.${customInstruction} If there is no text in the image, respond with exactly: [NO TEXT]`
}
const rawResponse = await callVisionApiText(config.endpoint, extractModel, base64Images, systemPrompt)
// Parse the response — structured JSON when a preferred language is set, plain text otherwise
let extractedText: string
let needsTranslation: boolean
if (preferredLanguage) {
const parsed = parseStructuredExtraction(rawResponse)
if (parsed) {
extractedText = parsed.text
needsTranslation = parsed.needsTranslation
} else {
// Malformed JSON fallback: treat raw response as plain text and attempt translation
extractedText = rawResponse
needsTranslation = true
}
} else {
extractedText = rawResponse
needsTranslation = false
}
console.log(`[ocr] llm used for ${itemKey} (mode=${ocrMode})`)
const extractedText = await callVisionApiText(config.endpoint, extractModel, base64Images, systemPrompt, config.maxTokensExtract)
if (!extractedText || extractedText === '[NO TEXT]') {
db.prepare('UPDATE media_items SET extracted_text = NULL, extracted_text_translated = NULL WHERE item_key = ?').run(itemKey)
return { extractedText: '', translatedText: null }
}
db.prepare('UPDATE media_items SET extracted_text = ? WHERE item_key = ?').run(extractedText, itemKey)
db.prepare('UPDATE media_items SET extracted_text = ?, extracted_text_translated = NULL WHERE item_key = ?').run(extractedText, itemKey)
// Only translate if the extraction step determined the text is not already in the preferred language
let translatedText: string | null = null
if (preferredLanguage && needsTranslation) {
const translateModel = config.modelTranslate || config.model
try {
translatedText = await translateText(config.endpoint, translateModel, extractedText, preferredLanguage, config.promptTranslate)
if (translatedText) {
db.prepare('UPDATE media_items SET extracted_text_translated = ? WHERE item_key = ?').run(translatedText, itemKey)
}
} catch (err) {
console.warn(`[ai-tagger] Translation failed for "${itemKey}":`, err instanceof Error ? err.message : err)
}
}
return { extractedText, translatedText }
return { extractedText, translatedText: null }
}
/**
@@ -656,7 +638,7 @@ export async function translateItemText(itemKey: string, sourceLanguage?: string
const preferredLanguage = getPreferredLanguage()
if (!preferredLanguage) return null
const translatedText = await translateText(config.endpoint, translateModel, row.extracted_text, preferredLanguage, config.promptTranslate, sourceLanguage)
const translatedText = await translateText(config.endpoint, translateModel, row.extracted_text, preferredLanguage, config.promptTranslate, config.maxTokensTranslate, sourceLanguage)
if (translatedText) {
db.prepare('UPDATE media_items SET extracted_text_translated = ? WHERE item_key = ?').run(translatedText, itemKey)
}
@@ -682,6 +664,7 @@ async function translateText(
text: string,
targetLanguage: string,
customInstruction = '',
maxTokens = 8192,
sourceLanguage?: string,
): Promise<string | null> {
let systemPrompt: string
@@ -691,7 +674,7 @@ async function translateText(
systemPrompt = `You are a translator. Determine if the following text is already in ${targetLanguage}. If it is, respond with exactly: [ALREADY_TARGET_LANGUAGE]. If it is not, translate it to ${targetLanguage}.${customInstruction ? ' ' + customInstruction : ''}`
}
const result = await callChatApiText(endpoint, model, systemPrompt, text)
const result = await callChatApiText(endpoint, model, systemPrompt, text, maxTokens)
if (!sourceLanguage && (result === '[ALREADY_TARGET_LANGUAGE]' || !result)) {
return null

72
src/lib/app-settings.ts
View File

@@ -46,6 +46,8 @@ const DEFAULT_PROMPT_EXTRACT =
'Be mindful of different colors of text that may indicate different speakers or emphasis.'
const DEFAULT_PROMPT_TRANSLATE = 'Return ONLY the translated text with no additional commentary.'
export type OcrMode = 'hybrid' | 'tesseract' | 'llm'
export interface AiConfig {
endpoint: string
model: string
@@ -58,6 +60,13 @@ export interface AiConfig {
promptTagger: string
promptExtract: string
promptTranslate: string
maxTokensTag: number
maxTokensDescribe: number
maxTokensExtract: number
maxTokensTranslate: number
ocrMode: OcrMode
ocrLanguages: string
ocrConfidenceThreshold: number
}
export function getAiConfig(): AiConfig {
@@ -76,9 +85,19 @@ export function getAiConfig(): AiConfig {
const promptExtract = promptExtractRaw !== null ? promptExtractRaw : DEFAULT_PROMPT_EXTRACT
const promptTranslateRaw = getSetting('ai_prompt_translate')
const promptTranslate = promptTranslateRaw !== null ? promptTranslateRaw : DEFAULT_PROMPT_TRANSLATE
const maxTokensTag = parseInt(getSetting('ai_max_tokens_tag') ?? '8192', 10) || 8192
const maxTokensDescribe = parseInt(getSetting('ai_max_tokens_describe') ?? '8192', 10) || 8192
const maxTokensExtract = parseInt(getSetting('ai_max_tokens_extract') ?? '8192', 10) || 8192
const maxTokensTranslate = parseInt(getSetting('ai_max_tokens_translate') ?? '8192', 10) || 8192
const rawOcrMode = getSetting('ai_ocr_mode') ?? 'hybrid'
const ocrMode: OcrMode = rawOcrMode === 'tesseract' || rawOcrMode === 'llm' ? rawOcrMode : 'hybrid'
const ocrLanguages = getSetting('ai_ocr_languages') ?? 'eng'
const ocrConfidenceThreshold = parseInt(getSetting('ai_ocr_confidence_threshold') ?? '70', 10) || 70
return {
endpoint, model, modelTagging, modelDescribe, modelExtract, modelTranslate, enabled,
promptDescribe, promptTagger, promptExtract, promptTranslate,
maxTokensTag, maxTokensDescribe, maxTokensExtract, maxTokensTranslate,
ocrMode, ocrLanguages, ocrConfidenceThreshold,
}
}
@@ -94,6 +113,13 @@ export function updateAiConfig(
promptTagger?: string,
promptExtract?: string,
promptTranslate?: string,
maxTokensTag?: number,
maxTokensDescribe?: number,
maxTokensExtract?: number,
maxTokensTranslate?: number,
ocrMode?: OcrMode,
ocrLanguages?: string,
ocrConfidenceThreshold?: number,
): void {
setSetting('ai_endpoint', endpoint)
setSetting('ai_model', model)
@@ -106,6 +132,13 @@ export function updateAiConfig(
if (promptTagger !== undefined) setSetting('ai_prompt_tagger', promptTagger)
if (promptExtract !== undefined) setSetting('ai_prompt_extract', promptExtract)
if (promptTranslate !== undefined) setSetting('ai_prompt_translate', promptTranslate)
if (maxTokensTag !== undefined) setSetting('ai_max_tokens_tag', String(Math.max(1, Math.floor(maxTokensTag))))
if (maxTokensDescribe !== undefined) setSetting('ai_max_tokens_describe', String(Math.max(1, Math.floor(maxTokensDescribe))))
if (maxTokensExtract !== undefined) setSetting('ai_max_tokens_extract', String(Math.max(1, Math.floor(maxTokensExtract))))
if (maxTokensTranslate !== undefined) setSetting('ai_max_tokens_translate', String(Math.max(1, Math.floor(maxTokensTranslate))))
if (ocrMode !== undefined) setSetting('ai_ocr_mode', ocrMode)
if (ocrLanguages !== undefined) setSetting('ai_ocr_languages', ocrLanguages.trim() || 'eng')
if (ocrConfidenceThreshold !== undefined) setSetting('ai_ocr_confidence_threshold', String(Math.max(0, Math.min(100, Math.floor(ocrConfidenceThreshold)))))
}
export function getPreferredLanguage(): string {
@@ -127,6 +160,10 @@ export interface LibraryAiOverrides {
promptTagger: string
promptExtract: string
promptTranslate: string
maxTokensTag: number | null
maxTokensDescribe: number | null
maxTokensExtract: number | null
maxTokensTranslate: number | null
}
interface LibraryAiSettingsRow {
@@ -138,6 +175,10 @@ interface LibraryAiSettingsRow {
prompt_tagger: string | null
prompt_extract: string | null
prompt_translate: string | null
max_tokens_tag: number | null
max_tokens_describe: number | null
max_tokens_extract: number | null
max_tokens_translate: number | null
}
export function getLibraryAiOverrides(libraryId: string): LibraryAiOverrides {
@@ -154,6 +195,10 @@ export function getLibraryAiOverrides(libraryId: string): LibraryAiOverrides {
promptTagger: row?.prompt_tagger ?? '',
promptExtract: row?.prompt_extract ?? '',
promptTranslate: row?.prompt_translate ?? '',
maxTokensTag: row?.max_tokens_tag ?? null,
maxTokensDescribe: row?.max_tokens_describe ?? null,
maxTokensExtract: row?.max_tokens_extract ?? null,
maxTokensTranslate: row?.max_tokens_translate ?? null,
}
}
@@ -164,7 +209,7 @@ export function setLibraryAiOverrides(libraryId: string, overrides: Partial<Libr
'INSERT OR IGNORE INTO library_ai_settings (library_id) VALUES (?)'
).run(libraryId)
const fields: Record<string, string | undefined> = {
const stringFields: Record<string, string | undefined> = {
model_tagging: overrides.modelTagging,
model_describe: overrides.modelDescribe,
model_extract: overrides.modelExtract,
@@ -175,7 +220,7 @@ export function setLibraryAiOverrides(libraryId: string, overrides: Partial<Libr
prompt_translate: overrides.promptTranslate,
}
for (const [col, val] of Object.entries(fields)) {
for (const [col, val] of Object.entries(stringFields)) {
if (val !== undefined) {
db.prepare(`UPDATE library_ai_settings SET ${col} = ? WHERE library_id = ?`).run(
val === '' ? null : val,
@@ -183,6 +228,22 @@ export function setLibraryAiOverrides(libraryId: string, overrides: Partial<Libr
)
}
}
const numberFields: Record<string, number | null | undefined> = {
max_tokens_tag: overrides.maxTokensTag,
max_tokens_describe: overrides.maxTokensDescribe,
max_tokens_extract: overrides.maxTokensExtract,
max_tokens_translate: overrides.maxTokensTranslate,
}
for (const [col, val] of Object.entries(numberFields)) {
if (val !== undefined) {
db.prepare(`UPDATE library_ai_settings SET ${col} = ? WHERE library_id = ?`).run(
val === null ? null : Math.max(1, Math.floor(val)),
libraryId,
)
}
}
}
export function getEffectiveAiConfig(libraryId: string): AiConfig {
@@ -200,6 +261,13 @@ export function getEffectiveAiConfig(libraryId: string): AiConfig {
promptTagger: overrides.promptTagger || global.promptTagger,
promptExtract: overrides.promptExtract || global.promptExtract,
promptTranslate: overrides.promptTranslate || global.promptTranslate,
maxTokensTag: overrides.maxTokensTag ?? global.maxTokensTag,
maxTokensDescribe: overrides.maxTokensDescribe ?? global.maxTokensDescribe,
maxTokensExtract: overrides.maxTokensExtract ?? global.maxTokensExtract,
maxTokensTranslate: overrides.maxTokensTranslate ?? global.maxTokensTranslate,
ocrMode: global.ocrMode,
ocrLanguages: global.ocrLanguages,
ocrConfidenceThreshold: global.ocrConfidenceThreshold,
}
}

17
src/lib/db.ts
View File

@@ -119,6 +119,10 @@ function seedAppSettings(db: Database.Database): void {
ai_model: '',
preferred_language: 'English',
ai_max_retries: '3',
ai_max_tokens_tag: '8192',
ai_max_tokens_describe: '8192',
ai_max_tokens_extract: '8192',
ai_max_tokens_translate: '8192',
}
const insert = db.prepare(
'INSERT OR IGNORE INTO app_settings (key, value) VALUES (?, ?)'
@@ -276,6 +280,19 @@ function migrateLibraryAiSettings(db: Database.Database): void {
prompt_translate TEXT
);
`)
// Add max_tokens columns if they don't exist yet
const row = db
.prepare("SELECT sql FROM sqlite_master WHERE type='table' AND name='library_ai_settings'")
.get() as { sql: string } | undefined
if (row && !row.sql.includes('max_tokens_tag')) {
db.exec(`
ALTER TABLE library_ai_settings ADD COLUMN max_tokens_tag INTEGER;
ALTER TABLE library_ai_settings ADD COLUMN max_tokens_describe INTEGER;
ALTER TABLE library_ai_settings ADD COLUMN max_tokens_extract INTEGER;
ALTER TABLE library_ai_settings ADD COLUMN max_tokens_translate INTEGER;
`)
}
}
function migrateLibrariesType(db: Database.Database): void {

31
src/lib/thumbnails.ts
View File

@@ -60,6 +60,19 @@ async function generateAiImage(src: string, dest: string): Promise<void> {
fs.renameSync(tmp, dest)
}
/** Generate a grayscale, contrast-normalised PNG for local OCR (Tesseract).
* PNG is lossless and avoids JPEG artefacts that can degrade OCR accuracy. */
async function generateOcrImage(src: string, dest: string): Promise<void> {
const tmp = dest + '.tmp'
await sharp(src)
.resize(AI_IMAGE_WIDTH, undefined, { withoutEnlargement: true })
.grayscale()
.normalise()
.png()
.toFile(tmp)
fs.renameSync(tmp, dest)
}
/** Run a child process and collect stderr. Resolves on exit code 0, rejects otherwise. */
function run(bin: string, args: string[]): Promise<void> {
return new Promise((resolve, reject) => {
@@ -190,6 +203,24 @@ export async function getAiImagePath(
return cacheFile
}
/**
* Returns the absolute path to a preprocessed PNG suitable for local OCR.
* The image is converted to grayscale and contrast-normalised for better
* Tesseract accuracy. Cached with an `_ocr` suffix.
*/
export async function getOcrImagePath(
absoluteFilePath: string,
libraryId: string
): Promise<string> {
ensureCacheDir()
const key = cacheKey(libraryId, absoluteFilePath)
const cacheFile = path.join(CACHE_DIR, key + '_ocr.png')
const cached = getCachedPath(cacheFile, absoluteFilePath)
if (cached) return cached
await generateOcrImage(absoluteFilePath, cacheFile)
return cacheFile
}
/**
* Returns the absolute path to a cached thumbnail JPEG for the given file.
* Generates it on first call (or when the source has been modified).