1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
|
import type { ZTagStyle } from "./types/users";
import { getTagStylePrompt } from "./utils/tag";
/**
* Remove duplicate whitespaces to avoid tokenization issues
*/
function preprocessContent(content: string) {
return content.replace(/(\s){10,}/g, "$1");
}
export function buildImagePrompt(
lang: string,
customPrompts: string[],
tagStyle: ZTagStyle,
) {
const tagStyleInstruction = getTagStylePrompt(tagStyle);
return `
You are an expert whose responsibility is to help with automatic text tagging for a read-it-later/bookmarking app.
Analyze the attached image and suggest relevant tags that describe its key themes, topics, and main ideas. The rules are:
- Aim for a variety of tags, including broad categories, specific keywords, and potential sub-genres.
- The tags must be in ${lang}.
- If the tag is not generic enough, don't include it.
- Aim for 10-15 tags.
- If there are no good tags, don't emit any.
${tagStyleInstruction}
${customPrompts && customPrompts.map((p) => `- ${p}`).join("\n")}
You must respond in valid JSON with the key "tags" and the value is list of tags. Don't wrap the response in a markdown code.`;
}
/**
* Construct tagging prompt for text content
*/
export function constructTextTaggingPrompt(
lang: string,
customPrompts: string[],
content: string,
tagStyle: ZTagStyle,
): string {
const tagStyleInstruction = getTagStylePrompt(tagStyle);
return `
You are an expert whose responsibility is to help with automatic tagging for a read-it-later/bookmarking app.
Analyze the TEXT_CONTENT below and suggest relevant tags that describe its key themes, topics, and main ideas. The rules are:
- Aim for a variety of tags, including broad categories, specific keywords, and potential sub-genres.
- The tags must be in ${lang}.
- If the tag is not generic enough, don't include it.
- Do NOT generate tags related to:
- An error page (404, 403, blocked, not found, dns errors)
- Boilerplate content (cookie consent, login walls, GDPR notices)
- Aim for 3-5 tags.
- If there are no good tags, leave the array empty.
${tagStyleInstruction}
${customPrompts && customPrompts.map((p) => `- ${p}`).join("\n")}
<TEXT_CONTENT>
${content}
</TEXT_CONTENT>
You must respond in JSON with the key "tags" and the value is an array of string tags.`;
}
/**
* Construct summary prompt
*/
export function constructSummaryPrompt(
lang: string,
customPrompts: string[],
content: string,
): string {
return `
Summarize the following content responding ONLY with the summary. You MUST follow the following rules:
- Summary must be in 3-4 sentences.
- The summary must be in ${lang}.
${customPrompts && customPrompts.map((p) => `- ${p}`).join("\n")}
${content}`;
}
/**
* Build text tagging prompt without truncation (for previews/UI)
*/
export function buildTextPromptUntruncated(
lang: string,
customPrompts: string[],
content: string,
tagStyle: ZTagStyle,
): string {
return constructTextTaggingPrompt(
lang,
customPrompts,
preprocessContent(content),
tagStyle,
);
}
/**
* Build summary prompt without truncation (for previews/UI)
*/
export function buildSummaryPromptUntruncated(
lang: string,
customPrompts: string[],
content: string,
): string {
return constructSummaryPrompt(
lang,
customPrompts,
preprocessContent(content),
);
}
/**
* Build OCR prompt for extracting text from images using LLM
*/
export function buildOCRPrompt(): string {
return `You are an OCR (Optical Character Recognition) expert. Your task is to extract ALL text from this image.
Rules:
- Extract every piece of text visible in the image, including titles, body text, captions, labels, watermarks, and any other textual content.
- Preserve the original structure and formatting as much as possible (e.g., paragraphs, lists, headings).
- If text appears in multiple columns, read from left to right, top to bottom.
- If text is partially obscured or unclear, make your best attempt and indicate uncertainty with [unclear] if needed.
- Do not add any commentary, explanations, or descriptions of non-text elements.
- If there is no text in the image, respond with an empty string.
- Output ONLY the extracted text, nothing else.`;
}
|
