Spring AI Movie Poster Copywriter

Spring AI Movie Poster Copywriter

Workflow

1. Collect request payload:

• Require movieTitle and movieDescription.
• Accept optional year, genre, and tone.

2. Retrieve context:

• Query Douban and encyclopedia pages using search first, then fetch detail pages.
• Extract only concise facts: genre, setting, themes, cast keywords, awards, and plot hooks.
• Ignore user-generated long comments and spoiler-heavy content.

3. Normalize context:

• Convert raw HTML into a compact context block under 1200 Chinese characters.
• Keep only deduplicated facts and keywords.
• Store retrieval source URLs for traceability.

4. Generate candidates:

• Load prompt templates from references/prompt-template.md.
• Ask model to generate 8-10 candidates first.
• Keep temperature moderate (0.6 to 0.8) to balance creativity and control.

5. Validate and filter:

• Run scripts/validate_slogans.py against candidate lines.
• Keep final 3-5 lines that satisfy length and uniqueness checks.
• If fewer than 3 survive, regenerate with stricter prompt constraints.

6. Return API response:

• Follow references/output-schema.json.
• Return final lines plus evidence URLs from Douban/encyclopedia retrieval.

Guardrails

• Respect robots and rate limits; cache page fetches by URL.
• Avoid sensitive, discriminatory, or explicit wording in slogans.
• Keep language punchy and cinematic; avoid spoilers and plot twists.
• Always validate line length in code, not by model assumption.

Resources

• references/prompt-template.md: system/user prompt templates.
• references/source-extraction.md: context extraction field rules.
• references/output-schema.json: API output contract.
• scripts/validate_slogans.py: strict 12-Chinese-character validator.

Source: microwind/design-patterns — distributed by TomeVault.