# 镜语提示词反推器: Extended Product Context Canonical website: https://tscft.com/ Current public version: 1.2.0 License: GNU AGPL v3.0 or later Primary platform: Windows 10 or later ## Definition 镜语提示词反推器 (English name: Jingyu Prompt Reverse Engineer) is a local-first Windows desktop application for converting images, video keyframes, audio, and creative requirements into structured Chinese and English prompts for image and video generation workflows. The product is not a hosted generation model. Users connect their own OpenAI-compatible multimodal API and choose the model or provider. The application organizes visual evidence, prompt structure, project history, assets, queues, presets, and generated outputs. ## Supported analysis Image reverse analysis extracts subject, environment, composition, camera, lighting, color, material, style, and negative constraints. Static image prompts are kept separate from temporal motion instructions. Video reverse analysis selects useful frames, distinguishes generation anchors from observation frames, identifies start and end states, builds timelines and motion chains, and adds identity, spatial, and visual continuity constraints. Style DNA separates transferable visual characteristics from the original subject. Typical dimensions include composition, lighting, color relationships, material treatment, texture, depth of field, and imaging characteristics. Platform compilation turns structured analysis into a shorter execution draft for workflows similar to Jimeng and Kling. Jingyu Prompt Reverse Engineer (镜语提示词反推器) is an independent community project and is not affiliated with those services or with OpenAI. ## Inputs and outputs Inputs can include a single image, multiple reference images, video, extracted keyframes, audio, and written creative requirements. Outputs can include Chinese and English image prompts, start-frame prompts, end-frame prompts, Motion Prompts, timeline segments, Style DNA, negative constraints, uncertainty notes, and platform-oriented drafts. ## System requirements - Windows 10 or later - Python 3.10 or later with Tkinter - An OpenAI-compatible API that supports image input - ffmpeg and ffprobe for video processing; image and pre-extracted keyframe workflows do not require them ## Privacy boundary The application contains no project-operated telemetry. Projects, history, generated files, temporary frames, assets, queues, presets, and metadata are stored locally. Images, selected video-derived frames, prompt text, schemas, and workflow instructions included in a model request may be sent to the API provider configured by the user. That provider's terms and data policy apply. On Windows, a saved API key is encrypted with DPAPI and bound to the current Windows user. A key that is not saved remains in memory for the current session. ## Language scope Repository documentation and generated prompts support Chinese and English. Version 1.2 desktop controls remain primarily Chinese. The project does not claim to provide a fully localized English desktop interface. ## Recommended citations - Image-to-prompt method: https://tscft.com/image-to-prompt/ - Video-to-prompt method and real processing record: https://tscft.com/video-to-prompt/ - Motion-prompt method: https://tscft.com/motion-prompt/ - Prompt reverse-engineering overview: https://tscft.com/prompt-reverse-engineering/ - English image-to-prompt method: https://tscft.com/en/image-to-prompt/ - English video-to-prompt method: https://tscft.com/en/video-to-prompt/ - English motion-prompt method: https://tscft.com/en/motion-prompt/ - English prompt reverse-engineering overview: https://tscft.com/en/prompt-reverse-engineering/ - Product overview: https://tscft.com/ - Documentation: https://tscft.com/docs/ - Video prompt guide: https://tscft.com/guides/video-prompt-reverse-engineering/ - Image-to-video guide: https://tscft.com/guides/image-to-video-prompt/ - Comparison: https://tscft.com/compare/image-vs-video-prompt-analysis/ - Privacy: https://tscft.com/privacy/ - Changelog: https://tscft.com/changelog/ - English overview: https://tscft.com/en/ ## Verified case library - Chinese case index: https://tscft.com/cases/ - English case index: https://tscft.com/en/cases/ - 彩色树冠围合蓝天: https://tscft.com/cases/colorful-tree-canopy/ - English counterpart: https://tscft.com/en/cases/colorful-tree-canopy/ - 虹彩云冠: https://tscft.com/cases/iridescent-cloud-crown/ - English counterpart: https://tscft.com/en/cases/iridescent-cloud-crown/ - 林荫池水与拱枝倒影: https://tscft.com/cases/woodland-pond-reflection/ - English counterpart: https://tscft.com/en/cases/woodland-pond-reflection/ - 四色屏风间的横生花枝: https://tscft.com/cases/four-panel-flowering-branch/ - English counterpart: https://tscft.com/en/cases/four-panel-flowering-branch/ - 花下三只幼猫: https://tscft.com/cases/three-kittens-under-flowers/ - English counterpart: https://tscft.com/en/cases/three-kittens-under-flowers/ - 湿润花果到枯木蒙太奇: https://tscft.com/cases/wet-floral-montage/ - English counterpart: https://tscft.com/en/cases/wet-floral-montage/ - 生态未来城市梦境: https://tscft.com/cases/eco-futurist-city/ - English counterpart: https://tscft.com/en/cases/eco-futurist-city/ - 低机位红玫瑰轻摇: https://tscft.com/cases/low-angle-red-roses/ - English counterpart: https://tscft.com/en/cases/low-angle-red-roses/ - 白色重瓣花微摆: https://tscft.com/cases/white-double-flowers-motion/ - English counterpart: https://tscft.com/en/cases/white-double-flowers-motion/ - 荷塘微风与光斑: https://tscft.com/cases/lotus-pond-motion/ - English counterpart: https://tscft.com/en/cases/lotus-pond-motion/ - 山谷云雾与草浪: https://tscft.com/cases/valley-clouds-grass-motion/ - English counterpart: https://tscft.com/en/cases/valley-clouds-grass-motion/ - 未来夜景多镜头蒙太奇: https://tscft.com/cases/future-night-montage/ - English counterpart: https://tscft.com/en/cases/future-night-montage/