交易员必备的 AI 预测K线走势工具
Kline-de-pre is an AI-powered financial time series analysis tool. Upload any K-line screenshot and get deep learning predictions using a three-stage pipeline: OpenCV image recognition → Transformer trend extraction → Diffusion model synthesis.
上传K线截图，AI自动预测未来K线走势。支持股票、加密货币、期货等所有K线图。深度学习驱动，开源可自部署。
三阶段深度学习AI预测流水线
Each inference passes through three specialized models for maximum prediction quality.
OpenCV.js loads the screenshot, converts to HSV color space, extracts green/red pixel masks, runs contour detection, and reconstructs OHLC data via projection methods.
60 candles transformed via GAF (Gramian Angular Field) into a 12-channel polar feature tensor. Transformer ONNX model infers DCT coefficients and a 128-dim feature vector.
1000-step DDPM reverse diffusion. UNet ONNX model predicts noise conditioned on DCT + features, iteratively denoising to generate complete OHLC predictions.
交易员实战应用场景
加密货币K线预测
Screenshot BTC, ETH, or any altcoin chart. AI predicts the next candlestick pattern using deep learning. Works across all timeframes — 1m, 5m, 15m, 1h, 4h, daily.
股票市场AI分析
Upload K-line screenshots from any stock trading platform. The AI financial analysis tool recognizes candlestick patterns and predicts trend continuation or reversal.
多时间框架确认
Run predictions on 1m, 5m, and 15m charts simultaneously. When all timeframes align, confidence in the predicted K-line trend increases significantly.
自动盯盘预测
Built-in BTC MarketWatch page with hourly auto-predictions. No manual screenshots needed — the system fetches live data and runs AI K-line prediction automatically.
核心功能
⚠ Disclaimer / 免责声明: Predictions reflect probabilistic extrapolation of historical patterns and do NOT constitute investment advice. Financial markets are highly uncertain. Always combine with fundamentals, market sentiment, and risk management. 预测结果仅反映历史形态的概率性外推，不构成投资建议。请务必结合基本面和风险管理。
价格方案
Pay per inference, or subscribe for higher limits.
Open Source / 开源 — Kline-de-pre's full source code is available on GitHub. Self-deploy for free with git clone and npm install. 源代码完全开源，可免费自行部署。
Kline-de-pre is an AI-powered financial time series analysis and prediction tool. Upload a K-line screenshot from any trading platform and the system uses a three-stage deep learning pipeline (OpenCV + Transformer + Diffusion) to predict future candlestick patterns.
Two core ONNX models: a Transformer that inputs GAF (Gramian Angular Field) feature tensors [1,12,60,60] and outputs DCT coefficients, and a Diffusion UNet that uses 1000-step DDPM reverse diffusion to generate OHLC residuals. Both support WebGL in-browser and Node.js server-side inference.
Financial markets are highly uncertain. Kline-de-pre reflects probabilistic extrapolation of historical patterns and does NOT constitute investment advice. Always combine with fundamentals, market sentiment, and risk management.
Online version: $1 per inference call. Pro ($19.90/mo): 100 inferences per 5 hours (no monthly limit). Gen ($49.90/mo): 400 per 5 hours + 2,000 monthly. The source code is open for free self-deployment.
The frontend screenshot detection works for any K-line chart regardless of trading pair. The MarketWatch backend is currently configured for BTC/USD via CryptoCompare API, but can be modified to support other pairs.
Chinese FAQ
Kline-de-pre是一款AI驱动的金融时间序列分析与K线预测工具。只需上传任意交易平台的K线截图，系统即可通过三阶段深度学习流水线（OpenCV图像识别 → Transformer趋势提取 → Diffusion扩散模型合成）自动预测未来K线走势。适用于股票、加密货币、期货等任何有K线图的金融产品，是交易员必备的AI金融分析工具。
Kline-de-pre的预测反映的是历史K线形态的概率性外推，不构成投资建议。金融市场具有高度不确定性，AI预测K线走势仅作为辅助分析参考，请务必结合基本面分析、市场情绪判断和风险管理策略做出交易决策。这也是为什么Kline-de-pre支持多时间框架分析——多个时间框架的预测一致时，可信度更高。
前端的K线截图检测功能适用于任何交易对的K线图表，无论是BTC/USD、ETH/USD等加密货币，还是A股、港股、美股等股票，甚至期货和外汇。只要K线图使用绿涨红跌的配色方案，上传截图后AI即可识别并预测K线走势。MarketWatch后端目前配置为BTC/USD自动预测，但开源代码可以修改以支持其他交易对。
在线版：每次AI推理调用 $1（按次付费，无订阅要求）。Pro专业版（$19.90/月）：每5小时100次K线推理，无月度上限，非常适合活跃交易者。Gen全能版（$49.90/月）：每5小时400次 + 每月2000次K线推理，同时涵盖Laintas平台所有产品。源代码在GitHub完全开源，技术用户可以免费克隆代码自行部署。
Kline-de-pre使用三阶段深度学习流水线：第一阶段，OpenCV.js将K线截图转换为HSV色彩空间，提取绿色/红色像素蒙版，通过轮廓检测识别蜡烛图形态并重建OHLC数据；第二阶段，60根K线通过GAF（格拉米角场）编码转换为12通道极坐标特征张量，Transformer ONNX模型推断DCT系数和128维特征向量；第三阶段，1000步DDPM反向扩散过程，Diffusion UNet模型预测噪声并逐步去噪，生成完整的OHLC预测数据。
可以。Kline-de-pre支持双运行时：浏览器端通过ONNX Runtime Web利用WebGPU/WebGL加速推理，无需安装任何软件；服务端通过Node.js运行，适合批量预测和自动化交易分析。浏览器推理不需要独立GPU——WebGPU会利用你电脑的集成显卡进行加速。