Description
The hard part of posting on Xiaohongshu isn't typing, it's pushing one topic into a full set of images: finding the selling point, dissecting hits, picking a title and laying out 9:16 cards, bouncing between tools at every step. Install xhs-writer-skill in Claude Code or OpenClaw, describe the topic in one sentence, and it produces the card set, copy and tags in one go.
The official demo promotes a GitHub project: give it the repo link and screenshots, and it scans the material, searches comparable hits, scores selling points, turns the real screenshots into Xiaohongshu-style cards with image-to-image, and ends with 7 cards plus a caption and hashtags.
Selling-point analysis: Scores points on scarcity, usefulness and perceptibility, then focuses on the strongest one.
Hit teardown: Searches similar high-performing notes and breaks down title patterns, card layouts and hooks.
Image-to-image: Keeps your real material and only restyles the text and layout, instead of inventing a fake image.
Five title styles: One option each for pain point, question, discovery, trending word and identity.
Two formats: 9:16 card sets, or a short-video script, saved to disk in a fixed layout.
Image generation uses gpt-image-2, so you configure that API yourself. Apache-2.0, Python 3.8+.
The official demo promotes a GitHub project: give it the repo link and screenshots, and it scans the material, searches comparable hits, scores selling points, turns the real screenshots into Xiaohongshu-style cards with image-to-image, and ends with 7 cards plus a caption and hashtags.
Features
Selling-point analysis: Scores points on scarcity, usefulness and perceptibility, then focuses on the strongest one.
Hit teardown: Searches similar high-performing notes and breaks down title patterns, card layouts and hooks.
Image-to-image: Keeps your real material and only restyles the text and layout, instead of inventing a fake image.
Five title styles: One option each for pain point, question, discovery, trending word and identity.
Two formats: 9:16 card sets, or a short-video script, saved to disk in a fixed layout.
Image generation uses gpt-image-2, so you configure that API yourself. Apache-2.0, Python 3.8+.
