# 1. hello liulian weekly
> The first cover is Liulian’s character sheet. The magnifying glass and teardown checklist capture what this newsletter is for: run it first, then explain the process clearly.
- Issue: 1
- Published: 2026/08/31
- HTML: https://weekly.zq4495.fun/en/posts/1
- JSON: https://weekly.zq4495.fun/api/en/posts/1.json
- Chinese original: https://weekly.zq4495.fun/posts/1.md
---
The first cover is Liulian's character sheet. The magnifying glass and teardown checklist capture what this newsletter is for: run it first, then explain the process clearly.
> **Weekly field notes from Liulian: AI content, tools, open source, and honest experiments. Confirmed results stay explicit; failures and limits stay in the record.**
## Field Test
**Wan3.0 finally makes IP-character video feel usable**
This week I tested Wan3.0 with the same Liulian IP character. Compared with my earlier Wan2.7 and HappyHorse attempts, character consistency and action completion improved enough to enter the usable range.
Cost still matters. The public price observed for this test was about RMB 1.2 per second, so the ten-second clip cost RMB 12. My current question is not “which model is best?” I first check whether the result clears the publishing bar, then count cost, failure rate, and rework together. The original thread also keeps the prompt and comparison material.
## Tool of the Week
**Bringing cloud-drive footage into an agent workflow**
For batch video work, editing is often not the slowest part. Finding the right clips inside hundreds of gigabytes of footage is. When folders are already organised, an agent can inspect the catalogue, choose material against a script, and download only what the edit needs.
The useful shift is not automating a few clicks. It turns a folder tree into production material that can be searched and reasoned over. This is still a workflow direction, not a claim that every cloud drive and editing project has been fully validated.
## Open Source
**Weekly: an Astro template made for long-running newsletters**
Liulian Weekly is adapted from this project. Markdown authoring comes with bilingual editions, RSS, full-text search, light and dark themes, comments, a static API, and Markdown endpoints that agents can read directly.
The template code declares the MIT licence, while the upstream author's writing and images remain copyrighted. I reused the code structure, removed the archive and personal configuration, and replaced the assets and content with my own. That separation is easy to miss when reusing an open-source project.
## Reading Note
**Build one small product before deciding whether to scale it**
I increasingly trust a simple order: publish around one concrete problem and observe what people save and ask. When the same question repeats, make a small template, consultation, or workshop. Only expand it into a course or long-term community after someone pays and gets a real result.
Automation matters for the same reason. Its job is not to look sophisticated. It lowers the time and cost of each experiment, leaving room to test more ideas and listen to more feedback.
## Notes
Issue 1 begins with the act of starting.
Future issues will record AI content experiments I actually ran, useful tools that may not be fashionable, open-source projects worth taking apart, and the less glamorous decisions involved in building a one-person business.
Not every experiment will work, and one result will never become a promise of a repeatable formula. I will keep the inputs, actions, costs, failures, and current conclusion on the table.
This is Liulian Weekly. See you next week.