How I Used Agent Plan to Automate the Final Step of Writing a WeChat Public Account Post

Problem

When publishing a WeChat public‑account article, the final step often stalls at finding a suitable, copyright‑free cover image. Searching image libraries can take a long time and still yield no appropriate picture.

Why previous AI packages couldn’t help

The author previously relied on Claude, but the account was unstable and Claude does not support image models, leaving a gap for automated cover‑image generation.

Agent Plan capabilities

Agent Plan extends the Coding Plan subscription with multimodal capabilities. It aggregates text, code, image, and video models—including GLM‑5.1, Kimi‑K2.6, and DeepSeek—under a single subscription. It also provides internet‑search quotas and the Doubao‑embedding‑vision model for accurate retrieval in long or fragmented conversations.

Desired workflow

Feed the finished article to the agent, let it decide the appropriate cover style, and output the image without manual description, sizing, or cropping.

Configure Hermes Agent

Select provider ：Custom endpoint (enter URL manually).

API base URL ：https://ark.cn-beijing.volces.com/api/plan/v3

API KEY ：use the dedicated Agent Plan API key.

Model name ：text‑generation model doubao‑seed‑2.0‑lite.

Display name ：volcengine‑agent‑plan hermes model After starting Hermes, verify that the model is set to doubao‑seed‑2.0‑lite.

Install image‑generation skill

Download the ZIP package from the following URL and unzip it into the directory ~/.hermes/skills/:

https://findskill.com/volcengine/agentplan/byted-ark-seedream-skill

Once installed, the skill can be invoked from Hermes Agent to generate images.

Adapt skill for WeChat cover

The generic skill is extended with two parameters: cover – generates a WeChat‑style cover image. title – renders the article title on the cover.

Usage

“Help me make a WeChat cover, the article is about …, title is ‘…’” → returns four candidate images with titles. “Generate a background image for the article, I’ll add the title myself” → returns a pure background.

Example prompt (Chinese):

帮我生成一张公众号封面图，文章讲得是火山方舟的Agent Plan，标题是【Agent Plan】

The first execution prompts for the API key; after entering the key the agent generates the image and saves it to the desktop. The same prompt can request four candidates at once, and a reference image can be supplied to guide the style.

Broader implications

Agent Plan’s multimodal subscription enables text, code, image, and video generation in a single package, allowing end‑to‑end workflows such as turning a document into a web page or creating short‑video assets without comparing individual model capabilities.