Direct answer: batch-producing Xiaohongshu posts efficiently
If you publish multiple Xiaohongshu (RED) posts every week, writing each caption and designing each image by hand makes a steady cadence hard to sustain. A more sustainable approach is an AI workflow: a large language model turns product details into titles, body copy, and image prompts, an image model renders the visuals in batch, and a human reviews everything before publishing.
RelayMe workflow templates chain these steps into one pipeline, which suits ecommerce sellers and content teams that need a constant output. Account positioning and topic strategy stay human decisions — the AI handles execution at volume.
The word pipeline matters here: each stage hands its output to the next automatically, so the marginal cost of the sixth post is minutes of review, not another hour of design. In an August 2026 test, one 1:1 product image on a budget-priced image model in RelayMe cost 20 credits — about $0.02 — and rendered in roughly half a minute.
Build your AI content workflow
Start with one fixed content format, for example a single-product recommendation post: the input is the product name, key selling points, and target audience; the output is one title, one body text, and three to four images. Encode that structure into the workflow prompt template so every run produces the same shape of content.
Keep separate templates for different formats — recommendations, listicles, tutorials — so they never interfere. Once the format is fixed, your product sheet becomes your publishing schedule: each row maps to one workflow run.
Two practical defaults keep the pipeline stable: pin the language model step to one model rather than letting it vary between runs, and set the image count explicitly in the template so a schedule slip never comes from a surprise fifth image. When a format evolves — say listicles move from five items to three — you change the template once and every later run inherits it.
Chain text and image generation automatically
In RelayMe, the output of the language model step can feed directly into the image generation step. In practice, the LLM writes the caption and, in the same run, produces a complete image prompt for each visual — subject, scene, palette, composition — and the image step renders them one by one.
Because copy and visuals come from the same run, style and content stay aligned naturally, with no manual copying of intermediate results between tools.
Cost stays predictable because each stage bills separately: at roughly 20 credits per image on a budget model, the visuals for a four-image post come to about 80 credits — under ten cents — and the text step adds only a small fraction on top. That per-post arithmetic is what makes daily posting financially boring, in the best sense.
Improve continuously with task history
After publishing for a while, compare your Xiaohongshu analytics with the RelayMe task history: which topics drive engagement, and which prompt patterns produce the most discarded images.
Fold winning patterns back into the template and rewrite prompts with high discard rates. A weekly review like this makes the template converge on your account style, and the share of manual editing keeps shrinking.
Two numbers are worth tracking from day one: discard rate per template (how many images you throw away before one passes review) and minutes of manual editing per post. Both fall as templates mature, and both are visible from the pipeline's own records — no extra spreadsheet needed. When a template's discard rate stays high for two weeks, the fix is almost always in the image prompt, not the model choice.
Publishing and rollout advice
Always review AI-generated content before publishing: confirm product details match reality, the copy meets platform rules, and the images carry no obvious generation artifacts.
Keep a steady cadence — for example one batch generation per week, published across several days — rather than flooding the feed at once. Pilot one or two formats first and expand templates only after they run smoothly; it preserves quality better than launching everything at once.
One more platform-specific habit: Xiaohongshu rewards consistency of persona, so keep the reviewing human constant per account instead of rotating editors through the pipeline. The reviewer is the persona's memory — the AI provides volume, and one consistent pair of eyes keeps ten posts a week sounding like one author rather than a committee.