AI-HIVE 电商图片

AI-HIVE 电商图片 (class AIHiveEcommerceImage) is the pack's flagship workflow node, and it's basically AIHiveGenerateImage with a platform dropdown bolted on and a job to do. Where the plain generate node lets you write whatever prompt you want, this one takes a target marketplace and prepends a chunk of platform-specific guidance to your prompt before it goes to the model - then runs the exact same generation under the hood.

The platform enum is the whole point:

• taobao - 淘宝/天猫 main image and detail pages: clear subject, concentrated selling points, Chinese title space reserved
• jd - JD-style commercial photography, structure and material focus
• douyin - Douyin/doudian product cards, mobile-first visual focus
• xiaohongshu - Xiaohongshu (RED) cover and lifestyle framing, title safe-area
• amazon - Amazon Listing/PDP: accurate subject, no invented brands, certifications, or claims
• instagram - IG social commerce: bold style, clean composition, room for a call to action
• generic - the catch-all when none of the above fits

The guidance is literally prepended text - check the source and you'll see the strings baked into COMMERCE_GUIDANCE. It's not a different model or a magic pipeline. That matters for two reasons: it means the framing is free (you get it without writing it), and it means what you see on the canvas is your prompt plus invisible Chinese boilerplate. If you're not targeting Chinese marketplaces, generic or instagram is usually the honest pick.

The workflow the README pushes

This is the pack's own demo, and it's a good one:

1. Wire Load Image into reference_image.
2. Set platform: taobao, model: public_model_nano_banana_pro.
3. Prompt something like: 保留商品结构与标签，生成高级浅色背景主图，主体居中，右上角预留标题空间
4. Take the images output straight into Preview Image or Save Image.

The reference_image socket is optional but for real product work you want it - the whole "keep the bottle and its label" trick depends on feeding the actual product in and telling the model what to preserve. That's the GPT Image / Nano Banana strength this node exists to expose: replace the background, keep the merchandise intact.

Other inputs carry over unchanged from AIHiveGenerateImage: model, routing_mode, batch_size (1–4, each billed), model_params_json, wait_for_result, timeout_seconds. Outputs are images, task_id, task_json.

Installing

ComfyUI Manager: search AI-HIVE, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/wubin1836/comfyui-ai-hive.git
pip install -r comfyui-ai-hive/requirements.txt

Just requests and Pillow, no model downloads. Needs AI_HIVE_API_KEY (or ~/.ai-hive/config.json) before launch.

Gotchas

Same as the whole pack: metered billing (check batch and model before running), data leaves the machine, and a timed-out poll means the job still finishes server-side - pull it with AI-HIVE 查询任务 rather than re-running. The one node-specific thing to remember: that baked-in "don't fabricate efficacy, prices, or certifications" guidance is doing real work, so it's on you to keep product claims honest in your own prompt too. This is a paid API wrapper around a filter-bearing closed model - the moderation is the server's, the accuracy is yours.