AI-HIVE 电商带货视频

AI-HIVE 电商带货视频 (class AIHiveEcommerceVideo) is AIHiveGenerateVideo wearing a marketing hat. Same three model families (seedance_2_5, minimax_h3, happyhorse), same five generation modes (t2v, i2v, r2v, edit, extend), same string-URL outputs. The only real difference is a platform field plus a block of ecommerce guidance prepended to your prompt before the job goes out.

One thing to notice: here platform is a free-text STRING (default 抖音电商 / 小红书 / Amazon / TikTok Shop), not an enum like the image-side ecommerce node. That's a clue to how it works - the platform text is just injected into the prompt. The node prepends something along the lines of: generate an ecommerce product, 带货, or lifestyle video; show the subject, a usage action, key details, and real selling points; frame for mobile; and don't fabricate efficacy, prices, certifications, or brand facts. Say your target platform in Chinese or English, and it tunes the framing.

Is it worth using over the base node?

That depends on how you work. This node saves you from writing the same platform boilerplate into every prompt, and it enforces the structure (subject → usage → details → CTA) so a lazy "make me a product video" prompt doesn't come back as a twenty-second closeup of a label. If you're churning out douyin or xiaohongshu product clips in volume, that enforcement is worth it. If you write long, deliberate prompts anyway, you lose nothing by using the base AI-HIVE 视频生成与编辑 node and skipping the injected text - it's the same engine, and your prompt is your prompt either way.

A realistic setup

platform: 小红书
model_family: seedance_2_5
generation_mode: i2v
prompt: 15 秒竖屏种草视频：产品特写开场，展示使用动作，关键细节特写，结尾留出行动号召空间

Feed a product image into reference_image, and the node uploads it and appends it to the image list automatically. Outputs are result_urls_json, task_id, task_json - remember, URLs, not a video tensor, so plan a download step.

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

Only requests and Pillow. Set AI_HIVE_API_KEY (or ~/.ai-hive/config.json) before launch.

Gotchas

The family rules from the base node apply unchanged - minimax_h3 has no edit or extend mode, for instance, and hitting an unsupported combo means filling custom_public_model_id (must start with public_model_). Billing is real and video is the expensive direction, so watch batch/length and use AI-HIVE 查询任务 after any timeout instead of resubmitting. And keep the honesty framing in mind: the guidance tells the model not to invent claims, but it's still your product and your prompt.