🍌 Banana｜普通三视图并发

This is the "no training wheels" version of the Hrio-Commerce image nodes. Where the flagship template node hides all its prompt magic behind a mode dropdown, the "🍌 Banana｜普通三视图并发" node hands you three plain text boxes - front_prompt, side_prompt, back_prompt - and fires three concurrent requests to the backend. That's the whole difference, and for a lot of people it's the better tool.

Why you'd pick this one: you know what you want each view to say. Maybe the template's fashion-replace phrasing doesn't fit your product, or you're shooting something that isn't clothing at all. You write the prompts, you get three consistent images with the same reference inputs, same model, same size. It's an API node in the cleanest form: no sampler, no VRAM, no local model - just your key, your text, and a cloud backend doing the generation (see the KB's external-api-nodes writeup for the honest tradeoffs: per-call cost and your data leaving the machine).

How it works

Same engine as its template sibling. The node takes whatever you typed into the three prompt boxes, uploads your optional reference images (up to ten slots, reused across all three requests), and sends three concurrent generation calls with the same model, image_size, and aspect_ratio. Results come back as IMAGE tensors on the canvas - ready for a Save Image or Preview node, as if a local sampler had made them.

The one genuinely nice mechanic is generate_scope. Default is "全部并发生成" (generate all three), but you can set it to regenerate just the front, side, or back. The other two views come back from this node's last successful run - cached, not re-billed. Bad side view? Re-run the side only. That's a real money-saver on a metered API.

Inputs and outputs that matter

• api_key - the vendor key; leave blank and it reads config.ini.
• front_prompt / side_prompt / back_prompt - your text. Each one is a separate multiline box, and the defaults are solid Chinese e-commerce phrasing (white background, single subject, clean lighting) if you just want to edit them slightly.
• model - Banano (default), Banano Pro, or Gemini 3.5 Flash. Display names; the backend maps them to its actual endpoints.
• image_size (default 2K) and aspect_ratio (default Auto) - shared across all three views.

Outputs: front_image, side_image, back_image (each a single IMAGE), a combined images batch, plus info (STRING) and mp4url (STRING, empty for image nodes). The only inputs a beginner actually touches are the three prompts and the model.

Install

Same pack as everything Hrio:

cd ComfyUI/custom_nodes
git clone https://github.com/moki-flank/Hrio-Commerce

or search Hrio-Commerce in ComfyUI Manager and restart. No model files to download - the pack's real dependencies are just requests, numpy, and pillow; torch comes from ComfyUI. You still need the vendor's API key, in the node field or in the pack's auto-created config.ini under [banana].

Common issues

• Missing key, missing output. No API key → the node returns a red error image as its output instead of failing loudly. Check the info string.
• Chinese defaults. The default prompts are in Chinese, matching the template logic the backend expects. If you type pure English, the results may drift from the product-photography style the defaults enforce. Keep the structure - white background, single subject, no collage, no watermark.
• It's proprietary and it phones home. This pack is closed-source, carries a credential, and calls a cloud endpoint by design. That's the category the KB flags most strongly: read the code before you hand it your key, and know every reference image goes to the vendor's server.
• After updates, delete old Banana nodes and hard-refresh the browser (Ctrl+F5) - stale nodes have caused issues in past releases.

Want more than three angles? The pack's five-view node does front, two 45° views, side, and back in one shot. Want the guided templates instead? That's the template panel node.