🎨 Hrio Design｜单图生成

Hrio_Design_Single_Image_Node (🎨 Hrio Design｜单图生成) is the smallest, most honest node in the Hrio Design pack: you give it a prompt and some reference images, it gives you exactly one IMAGE back. It's a cloud call dressed as a ComfyUI node, part of a Chinese-origin pack aimed at graphic designers and interior designers cranking out proposal images (平面设计 and 室内设计 pitches, if you want the pack's own framing).

The name is a small lie that works in your favor: nothing runs on your machine. No checkpoint, no VAE, no LoRA loading, no GPU at all. The node uploads your reference images to a hosted backend and calls a Gemini-format /v1beta/models/{model}:generateContent endpoint on a private server the author runs. Your only real job is supplying an API key.

How it works

Under the hood the flow is: resolve the key (node field, or the config.ini fallback), upload any reference images to the pack's object storage via presigned URLs, POST the prompt plus the uploaded image URLs to the backend, then download the returned image and convert it to a ComfyUI tensor. All of it happens in one blocking call - expect it to take a while at 4K, it's not instant.

That's also the two things to keep in mind before you wire it into a workflow. It's a per-call metered service, not a free sampler, and your prompts and reference images physically leave your machine and sit on a server you don't control. Fine for client work, something to think about twice for anything sensitive.

The inputs that matter

• prompt - your image prompt. The default is a solid Chinese design-proposal brief (clean composition, natural materials, tidy lighting) that you'll almost always replace. It's multiline, so use it.
• model - banano2 (default), banano-pro, veo3.1, or gemini3.1-pro. banano2 is the pack's own hosted image model; veo3.1 and gemini3.1-pro are Google's closed models reached through the author's backend, in the reseller-proxy pattern.
• api_key - leave blank and it tries config.ini's [banana] api_key. The README never tells you where to get a key, which is the pack's real gatekeeper: the author hands them out.
• image_size - 1K / 2K / 4K / 8K. Default 2K is a good balance; 8K is a long wait.
• aspect_ratio - Auto plus presets from 1:1 up to 4:1 and 1:8.

The optional image_1 through image_10 slots are reference images - up to ten of them, each uploaded and sent alongside the prompt. And there's automation_payload, a JSON blob the pack's web UI uses for batch runs; ignore it unless you're scripting.

Output

One output, image, type IMAGE - the node's tooltip says it plainly: this node outputs exactly one image. Wire it into a preview or a save node like any other generator.

Installing it

Same story as the rest of the pack. In ComfyUI Manager, search Hrio-Design and install, or:

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

Then restart ComfyUI. Dependencies are just requests, numpy, and pillow - all but certain to already be in your environment. There are no model downloads, because there are no local models.

Where people get burned

The API key is the whole ballgame: without one the node renders an error image telling you to fill in the field or edit config.ini. Check verify_ssl = false in that same config.ini - the pack talks to its backend with TLS verification disabled, which is normal for this pack but a good reminder that you're trusting a third-party server with your traffic. Reddit has effectively no chatter about this pack, so there's no community troubleshooting to lean on - if it breaks, it's you and the author.