OneThingAI Hunyuan Image

The name "Hunyuan" sets an expectation in this community, and it's a trap. Tencent's open-weights Hunyuan video line is what the local crowd actually runs - and its image releases are famously huge, up to an 80B image MoE that wants roughly 320GB of VRAM. OnethingAIImageHunyuan ("OneThingAI Hunyuan Image") is the opposite of all that: hunyuan-image served over an API through OneThing AI. No weights, no VRAM, no license paperwork - just a prompt and a billed generation. If that tradeoff sounds like the whole point of the pack, it is.

How it works. It's the same OpenAI-compatible client as the other per-model nodes in ComfyUI_Onething_Image: POST your prompt and key to OneThing AI's /v1/images/generations, get base64 JSON back, and it lands in your graph as a standard IMAGE tensor you can save, upscale, or feed into anything else. This is the simplest node in the pack - one model, size knobs, done.

Inputs that matter. model is locked to a single choice (hunyuan-image), so the only things you set are:

• api_key - your OneThing AI key
• prompt - the whole job
• image_size - presets (1536x1024 / 1024x1024 / 1024x1536) or 自定义, with custom_width/custom_height (512–4096, step 64)
• retries / timeout - 3 and 120 by default

There's an extra optional input, too: a JSON string that gets merged straight into the request payload, which is your escape hatch if you need to pass an API parameter the node doesn't surface. The output is a single IMAGE.

One thing worth flagging before you wire anything: the schema lists an optional reference_image, but the pack hard-disables it for this model. Wire an image in and it gets filtered out before the request is sent - this is text-to-image only. If you want image-edit in this pack, the Gemini, OpenAI, or Seedream nodes are the ones that actually do it.

Install. Standard, and short, because there's nothing heavy to install:

cd ComfyUI/custom_nodes
git clone https://github.com/OneThingAI/ComfyUI_Onething_Image.git
pip install Pillow requests

Restart ComfyUI - or find "ComfyUI OneThing AI Node" in ComfyUI Manager. No model files, no dependencies beyond what ComfyUI already ships.

Troubleshooting. There isn't much that can go wrong that isn't pack-wide. A 401/403 means the key or credit is wrong. A hung queue is the timeout - bump it toward 180 if hunyuan-image is running slow. And remember the always-executes quirk: the node re-runs on every queue pass and each pass is a billed call, so re-queuing an unchanged workflow still costs you.

The honest framing: if you want the Hunyuan image model and you have no way to run it locally, this is the easy on-ramp. If you just want a Tencent-family model and have a GPU, the open video line is where the community actually lives - this node is strictly for the API path.