EigenAI Qwen Generator

The EigenAI Qwen Generator is the modular text-to-image node in this pack, and the honest way to describe it is: it's the FLUX Generator's source file with the docstrings changed. Compare them side by side and the payload construction is identical - prompt, width, height, guidance_scale, optional seed, an optional loras array - and it hits the exact same /generate endpoint. Whatever "Qwen-compatible" means to whoever runs the server, the client never sees the difference.

Like its sibling, nothing runs on your machine. It's an HTTP POST with a five-minute timeout, then a download, then a tensor. The README insists the Qwen default API is http://74.81.65.108:8010, but the shipped default in the code is http://74.81.65.108:8000 - same port as the FLUX node. Since that hardcoded IP is almost certainly not your server, the practical upshot is the same either way: you're setting api_url yourself, and whichever one is right depends on which server the Eigen AI API you're actually using exposes.

The inputs that matter

• prompt (PROMPT) and lora_config (LORA_CONFIG) - from the Text and LoRA nodes, same as the FLUX generator. Send nothing in and the node uses a default LoRA config on the server side.
• width / height - 256–1024 in steps of 64, default 512.
• guidance_scale - default 3.5, range −10 to 10. The default aligns with where Flux-family guidance usually sits; if your server runs a Qwen model with different expectations, this is a slider you'll actually touch.
• seed - -1 for random, or pin it for reproducibility.
• api_url - the field that makes or breaks the node.

Output: image (IMAGE). Wire it to a Save Image node and you'll find results in ComfyUI/output.

The trap you need to know about

Same as the FLUX generator: on any failure it returns a light-red placeholder image instead of raising an error. If your graph "completes" but hands you pink squares, the request never made it - check that api_url resolves and the server is up. There's no API key field anywhere in the code, so either the endpoint is open or authentication happens at a layer this plugin never touches.

Installing

cd ComfyUI/custom_nodes
git clone https://github.com/jialuw0830/flux_api_comfyui_plugin

Restart, or search "Eigen AI FLUX API" in ComfyUI Manager. Dependencies (requests, pillow, numpy, torch) come with ComfyUI. And if the modular three-node chain feels like overhead, EigenAIQwenNode is the unified version - same request, but with the prompt field, three LoRA name inputs, and an upscale toggle all on one panel.