EigenAI FLUX Generator

Here's the one fact that explains this entire pack: the EigenAI FLUX Generator does not generate anything on your machine. It's a thin HTTP client with a ComfyUI face. You give it a prompt and some numbers, it POSTs them to a remote Eigen AI FLUX API server, waits up to five minutes, downloads the resulting image, and hands you a tensor. Your GPU can stay asleep; the README's talk of "quantized models for quick generation" describes the server, not your box.

This is the modular version of the FLUX generator - the one that expects its prompt from an EigenAITextNode and its lora_config from an EigenAILoraNode rather than carrying those fields itself. Same API, same behavior, split across three nodes for people who like their graphs tidy. The README frames the endpoint as FLUX.1-schnell, and the node itself couldn't care less: it just POSTs to /generate and trusts whatever the server runs.

The inputs that matter

• prompt (PROMPT) and lora_config (LORA_CONFIG) - wire these in from the Text and LoRA nodes. The prompt goes into the request verbatim; the LoRA config becomes the loras array in the payload.
• width / height - 256–1024, step 64, default 512. Note the server caps at 1024, which is below Flux's comfortable native territory; 512 squares are what the API is specced for here.
• seed - default -1 means the server picks one. Set a real value to reproduce an image; the rest of the payload also has to match, obviously.
• guidance_scale - default 3.5, which matches the Flux guidance sweet spot from the KB rather than SDXL-style CFG. Leave it unless you have a reason.
• api_url - the one you actually have to care about. Default is http://74.81.65.108:8000, a hardcoded public IP. That server may or may not be yours or reachable; every workflow built on this pack needs to point this at a live Eigen AI FLUX API endpoint.

Output: a single image (IMAGE tensor) that you can preview, save, or chain into an upscaler.

Where people get burned

• Silent failure. If the API call, download, or anything else throws, the node doesn't error out - it returns a light-red placeholder image at your requested size. A batch of pink squares that "succeeded" is the classic symptom. Check api_url first.
• It's a remote call with your text in it. You're sending prompts to a server you don't control. If that's a problem, this isn't your node.
• No key in the code, for what it's worth - the plugin never asks for an API key, so access control (if any) lives entirely on the server side.

Installing

No models, no heavy dependencies, nothing to download:

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

Restart, or install via ComfyUI Manager by searching "Eigen AI FLUX API". Dependencies are requests, pillow, numpy, torch - all already present in stock ComfyUI. If you'd rather have one node instead of three, EigenAIFluxNode is the unified version with the prompt, LoRA names, and upscale controls all on a single widget panel.