FluxProOnline(NYJY)
Flux Pro on someone else's GPU, no API key of your own
- image
Flux Pro is the top-tier version of the model family that reset the open-weights quality bar - the one that never shipped open weights, only as a paid API. This node doesn't change that. What it does is give you a free, keyless path to it from inside ComfyUI by proxying through a third-party Hugging Face Space. You type a prompt, pick a resolution and seed, and get back an IMAGE tensor - no local Flux weights, no VRAM spike, no BFL API account.
Fair warning up front: this rides on someone else's generosity, so it's a sometimes node, not a production pipeline. When it works it's magic; when the free Space is overwhelmed (or the upstream changes), it fails. Set expectations accordingly and it's a fun tool to have.
How it works
Under the hood the node is a thin gradio_client wrapper around the HF Space NihalGazi/FLUX-Pro-Unlimited, which exposes a /generate_image endpoint that proxies requests to upstream Flux Pro instances. The server_choice dropdown selects which upstream server the Space routes you to - mostly named after hosting setups (google_us, azure_lite, artemis, nb_dr, pixelnet), with a few labeled nsfw_core for uncensored routing. Your prompt, width, height, and seed are sent, the Space returns an image, and the node loads it back as a standard IMAGE tensor for the rest of your graph. There's no API key anywhere in this flow - the Space is doing the authenticated calling, you're just a passenger.
The inputs that matter
- prompt - multiline; the thing you're asking for. Flux Pro is a natural-language model, so write sentences, not tag soup.
- width / height - both default 1280, range 512–2048. Set your canvas.
- seed - default 1234; fixed seeds give reproducible results.
- server_choice - which upstream server to use. If one is slow or erroring, trying another is the standard fix. The
nsfw_coreoptions exist for uncensored routing - whether you actually get NSFW output depends entirely on what the upstream allows, and it's a third-party service, so don't lean on it for anything you care about.
Output: a single image (IMAGE).
Installing
Standard NYJY pack install: ComfyUI Manager → Install via Git URL → https://github.com/aidenli/ComfyUI_NYJY, restart. The gradio_client dependency (bundled in requirements.txt) is the one this node needs. No model files, no keys.
Common issues & honest caveats
Failures are expected, not bugs. Free proxy Spaces get rate-limited and saturated; a timeout or empty result usually means "the Space is busy," and switching server_choice or re-running is the community-standard workaround. This node is the most fragile thing in the pack, and that's structural - you're a guest on someone's free tier.
Don't build a business on it. It's free, which means no SLA, no consistency guarantee, and upstreams that can change or vanish. If you need dependable Flux Pro output, get a real BFL API key and use a proper OpenAI-style integration - this node is for experiments and "what would Flux Pro even do with this?" moments. And when the servers behave, the answer to that question is genuinely nice - Flux's natural-language handling and prompt adherence are the reasons it became the quality benchmark in the first place.
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | — | |
| width | INT | 1280512–2048 | — |
| height | INT | 1280512–2048 | — |
| seed | INT | 12340–2147483647 | — |
| server_choice | COMBO | google_us | 9 options: google_us, azure_lite, artemis, nb_dr, pixelnet, nsfw_core, +3 |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |