OneThingAI Flux Image

Flux is the model name this community actually runs locally - flux-dev and flux-schnell live on plenty of rigs, and flux-dev's licensing has been a discussion topic since BFL shipped it. OnethingAIFlux ("OneThingAI Flux Image") is the version of that where you never download anything: flux-dev served through OneThing AI's API, one key, billed per image. No 24GB fp16 download, no 8GB+ VRAM, no GGUF hunting. The trade, naturally, is money and control.

How it works. Same OpenAI-compatible client as the rest of the ComfyUI_Onething_Image pack: 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 or pipe anywhere. Under the hood there's retry logic for 429/5xx responses and a 120s default timeout. Two things to note: the node always re-runs on every queue (each pass is a billed generation), and while the schema lists an optional reference_image, the pack hard-disables it for this model - wired or not, it's dropped before the request goes out. Text-to-image only.

Inputs that matter. model is locked to a single choice (flux-dev) - no Schnell, no Kontext, no fine-tunes, which is a real limitation. What you set:

• api_key - your OneThing AI key
• prompt - flux-dev responds well to natural-language scene descriptions
• 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

Plus the pack-wide optional extra JSON input, which gets merged into the request payload if you need to pass something the node doesn't expose. Output is a single IMAGE.

Install. The shortest in the pack:

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

Restart ComfyUI, or install "ComfyUI OneThing AI Node" from ComfyUI Manager. No model files - the model runs on OneThing AI's servers, so there's also nothing to license-check on your end beyond OneThing AI's own terms.

Troubleshooting. The 401/403-for-bad-key and hung-queue-timeout cases apply like everywhere else. The bigger question is whether this node earns its place. Honestly, it's the one I'd least reach for: flux-dev via API is strictly worse than running it locally if you have a GPU - you lose LoRAs, ControlNet, any of the ecosystem tooling that makes Flux worth using, and you pay per image for the privilege. Where it makes sense: you have no GPU at all, you need one quick Flux-quality image, or you're building a multi-provider comparison workflow and want Flux in the same graph as Gemini and Seedream. For that last job, it's handy to have. For anything serious, run the weights.