Flux助手(高级)
Run Flux without downloading a single gigabyte
- image
Here's the honest pitch: local Flux is heavy. The dev model alone is over 30 GB of weights, and you need a serious card to run it happily. This node does the whole thing on Alibaba's hosted Flux endpoints instead - type a prompt, get a real Flux image out as an IMAGE tensor, and the only thing you downloaded was a few lines of Python. It's the fastest way to "run" Flux that exists, and for anyone on a modest GPU it's the only way.
Pick model from flux-schnell, flux-dev, or flux-merged. The advanced variant gives you the dials: seed (with a huge max), steps (leave at 0 and it auto-configures - the tooltip spells it out: schnell defaults to 4 steps, dev to 50, merged to 30), guidance (default 3.5, which tunes how tightly the output obeys the prompt - higher is more literal, lower is more creative), size from a 6-entry preset list (1024×1024, 512×1024, 768×512, 768×1024, 1024×576, 576×1024), and offload (a True/False toggle that offloads heavy components to CPU to ease memory pressure - the tooltip's recommendation, leave it False unless you're resource-limited). prompt is your prompt, and api_key is the pack's standard DashScope key.
The mechanism is a single dashscope.ImageSynthesis.call, and the returned image URL is downloaded and converted back into a ComfyUI tensor that lands on your GPU. Output: one IMAGE.
For the casual use case - you just want a Flux image and don't care about the dials - the "Flux助手(简易)" sibling in this pack is the same thing with steps, seed, guidance, and offload stripped out. This node is for when you want to lock a seed, tune guidance, or experiment with step counts on the hosted model.
Install: ComfyUI Manager → "ComfyUI-My-Nodes", or cd ComfyUI/custom_nodes && git clone https://github.com/Tagbliton/ComfyUI-My-Nodes and restart. The pack's requirements.txt installs dashscope, openai, requests, soundfile, numpy, and pins urllib3==1.26.20 - don't touch that pin.
Gotchas - and this is the one to remember: the README explicitly calls out that choosing flux-schnell here throws list index out of range, and it's currently unfixed. The workaround is to use flux-dev or flux-merged instead. Also keep the API framing in mind: this is metered pay-per-image via DashScope, so it's not free like local Flux is after the download, and getting a key at all requires a real-name-verified mainland China account - the single biggest reason people outside China can't make this node work.
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| api_key | STRING | — | |
| model | COMBO | 3 options: flux-schnell, flux-dev, flux-merged | |
| seed | INT | 00–4294967290 | — |
| steps | INT | 00–100 | 如果为0,自动配置。 flux-schnell 模型官方默认 steps 为4,flux-dev 模型官方默认 steps 为50,flux-merged 默认 steps 为30。 |
| guidance | FLOAT | 3.50–100 | 指导度量值,用于在图像生成过程中调整模型的创造性与文本指导的紧密度。较高的值会使得生成的图像更忠于文本提示,但可能减少多样性;较低的值则允许更多创造性,增加图像变化。默认值为3.5。 |
| size | COMBO | 6 options: 1024*1024, 512*1024, 768*512, 768*1024, 1024*576, 576*1024 | |
| offload | COMBO | False | 一个布尔值,表示是否在采样过程中将部分计算密集型组件临时从GPU卸载到CPU,以减轻内存压力或提升效率。如果您的系统资源有限或希望加速采样过程,可以启用此选项,默认为False。 |
| prompt | STRING | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |