TextToImage Siliconflow ♾️Mixlab

TextToImage Siliconflow is the pack's cloud-generation escape hatch: instead of running a checkpoint on your own GPU, it calls SiliconFlow's hosted API and gets an image back over the network. Per the pack's changelog, this node "can directly call SiliconFlow's provided Flux image generation" - and specifically, the one model it targets is black-forest-labs/FLUX.1-schnell, the fast, Apache-licensed, distillation-trained member of the Flux family built to produce a usable image in very few diffusion steps. That licensing detail matters here: schnell is the openly-licensed Flux variant (unlike Flux-dev, which restricts commercial use), which is presumably why it's the one a hosted API is comfortable serving.

What this buys you is speed of a different kind than local GPU speed - no checkpoint download, no VRAM budget to manage, no sampler tuning. You send a prompt, SiliconFlow's servers run schnell, and an image comes back. That's genuinely useful if you're on a low-VRAM machine, want to prototype a composition before committing local compute to it, or you're building a workflow meant to run somewhere without a GPU at all.

The required fields are api_key (from your SiliconFlow account), prompt, width and height (512 to 4096, in steps of 8, default 512 each), model - currently a one-item dropdown fixed to black-forest-labs/FLUX.1-schnell, with an optional custom_model_name field if SiliconFlow adds more image models to their catalog and you want to target one directly - and seed. There's no steps, cfg, or sampler control exposed here the way there would be on a local KSampler; schnell's whole design point is that it doesn't need them, and the API almost certainly runs it at the low step count the model was distilled for regardless of what you'd want to configure locally. The single output is image, a plain IMAGE you can preview, save, or feed into any downstream node exactly like a locally-sampled one.

Install is the pack-wide standard:

cd ComfyUI/custom_nodes
git clone https://github.com/shadowcz007/comfyui-mixlab-nodes.git
cd comfyui-mixlab-nodes
install.bat

or via ComfyUI Manager (search "comfyui-mixlab-nodes"), or pip3 install -r requirements.txt in a venv. Nothing to download locally for this node specifically - the model lives on SiliconFlow's servers, so the only setup is getting an api_key from their dashboard.

The most common snag is the same as any hosted-API node: a missing, wrong, or expired api_key fails the call before it ever generates anything, so check that first if a run errors out. Because this is a network call, expect it to be noticeably slower and less predictable than a warmed-up local sampler - a slow or flaky connection means a slow or flaky generation, and there's no local fallback if the API is down or rate-limiting you. Keep in mind too that width and height step in multiples of 8, same constraint as most diffusion models - an odd value will get rounded or rejected rather than honored exactly. And if you're expecting Flux-dev-level output quality or prompt adherence, recalibrate: schnell trades some of that for speed by design, so a soft or slightly-off result at default settings is usually the model doing what it was built to do, not a misconfigured node.