SynVow-Nano2 T2I

SynVow-Nano2 T2I has a small identity problem. The name makes it sound like a tiny distilled model you load from a file - it isn't. There's no model file, no checkpoint, no VRAM involved at all. This is a wrapper around an OpenAI-compatible image-generation API, and the only "nano" about it is the pack. Text goes out, an image comes back, and the actual diffusion happens on someone else's servers. That's the whole pitch: if you've got a working API key and an internet connection, you can generate images from inside ComfyUI without a GPU breathing down your neck.

It's part of the wider trend of API-image nodes creeping into the ComfyUI ecosystem. The default model is gemini-3-pro-image-preview, so you're really renting Google's Gemini image model through a reseller's endpoint rather than running anything local. Worth knowing before you wire it in: your prompt, and eventually your images, are leaving your machine. Same as any cloud generation, just know who you're sending them to.

How it works

The node builds a plain chat-completions request - a model, a user message, and a seed - and POSTs it to the provider's /v1/chat/completions URL. The model's reply comes back as JSON, and the node regex-scrapes it for images in two forms: base64 data URIs embedded in the message content, or URLs (it filters for google/image/oss/ufile/png/jpg hosts). Either way, they get decoded, resized, and returned as a normal IMAGE tensor you can pipe straight into a Save Image node.

"SynVow" is the service brand, but here's the first trap: the README advertises three API sources - SynVow AI, T8Star, and Apicore - while the node's dropdown only actually offers T8Star and Apicore. The SynVow branch exists in the code but no longer has a UI choice wired to it. Pick T8Star or Apicore, get a key from that provider's site, and don't hunt for a SynVow option that isn't there.

The inputs that matter

• api_key - your key from the chosen provider. It's a plaintext string widget, which means it gets baked into the workflow JSON when you save. Don't share workflows with your key still in them.
• prompt - the actual text-to-image prompt. Defaults to 4k wallpaper, masterpiece, so it's not shy about taste.
• aspect_ratio - 1:1, 16:9, 9:16, 4:3, or Custom Size. Pick Custom Size and the custom_w/custom_h pair (64–16384, step 8) kick in.
• count - 1 to 4 images per request. Output is a batch stack, so a Save Image node writes them all.
• seed - 0 means random. The code also forces any oversized value into Int32 range with a modulo, so don't stress about the max.

model_select is free text, not a dropdown, so you can swap in whatever the endpoint supports - the default gemini-3-pro-image-preview is just a sensible guess. Output is a single images IMAGE that feeds Preview/Save as usual.

Installing it

Install via ComfyUI Manager (search SynVow-Comfyui-Nanoapi), or clone manually:

cd ComfyUI/custom_nodes
git clone https://github.com/AJbeckliy/SynVow-Comfyui-Nanoapi

Then restart ComfyUI. One honest warning: the README's own clone command is a copy-paste trap - it literally contains YOUR_USERNAME and the wrong repo name (Synvow-nano2). The URL above is the real one. Dependencies are just torch, numpy, Pillow, requests, and urllib3, all of which a stock ComfyUI already has, so there's nothing extra to download and no model files at all.

Gotchas

The request timeout is 1000 seconds, so slow generations don't die instantly - but they can hang for a while if a provider is struggling. If you get an "API Error" exception in the console, that's the provider's JSON error field passed straight through; check the key and the model name first. "No images found" means the response came back with no parseable image at all, and the console log prints the first 200 characters so you can see what actually arrived. SSL verification is disabled (verify=False), which the README justifies for internal APIs - fine for a trusted endpoint, just another reason not to point this at an endpoint you don't control.

If your GPU is idle and your wallet tolerates per-image API pricing, this is a genuinely easy way to get Gemini-grade results in the node graph without a local install. Just set your key once, keep it out of shared workflows, and remember the model you're renting is only as honest as the provider you bought it from.