🎨 Agnes Text-to-Image

Every local workflow has a moment where you wish you could just call a hosted model and get a result back in your graph, no GPU fan spinning up. That's this node. Agnes_Text2Image sends your prompt to Agnes AI's cloud, downloads the finished image into a ComfyUI IMAGE tensor, and hands it to you like it was generated locally - plus a URL, which is the part that matters if you're chaining toward video.

The pack's pitch is "no local GPU required," and it's accurate: nothing here runs diffusion on your machine. You're renting inference, which is the right trade when you have a laptop, a deadline, and no appetite for a 12GB model download. ComfyUI's node graph becomes a nice front-end for a hosted service instead of a heavy local pipeline.

How it works

The node POSTs to {base_url}/images/generations - your config's base URL plus /images/generations - with the prompt, the model name agnes-image-2.1-flash, the chosen size, and your seed. It parses the response for an image URL, downloads the file, and converts it to a [1, H, W, C] float tensor that plugs straight into any ComfyUI image socket. The url output carries the same image's public URL, which is what lets you pipe the result into Agnes_Image2Video in a single wire.

The inputs that matter

The required set is small:

• config - from Agnes_Config. No key, no run.
• prompt - describe the image. The default is a landscape-at-sunset boilerplate, so it's genuinely optional to change.
• size - a dropdown of nine presets, defaulting to 1152x768. Handy landscape, portrait, and square options without worrying about model-native resolution.
• seed - -1 means random; set a fixed value to reproduce a result. Only seeds >= 0 are actually sent to the API, so don't expect -1 to be meaningful as a number.

The one optional is negative_prompt, which is empty by default - fill it only if you know Agnes's image model reacts well to it; with a hosted model you can't easily A/B it the way you would a local sampler.

Install

Grab it via ComfyUI Manager (search "AgnesAI") or clone:

cd ~/ComfyUI/custom_nodes
git clone https://github.com/Watchcats211/ComfyUI_AgnesAI_Nodes.git

Restart ComfyUI. Dependencies are torch, numpy, and Pillow - all standard. This node doesn't even need OpenCV; that's the video nodes' job.

Common issues

• A red square where your image should be. On any API error, the node returns a 512×512 red placeholder tensor. That's not a rendering bug - it's the pack's built-in "something broke" signal. Check the ComfyUI console for the actual error line.
• Auth errors - the Config node's connection test should catch a bad key before you ever reach generation. If it 401s here anyway, your key is invalid or out of credits.
• It's a cloud service, so quality is out of your hands. You get agnes-image-2.1-flash and the nine sizes, full stop. If you're used to tweaking samplers and CFG on a local model, this will feel like a blunt instrument - but that's the deal you signed up for, and it's fast.

The single best thing about this node is the url output. Text-to-image is nice; text-to-image that becomes the opening frame of a generated video, with one wire, is the actual reason this pack exists.