collbrogtr/comfyui-ideogram-autoprompter

This is a plugin for ComfyUI, which is a visual tool people use to build AI image-generation workflows by connecting nodes on a canvas. This particular node is for Ideogram 4, an image generator that requires prompts in a specific structured format, a JSON object that describes the background, the individual elements in the scene, where each element sits in the image, any text that should be rendered, and the color palette to use. Writing that JSON by hand is tedious, and this tool automates that step. You type a short description of what you want, and optionally drop in a reference image. When you click Generate, a vision AI model reads your input and builds the full structured caption for you, filling in all the required fields including bounding boxes that describe where each element should appear. The result is then shown on a visual canvas where you can drag, resize, and edit the regions before they get passed to Ideogram 4. The plugin supports a few different AI engines for building the caption. The default is a local model called Qwen3-VL that downloads automatically the first time you use it. To avoid keeping a large model in memory, it unloads after each generation by default. You can also use Gemini by pasting in a free API key, and the key is held only in memory for that session and never saved into your workflow file. Installation is done by copying the folder into ComfyUI's custom nodes directory and running a pip install for the dependencies. The generation step happens interactively inside the node's UI by pressing a button, not when the overall workflow graph runs. The node's output is the finished JSON string, which connects to Ideogram 4's prompt input in the rest of your workflow.