GPT Image 2 Prompt Preview 🖼️

GPT Image 2 Prompt Preview is the no-edit sibling of the pack's main Selector node. Same 500+ prompt library, same thumbnail support, but no editable text box - it just shows you a prompt and its image, and hands the prompt text out as a string. If the Selector is the storefront, this is the window display.

When do you actually reach for it? Two cases. First, you're browsing the library to find inspiration and don't want to risk accidentally editing the prompt text - this node can't touch it. Second, you want a second prompt source in a workflow that already has the Selector doing the editing; the Preview can sit on the canvas showing what's selected elsewhere, or feed a different downstream branch.

How it works

Mechanically it's nearly identical to the Selector: the dropdown choices come from data/local_prompts.json, the selection strings encode the entry ID ([preset_0] [character] Character Case1), and the thumbnail is served from local disk through the pack's /gpt_image2_prompt/image route rather than an external URL. The node is an output node, so it pushes its UI payload (image path plus the prompt text) to the canvas when the workflow runs.

Inputs and outputs

Two inputs, both dropdowns:

• category - filters the library (10 options: all, portrait, poster, character, ui, comparison, ad_creative, ecommerce, other, custom).
• prompt_selection - the 529-entry list of prompts. This is the whole job.

Outputs:

• prompt (STRING) - the prompt text. This is the useful one; wire it into your GPT Image API node.
• image_path (STRING) - a relative path like images/portrait_case1/output.jpg. It's mostly there so the frontend can render the preview; you won't usually wire it anywhere, though you could feed it into a file-path-aware node if you want to reuse the reference image itself.

Install and troubleshooting

It ships in the same pack as the Selector, so the install is identical - ComfyUI Manager, search "GPT Image 2 Prompts", or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/comfyui-gpt-image2-prompt-T8

Restart, no models, no pip installs. The data builds automatically on first launch. The one quirk to remember: the Preview updates its canvas display when the workflow executes, not the instant you change the dropdown - so hit Run once and the preview catches up. If the image area stays stuck on "Select a prompt to see preview," the local data build probably failed, and the pack's Execution Checker node will tell you exactly which file is missing.