Prompt Vault — Browse

Ever sat in front of an empty positive prompt box and realized you don't actually know how to describe a studio headshot with the right lighting, the right film grain, and skin that doesn't scream "AI"? That's the exact problem Prompt Vault - Browse exists for. It's a shelf of 21 hand-written photographic templates in 8 categories (portrait, fashion-editorial, beauty-closeup, product, architecture, landscape, street-documentary, food-drink), each written the way a shoot happens: camera, lens, lighting setup, film stock, what the skin texture should look like.

And despite the name, there's no vault API involved. No key, no network call, no model download. It's a dropdown, a text box, and a string output. The name is a lie only in the sense that all local tools are - the "vault" is a Python file full of templates sitting in your custom_nodes folder.

How it works

Pick a template and its full prompt text loads into the edit_prompt box - the placeholder says "Selected template appears here. Edit freely before output," and it means it. The one gotcha to internalize: if edit_prompt has any text in it, that's exactly what the node outputs. The template is a seed, not a lock. Change the template dropdown and the old edited text stays put, so if you want to swap in a fresh template you have to clear the box first. It's a footgun, but it's also the reason this node is actually useful - the templates are strong starting points, and editing them is the whole point rather than a fight.

The pieces that matter:

• template - the dropdown with all 21, labeled like [portrait] Professional Studio Headshot.
• edit_prompt - where the template lands and where you make it yours. This is the field you'll live in.
• category_filter - heads up: it doesn't actually filter the dropdown. The list always shows all 21. What it does is shape the "no template selected" listing you get back if you run the node with nothing chosen.
• append_text - tacks extra text onto the end of whatever the prompt ends up being. Handy for adding a subject when a template's too generic.
• save_to_favorites - boolean; true writes the final prompt to the pack's favorites JSON so the Favorites node can pull it back later.

Outputs are two strings: prompt - the finished prompt, wire it into a CLIP Text Encode's positive input - and prompt_dna, a JSON blob recording which template, whether you edited, and the character count. That's a diagnostic, not something to feed a sampler. Ignore it unless you're debugging.

Install

Easiest path is ComfyUI Manager: search "ComfyUI-Prompt-Vault" and install. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/jeremieLouvaert/ComfyUI-Prompt-Vault.git

Restart ComfyUI and the three nodes appear under AKURATE/Prompt Vault. The requirements file is a single comment line - pure Python, zero dependencies, no models to fetch.

Common issues

• Feeding the "no selection" output to CLIP. If you run the node without picking a template, the prompt output is a plain-text listing of available templates. Wire that into CLIP and you'll render text, not a portrait. Always pick a template.
• The edit box staleness thing above - clear edit_prompt to reload a template.
• Wrong model, wrong flavor. These are long natural-language prompts written for photo-real generation. They sing on SDXL-lineage and newer natural-language-encoded models, and they're wasted on tag-based anime checkpoints like Illustrious or Pony - the KB's own advice is to check what encoder your checkpoint uses before writing prose. Also, by design, this pack has zero anime, 3D, or cartoon content. It's a photography shelf; don't expect it to prompt your waifu.