GR Prompt Generator1

The author built this one for a specific, relatable annoyance: writing bland, too-short prompts and not having the energy to flesh them out. In his own words, describing this node when he shared it: "if like me, you make bland prompts describing things in as few words as possible and want to expand it, you can use this node... it can take your own short text prompt and expand on it as well." That's the whole pitch - either roll something from scratch, or hand it three words and let it build a fuller prompt around them.

Don't confuse this with GR Prompt Generator (the vision-language one that describes an image) - this one doesn't look at pixels at all. It's a template/category-driven text generator, closer in spirit to a wildcard system than a captioner.

How it works

Two ways to use it. Leave positive and short_text empty and it generates something from scratch, steered by category (subjects, moods, colors, compositions, details, weather, time_of_day, objects, styles) and type_of_image (a long list of visual styles - 3D Render, Anime-style, Cinematic, Cyberpunk, and dozens more). Or give it a short_text - your bare-bones idea - and turn expand on, and it builds that idea out into something fuller rather than generating unrelated content. seed controls which random elements get pulled, same as any seeded randomizer - reuse a seed to get the same generated prompt back.

positive is there as a base you're building on top of, distinct from the throwaway short_text - think of positive as an existing prompt you want the generator to extend rather than replace.

The inputs and outputs that matter

• short_text (STRING, multiline) - your minimal idea. This is the input that makes the "expand a bland prompt" use case work.
• expand (BOOLEAN, default off) - turn this on when you want the node to build on short_text/positive rather than generate independently.
• category - which axis the generator leans on when building content (subjects, moods, colors, etc.).
• type_of_image - the visual style applied to the result, from a large preset list.
• seed - reproducibility; same seed, same generated prompt.
• generated_prompts (STRING, output) - the result, ready to wire into a CLIPTextEncode.
• seed (INT, output) - the seed that was actually used, handy for reproducing a result you liked later.

How to install it

Via ComfyUI Manager: search "GraftingRayman", install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/GraftingRayman/ComfyUI_GraftingRayman

then restart. Every node in this pack, this one included, needs OpenAI's CLIP package installed separately from ComfyUI's own CLIP:

# portable build
.\python_embeded\python.exe -m pip install git+https://github.com/openai/CLIP.git

# system python
pip install git+https://github.com/openai/CLIP.git

Skip it and the whole pack fails to import.

Common issues & troubleshooting

Node missing from search. That's the CLIP dependency above, a pack-wide install requirement rather than anything specific to prompt generation.

Output ignores your short_text. Make sure expand is actually on - with it off, the node generates independently from category/type_of_image rather than building on what you typed.

Same prompt every time. Expected if seed isn't changing - wire it to a randomizing widget if you want fresh output each queue, and keep it fixed only when you deliberately want to reproduce a specific generated prompt.

Generated text doesn't fit your model's prompting style. This node produces prose/tag-style descriptive text aimed at general use - if you're on an SDXL-lineage checkpoint that wants comma-separated Danbooru-style tags specifically, or an LLM-encoded model that wants a clean instruction rather than a stacked description, treat the output as a starting draft to edit rather than something to paste in unmodified. What a prompt should look like varies a lot by which text encoder your checkpoint actually uses.