GR Prompt Generator Extended

This is GR Prompt Generator1 with every axis broken out into its own control instead of blended into one category/type_of_image choice. If the plain version's "pick a category, pick a style" felt too coarse - you wanted the mood dialed independently from the color palette, and the composition independently from both - this is the version that lets you do it.

How it works

The base idea is the same as its sibling: give it a short_text idea and turn expand on to build a fuller prompt around it, or leave it empty to generate from scratch, seeded and reproducible via seed. What's different is the sheer number of independent axes you can steer: mood_type, color_type, composition_type, detail_type, weather_type, time_of_day_type, object_type, style_type, and subject_type are each their own enum, most offering a none (skip this axis entirely) and a random (let the generator pick) alongside dozens of specific named options - subject_type alone runs to nearly 190 choices.

subject_only is worth calling out specifically: turn it on and the generator narrows to just the subject rather than building out the full scene around it, useful when you want a tight, focused prompt rather than an elaborate multi-clause description.

Practically, this node rewards knowing what you actually want before you open it. Scrolling ten separate enum dropdowns to build one prompt is a lot of clicking if you just wanted "something moody and cinematic" - for that, the plain GR Prompt Generator1 with its single category/type_of_image pair gets you there faster. Reach for this one when you have specific, independent requirements across several of those axes at once and the coarser version keeps giving you combinations you don't want.

The inputs and outputs that matter

• short_text / expand - same as the base version: your seed idea, and whether to build on it.
• subject_type, style_type, type_of_image - the three axes with the largest option lists; these do the most work in shaping the overall feel of the result.
• mood_type, color_type, composition_type, detail_type, weather_type, time_of_day_type, object_type - finer per-axis controls, each defaulting to none so you only pay attention to the ones you actually care about.
• subject_only (BOOLEAN) - narrow the output to just the subject, skipping the broader scene-building.
• seed - reproducibility, as with any seeded generator.
• generated_prompts (STRING, output) - the assembled result.
• seed (INT, output) - the seed actually used.

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, including this one, needs OpenAI's CLIP package installed alongside ComfyUI's own:

# 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

Common issues & troubleshooting

Node missing from search. Check the CLIP dependency above - it applies to the whole pack, not just prompt generation.

Result feels like it's fighting itself. With nine independent axes, it's easy to pick combinations that pull in different directions (a weather_type and a time_of_day_type that don't naturally go together, say). Leave axes you don't have a real opinion on set to none rather than filling every field just because it's there.

Prefer the simpler version. If you find yourself leaving most of these axes on none most of the time, GR Prompt Generator1's single category/type_of_image pair is doing the same underlying job with far less clicking - this node earns its complexity specifically when you need several axes controlled independently at once.

Output doesn't match your model's expected prompt style. As with the plain version, treat the generated text as a draft. Whether it should read as comma-separated tags or a natural-language sentence depends on which text encoder your checkpoint actually uses, not on anything this node knows about.