CGA_NegativeSelector
One dropdown for the negative-prompt boilerplate you keep retyping
- STRING
You know that string of negative prompt keywords you paste into every SDXL workflow - "cgi, 3D, digital, airbrushed," the whole nine yards? CGA_NegativeSelector is a dropdown full of exactly those presets, so you stop retyping them. Pick one, optionally append your own text, and it spits out the combined prompt as a single STRING you wire into your CLIP Text Encode (negative).
That's the entire value proposition. It's a convenience node, and it's honest about being one.
How it works
- preset - a dropdown with six options. Five are hardcoded negative prompt presets, each tuned to push a photorealistic result away from the CGI/anime/illustration look ("fake eyes, deformed eyes, bad eyes, cgi, 3D…", "(worst quality, low quality, illustration…)", and so on). The sixth is None.
- custom_prompt (STRING, optional, multiline) - your own text, which gets prepended to the preset.
If you pick None, you get just your custom prompt. Otherwise you get custom_prompt followed by the preset. Output is a single STRING.
The dropdown shows truncated versions of each preset (the full text is long), but the full string is what comes out the other end.
The two things you have to know
First, and this one will actually bite you: the fifth preset contains embedding:BadDream. That's a reference to an A1111-style textual-inversion embedding that must exist in your ComfyUI embeddings folder, or the CLIP encode downstream will fail with a missing-embedding error. It's not included with the pack. Either install BadDream from Civitai, or avoid that preset.
Second, a reality check on negative prompts in general: they only do anything when CFG is meaningfully above 1. On guidance-distilled models - Flux, the Turbo and Lightning checkpoints - at CFG 1 the negative prompt is not even computed, so this node is a no-op there. It earns its keep on SD 1.5, SDXL, and the SDXL-derived line (Illustrious, Pony, and friends) at normal CFG values, which is exactly the audience these presets were written for. And the presets are firmly 2023-era kitchen-sink boilerplate; the community has largely moved toward naming one specific defect or using a coherent source region instead of forty keywords. Fine to start from, but don't treat the dropdown as gospel.
Install
ComfyUI Manager → search ComfyUI_CGAnimittaTools → install → restart, or:
cd ComfyUI/custom_nodes
git clone https://github.com/cganimitta/ComfyUI_CGAnimittaTools
The pack itself has no models and no heavy deps - just torch, numpy, pathlib. The only external file this node's presets reference is that BadDream embedding, which is on you to grab. You'll find the node under CGAnimittaTools in the menu.
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| preset | COMBO | 6 options: fake eyes, deformed eyes, bad eyes,..., cgi, 3D, digital, airbrushed, (worst quality, low quality, illust..., color, cartoonish, digital, (modern..., (deformed iris, deformed pupils, se..., None | |
| custom_promptopt | STRING | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| STRING | STRING | — |