🌊Trix Prompt AIO

🌊Trix Prompt AIO is the "all-in-one" prompt node that ships inside ComfyUI-TrixNodes - a custom panel with a style gallery, a translator, a prompt stack, and a negative-prompt box, all feeding two plain text outputs: positive and negative. If you've built workflows where your prompt lives in a pile of separate text nodes and you keep copy-pasting style snippets around, this is the thing that collects all of it in one place.

It's a heavy-feeling node for a simple job, and that's the point: the heavy lifting is in the UI, not in the outputs. You wire positive and negative into your CLIP Text Encode(s) exactly like you would any text node.

How it works

Under the hood it's a few moving parts working together:

• Prompt stack + separator logic. You can feed an optional text_in string, and the node's own stack UI adds more lines. It then joins them with your choice of comma, period, space, and newline separators, skipping empties and not double-punctuating. It's a fancier version of the text-concat nodes you already have.
• Style tags. Put @style_name or *category in your prompt and it expands them from the bundled Excel style banks - fooocus_styles.xlsx, the no8d_* files, krea2_styles, and others it ships in styles/. It's wildcards-by-any-other-name, but the gallery UI makes them browsable instead of memorized.
• Translation. The headline feature: type in your native language, get English out. There's online_google and online_bing via the deep_translator library, or offline via NLLB-200 (600M or 1.3B) or M2M-100 running through CTranslate2, int8-quantized, on CPU. It can translate live in the UI or on the final output.

The Russian-language README and the ru→en default tell you where this node grew up: it's built for people who prompt in their own language and need English for the model.

The inputs and outputs that matter

The only real input is text_in - an optional, multiline STRING. Everything else you set lives in the node's custom UI panel (style gallery, translation engine and languages, prompt stack), not as sockets.

Outputs are exactly two:

• positive - your merged, style-expanded, optionally translated prompt. Wire into your positive CLIP Text Encode.
• negative - same treatment for the negative box.

One word of caution from the wider prompting landscape: that negative output only does work on SDXL-lineage models (Illustrious, NoobAI, Pony). On LLM-encoded checkpoints like Flux 2 Klein, Z-Image, or Anima, the negative prompt is largely inert - the encoder was never trained on it. Know which encoder you're talking to before you invest in a big negative library.

Gotchas

The first real one is install weight: this pack's requirements.txt pulls in ctranslate2, huggingface_hub, deep_translator, openpyxl and xlsxwriter whether you ever translate anything or not. It's all needed for the translator and the Excel style storage, but it's a chunk of dependencies for what looks like a text node.

Second: the online Google route is the flaky one. deep_translator's Google endpoint gets rate-limited and breaks unpredictably - if translations start failing, switch to the offline engine. Just know the first offline run downloads a model from Hugging Face (roughly 600 MB for NLLB-200 600M, ~1.3 GB for the 1.3B) and runs on CPU, so the first batch is slow. After that it's cached and the pack unloads idle models to keep RAM reasonable.

Third: this is a front-end-heavy node with its own JavaScript. If the custom panel renders wrong or the style gallery is empty after an update, hard-refresh the browser (Ctrl+Shift+R) before you blame the node - stale JS caches cause most of the weirdness.

Install

ComfyUI Manager → search ComfyUI-TrixNodes (publisher trix7111) → install → restart. Or the manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/trx7111/ComfyUI-TrixNodes

Ignore the README's stale pixaroma/ clone URL - that's an old account name; the repo is trx7111. Restart and you'll find the node under the Trix/Prompting category, ready to hoard your whole prompting life in one box.