Agnes-AI Text

The least glamorous node in this pack is arguably the most useful. Text-to-image quality lives or dies on the prompt, and Agnes-AI Text runs four LLM presets on Agnes's servers and hands you back a clean string you feed straight into your existing text encoder. No GPU, no local LLM download, no VRAM for a 7B model you only wanted for prompt polish.

How it works

It's a chat-completions call to https://apihub.agnes-ai.com/v1/chat/completions with a hardcoded system prompt per preset, temperature fixed at 0.3 and max_tokens at 2048. After the reply comes back, the node strips common prefixes like Prompt: and **Prompt:** so the output is usable without cleanup. The two vision presets are the interesting part - they attach your image to the request as a base64 PNG data URI, so the model is actually looking at the picture, not guessing.

The four presets:

• Prompt Enhance - expands a brief prompt with lighting, palette, composition, camera angle. No image needed.
• Translate to English - translates while preserving visual terms. No image needed.
• Extract Art Style from Image - pulls the art movement, technique, and palette out of a reference. Needs an image.
• Image Detailed Description - writes an extremely detailed, generation-ready prompt from an image. Needs an image.

Inputs and outputs that matter

• preset - pick one of the four above.
• prompt - the text to enhance or translate; required for the two text presets.
• system_prompt (optional) - overrides the baked-in system prompt. This is the secret weapon: paste your own style guide here and the preset becomes yours.
• image (optional) - required for the two vision presets.

The single output, output (STRING), wires straight into a CLIP Text Encode node, or into AgnesImage's prompt input if you want a cloud-to-cloud pipeline. Anywhere a prompt string goes, this can feed it.

A quirk you'll hit

This node returns problems as strings rather than raising errors. Forget the image on a vision preset and the output is literally Image required for this mode. Miss the API key and you get API key required - set it in ComfyUI Settings Panel → Agnes-AI sitting in your text field. Both are fine - they're debuggable - but glance at the actual output before you chain it into a text encoder, or you'll generate from the word "Error."

Installing and configuring

Same install as the rest of the pack: ComfyUI Manager, search Agnes-AI, restart. No extra pip dependencies. The default text model is agnes-2.5-flash; agnes-2.5-pro-alpha is the paid tier, and agnes-2.0-flash / agnes-1.5-flash are the older options - all selectable in Settings under the Agnes-AI entry. You need a free key from platform.agnes-ai.com either way.

The caveat, briefly

Same one as every node in this pack: your prompts (and for the vision presets, your images) leave the machine and go to a server you don't control, on a platform that's young and largely unproven in the community. The API-wrapper category has produced malware before, so skim the pack source before trusting it with anything sensitive. For the common case - "make this vague sentence into something SDXL won't choke on" - it does the job and does it for free.