Flux Context Preset

Flux Context Preset is a prompt-engineering cheat sheet that outputs exactly one thing: a STRING of instructions for an image-editing model. Pick "Teleport" and it produces a carefully worded brief telling an editor to re-contextualize your subject somewhere unexpected; pick "Product" and it writes a brief for turning your image into a professional catalog shot; pick "Haircut" and it asks for four different hair restyles. You get 13 transformation presets, each one a pre-written "you are a creative prompt engineer" system prompt ready to paste into any editing model that takes natural-language instructions - Flux Kontext, Qwen-Image-Edit, Gemini Flash Image, gpt-image edits, you name it. No API keys, no model downloads, no magic. It's a text template node.

Why you'd reach for it

Because writing good edit instructions is fiddly, and the presets are genuinely well-crafted. The "Move Camera" preset doesn't just say "change the angle" - it demands "the camera now gives a top view of the room," with the scene-aware specificity that actually gets good results out of instruction-editing models. If you're new to this, the gap between "rotate the camera" and a prompt like that is the difference between a mediocre edit and one that looks intentional.

It's the one node in this pack that always loads - the pack's __init__.py deliberately imports flux_context_preset first with a try/except, so even if every other node in the pack fails to import (missing easyocr, broken deps), this one still shows up. That's a decent hint about where the author put the most care.

How to use it

Two inputs: preset (the dropdown of 13) and additional_text (optional extra instructions appended after the preset). Output: a STRING. From there you've got two natural paths:

1. Straight into an image-editing model - wire the STRING into any node that takes a natural-language edit prompt (Kontext, Gemini, gpt-image). Combine with the pack's polymath_chat node by passing an image in and using the preset as the prompt, and you've got an automated "transform my render" pipeline.
2. Into polymath_chat for iteration - let an LLM read the preset and refine it for your specific image.

The presets are tuned for instruction-editing models, which is worth being clear about: this is not a diffusion prompt. "Relight" and "Product" and "Remove Text" are the same jobs people used to do with masked inpainting, and as of the last couple of years the community consensus is that instruction editors took those jobs over - see the inpainting discourse for the full argument. The one thing a mask still owns is pixel-identical unmasked regions; if you need that, the pack's polymath_text_mask node is the companion to reach for instead.

Installation

It's in the comfyui_LLM_Polymath pack, so ComfyUI Manager → search "comfyui_LLM_Polymath" → install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/lum3on/comfyui_LLM_Polymath
pip install -r requirements.txt

then restart. Because this node is pure string output it works with zero configuration - no keys, no Ollama, no model files. You only need the API/model side set up downstream.

A couple of notes

The presets are opinionated and mostly excellent, but they're hard-coded - "Haircut" always asks for exactly 4 variations, "Movie Poster" always assumes a random genre. That's fine (random is kind of the point), but if you want a specific genre you'll be editing the additional_text field to steer it. And remember this is an indie pack released in early 2025 - the preset list is what it is, and there's no way to add your own through the UI. If you outgrow it, the output is just text; you can copy the preset, tweak it, and stop using the node entirely. That's the beauty of a node whose whole job is a string.