ComfyUI LM Studio FLUX.2 Prompt Generator

Simple ComfyUI custom nodes for generating FLUX.2-style prompts with a local LM Studio model.

This version is intentionally simplified. It removes advanced controls that can easily cause bad output, such as visible fallback mode, thinking mode, image format, prompt focus, and base prompt influence.

Nodes

LM Studio FLUX.2 Prompt Generator
LM Studio FLUX.2 Prompt Generator from Image

The first node converts a short text idea into a longer FLUX.2 prompt. The second node uses an input image plus an optional base prompt to generate a FLUX.2 prompt.

Recommended model

Recommended model:

HauhauCS/Qwen3.5-35B-A3B-Uncensored-HauhauCS-Aggressive

Hugging Face:

https://huggingface.co/HauhauCS/Qwen3.5-35B-A3B-Uncensored-HauhauCS-Aggressive

NVIDIA RTX 5090 class GPU recommended for the recommended model. If your GPU is smaller, choose a different vision model or a smaller/more heavily quantized GGUF vision model.

Installation

Clone this repository directly into your ComfyUI custom_nodes directory:

cd ComfyUI/custom_nodes
git clone https://github.com/Slartibart23/comfyui-lmstudio-flux2-prompt-generator.git

The important file path must be:

ComfyUI/custom_nodes/comfyui-lmstudio-flux2-prompt-generator/__init__.py

Restart ComfyUI completely.

LM Studio setup

1. Open LM Studio.
2. Load your text or vision model.
3. Open the Developer area.
4. Start the local server.

Default endpoint used by the nodes:

http://127.0.0.1:1234/v1/chat/completions

Get the exact model ID with PowerShell:

Invoke-RestMethod http://127.0.0.1:1234/v1/models | ConvertTo-Json -Depth 10

Paste the exact id into the node's model field.

Simplified node settings

Recommended starting values:

prompt_mode: photoreal or cinematic
target_word_count: 170
seed_mode: fixed
temperature: 0.3
max_tokens: 4096

The nodes automatically use a no-think style instruction and automatically clean common reasoning output such as Thinking Process:. You do not need to choose fallback settings manually.

When the output is not a clean FLUX prompt

If the output is incomplete, too analytical, or not cleanly written as a FLUX-style prompt, increase both:

ComfyUI node max_tokens
LM Studio context length

Suggested values:

ComfyUI max_tokens: 4096 to 8192
LM Studio context length: 16384 or higher, if your hardware allows it

Prompt modes

neutral
cinematic
photoreal
illustration
product
character
NSFW
Strong NSFW

The NSFW modes are intended only for legal, consenting-adult workflows. Do not use this project for minors, non-consensual imagery, exploitation, or illegal content.

Troubleshooting

Node does not appear

Check that the root __init__.py is here:

ComfyUI/custom_nodes/comfyui-lmstudio-flux2-prompt-generator/__init__.py

Then restart ComfyUI.

LM Studio is not reachable

Make sure LM Studio's server is running and the node endpoint is:

http://127.0.0.1:1234/v1/chat/completions

Text still starts with Thinking Process

Increase max_tokens and LM Studio context length, or use a non-reasoning model for prompt generation. This simplified node tries to clean reasoning output automatically, but a dedicated non-reasoning instruct model may be cleaner for text-only prompt generation.