ArchAi3D_Simple_Position_Prompt

Qwen-Edit and Qwen-Image don't do coordinates. Give them "put a bike at x=230, y=400" and they'll nod and ignore you. The trick the community converged on instead is the position guide: you draw numbered red rectangles on a copy of your image, show it to the model as a second input, and write your prompt as "rectangle 1 = ..." - the rectangles are the coordinates. This node writes that prompt for you, consistently, with the formatting Qwen actually expects.

It's from ComfyUI-ArchAi3d-Qwen, Amir Ferdos's pack. He's an architect turned ComfyUI developer, and this is his take on a workflow his Qwen encoder nodes are built around: prompt + numbered position guide image → precise object placement or removal without pixel coordinates.

How it works

You describe what goes in each rectangle, separated by slashes, and the node wraps each one in parentheses - the format Qwen was trained on. Slash order maps to rectangle number in order, so add a bicycle / add a man sitting on chair becomes:

(rectangle 1 = add a bicycle.)
(rectangle 2 = add a man sitting on chair.)
(Maintain all original elements in Image 1 unchanged.)

It works in two modes:

• add_objects - place new objects at the rectangle positions, keeping everything else.
• remove_objects - erase whatever's inside each rectangle and fill it with seamless background (extend the floor, wall, sky, whatever's there).

Each mode ships an optimized system prompt, because "remove the person and fill naturally" needs a very different instruction than "add a bicycle." The full system prompt for remove mode reads like a small spec sheet - it explicitly tells the model the red rectangles are invisible guides and not to draw them, which is exactly the failure mode people hit with this workflow.

Inputs

• operation_mode - add_objects or remove_objects. Start here; it changes the defaults everywhere.
• object_descriptions - your objects, /-separated, one per rectangle in order. This is the input that matters.
• start_description / end_description - optional bookends. If you leave end_description empty it picks a smart default per mode (maintain-everything for add, seamless-fill for remove).
• use_system_prompt - yes/no; no returns an empty system prompt if you'd rather supply your own.

Outputs: formatted_prompt (STRING) and system_prompt (STRING). Both feed your Qwen encoder's text inputs.

Installing

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/amir84ferdos/ComfyUI-ArchAi3d-Qwen.git
cd ComfyUI-ArchAi3d-Qwen
pip install -r requirements.txt

Or ComfyUI Manager → "ArchAi3d Qwen". No model downloads for this node - it's pure text formatting. Pack is free for personal use; commercial work needs a license from the author.

Honest caveats

The node formats the prompt, but the placement quality is still up to the model - keep your subject centered and describe it concretely, and expect approximate positions rather than pixel-exact ones. Also, the rectangles come from a separate image you have to make (any editor with a rectangle tool works, or another node that draws boxes). This is a utility that saves you from hand-formatting a bracket-heavy prompt fifty times, not a magic bullet for perfect composition - but for interior-design "add this lamp over there, remove that chair" work, it's exactly the shape the community found works.