📦 Object Focus Camera

This is the original Object Focus Camera, and it's the one to start with if the v6 "Ultimate" edition's forty-odd settings make your eyes glaze over. Five positions, four distances, three lenses, one optional details box. It generates a camera-focus prompt for Qwen-Image-Edit, and - like its siblings - it's designed around the dx8152 camera LoRAs, which is why there's a lora_mode dropdown even on the "simple" version.

The inputs are self-explanatory, which is the point:

• target_object - what the camera should focus on ("the watch", "the door handle").
• camera_position - Front, Angled 30°, Side 90°, Top-Down, or Low Angle. That's it.
• camera_distance - Very Close (Macro) through Far.
• lens_type - Normal, Close-Up, or Macro. The tooltip is honest: Close-Up and Macro are the ones optimized for the dx8152 LoRA.
• lora_mode - tells the builder which dx8152 LoRA you're running: Multiple Angles (camera movements) or Next Scene (scene transitions). Picking wrong means the generated prompt format won't match what the LoRA expects.
• show_details - optional free text for "showing fine texture and engravings" style specifics.

The outputs are the same triple as the rest of the camera family: prompt, system_prompt, and description. Wire prompt and system_prompt into the pack's Qwen encoder.

How it works

Under the hood it's pure prompt assembly - no model files, no API. It composes a Chinese-base camera instruction tuned for the dx8152 LoRA's training data (the author's testing found Chinese prompts track better), and the dropdowns just pick the phrasing. There's genuinely nothing magic here; the value is not having to memorize the phrasing that works.

Installation

It's in the ComfyUI-ArchAi3d-Qwen pack. ComfyUI Manager ("ArchAi3d Qwen") or:

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

Restart, and it's under ArchAi3d/Qwen/Camera. Personal use free; commercial use needs the author's paid license.

Where people get burned

If you use it without a dx8152 LoRA, the Chinese-base output can read as odd but usually still works - Qwen-Edit understands enough. The bigger trap is the version sprawl: the pack ships v1 through v7 of essentially this node, and people grab the latest and wonder why the settings changed. If you just want "front/angled/side of this object," v1 is fine and honest. And remember the universal Qwen camera caveat: it's directionally consistent, not angularly precise - "Side View" will reliably put you to the side, but the exact degrees are a suggestion.