📦 Object Focus Camera v6 (Ultimate)

Qwen-Image-Edit is great at "change the camera angle," but it's also great at ignoring you if you phrase the request like a spec sheet. Object Focus Camera v6 is this pack's most complete attempt at turning camera intent into prompt text the model actually follows - the "Ultimate" edition of a node line that went through seven versions. It builds a ready-to-paste camera prompt around a target object, and if you're using the dx8152 camera LoRAs the community built for Qwen Edit 2509, this is the node the author tuned them for.

The core is target_object - "the chandelier crystal", "the brass handle" - plus camera_position, a 29-way dropdown mixing orbit patterns (Front, Angled 15-60°, Side, Back, Top-Down, Low Angle, plus Orbit Left/Right in degrees) with vantage-point phrasing. camera_movement adds dolly/tilt/pan on top; the tooltip notes dolly is the most consistent with Qwen, which matches the community finding that distance-based movement beats arbitrary degree rotations.

The settings you'll actually tune

• distance_mode + camera_distance or distance_meters - preset words (Very Close → Very Far) or a numeric meter value, which the node converts to words because a literal "1.5m" in the prompt reliably renders wrong.
• prompt_language - Chinese (best with dx8152 LoRAs), English, or Hybrid. If you're not using those LoRAs, English is fine and readable.
• lens_type - 10 options from macro to fisheye to tilt-shift. Close-Up and Macro are the dx8152-trained sweet spot.
• auto_facing - adds "facing [target]" to vantage prompts so the camera actually points at the object.
• material_detail_preset (37 entries) and photography_quality_preset (15) - pre-written clauses like "Crystal Facets" or "Commercial Product" you bolt on instead of typing yourself. This is where the node earns its keep: two dropdowns instead of three sentences of trial-and-error.
• focus_transition_mode - "Standard" keeps the position, "Focus Transition" explicitly repositions from one object to another (e.g., corner → refrigerator).
• show_details - free text for anything the presets miss.

Outputs

prompt (the camera prompt for your encoder), system_prompt (a system-level instruction framing the shot), and description (a human-readable explanation of the camera setup - handy for debugging why the model did what it did). Feed the first two into the pack's Qwen encoder nodes.

Installation

Part of ComfyUI-ArchAi3d-Qwen - 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, find it under ArchAi3d/Qwen/Camera. Free personal, paid commercial license.

Where people get burned

This is a prompt builder, not a guarantee - Qwen re-emits the whole frame, so pixels around your object will drift, and the model can rotate a person in frame instead of the camera. Centered subjects work better than off-center ones, and "left/right" in a prompt means picture left/right, not the subject's. The version numbers (v1 through v7) are genuinely different behavior, not marketing: v6 combines Chinese base prompts with vantage-point English details in one prompt, which is a specific style - if you want the simple English-only version, the plain Object Focus Camera node or the Cinematography Prompt Builder are calmer starting points.