📦 Object Focus Camera v4 (Enhanced)

Most camera nodes assume the camera is already pointed at the thing. V4 ("Enhanced") is the version that answers the question nobody asks until they hit it: what if I'm standing in the kitchen corner and I want to be facing the refrigerator? That "reposition" ability is the whole reason this version exists, and it's surprisingly rare in Qwen prompt builders.

What's new in V4

The headline is focus_transition_mode, a two-way toggle:

• Standard (Maintain Position) - behaves like V3: distance-aware positioning that keeps the camera roughly where it is, gentler at far distances.
• Focus Transition (Reposition to Object) - this is the new one. It tells the model to intentionally move the camera from the wide environmental view to stand directly in front of the target object. The tooltip's example says it best: "kitchen corner → face refrigerator."

Everything else carries over from V3 - camera_position (19), camera_movement (7), camera_distance (5), lens_type (10), prompt_language (English/Chinese/Hybrid, the Chinese option still being the right pick if you run dx8152's Qwen Edit LoRAs), and add_detailed_explanation (None/Basic/Detailed). target_object is still "the watch", "the ring", "the refrigerator", and show_details still takes texture requests.

Why the toggle matters

Think about the failure mode you've hit with every Qwen edit model: you ask for a close-up of the countertop, and it either hallucinates a brand-new countertop or produces a zoomed crop that reshapes the whole room. "Focus Transition" is the pack's attempt to phrase the move as a camera journey - reposition, then frame - rather than as an object edit, which is what keeps the scene intact. It pairs with the distance control: the tooltip notes that Standard mode is "distance-aware" and dials prompt strength back at far distances, because a 5m move and a 0.5m move are different requests and shouldn't be phrased the same way.

Mechanism

Same as the rest of the family: it composes a plain-English edit instruction for Qwen-Image-Edit, with the transition mode changing how the sentence is structured. Outputs stay the same three strings - prompt (ready for the encoder's prompt input), system_prompt (the preserve-the-scene camera-operator framing, which matters more in transition mode since the scene is doing more work), and description. Wire all three into an ArchAi3D encoder with a Qwen Edit 2509/2511 model and you're set.

Install

Part of the pack - ComfyUI Manager, search "ArchAi3d Qwen", or manually:

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. Don't forget the actual Qwen-Image-Edit weights - quantized (GGUF Q4 or fp8) for consumer cards.

When to use it

Honest take: if you shoot objects already in front of the camera, V3 is enough and V4's extra toggle just sits there. Reach for V4 when your workflow is the interior exploration one - camera somewhere, object across the room, "now go stand in front of it." That's the use case it was built for, and it does it better than any prompt you'll type by hand. If you're on the fence between V3, V4 and V5, the real question is whether you find yourself asking the model to walk the camera somewhere. If yes, V4 is your version.