📦 Object Focus Camera v3 (Ultimate)

If V2 is the "orbit it around" quick shot, V3 ("Ultimate") is when you want a real camera setup and not a lucky guess. It keeps the same DNA - one node that turns your intent into an edit prompt for Qwen-Image-Edit - but replaces the single action dropdown with a full stack: position, movement, distance, lens, and even the language you prompt in.

What's on the panel

The inputs you'll actually touch, in order:

• camera_position (19 choices) - the big one. Front, angled views at 15/30/45/60°, side, back, bird's-eye, worm's-eye, plus the orbit moves that V2 made famous. This sets the static framing.
• camera_movement (7 choices) - layered on top for motion: static, dolly in/out, tilt up/down, pan left/right.
• camera_distance (5 choices) - "Very Close (Macro)" through "Very Far". Picks how intimate the shot is.
• lens_type (10 choices) - 50mm normal, close-up, macro, 24mm wide, fisheye, 85mm telephoto, 135mm with bokeh, tilt-shift, panoramic. The backend expands these into technical descriptions so the model actually knows what "tilt-shift" looks like.
• prompt_language - English, Chinese, or Hybrid. The Chinese option exists because the dx8152 camera LoRAs for Qwen Edit 2509 were trained on Mandarin prompts, and they respond far better to it. If you're running dx8152's Multiple-Angles or Next Scene LoRA, Chinese (or hybrid) is genuinely the better default.
• add_detailed_explanation - None/Basic/Detailed, how much perspective reasoning to append after the base prompt.

target_object stays front and center ("the watch", "the door handle"), and show_details is still there for texture and engraving requests.

How it works

Same mechanism as V2 - plain-language camera instructions - but composed from six independent dimensions instead of one canned phrase. The output prompt reads like "change the view to a 45-degree angled view of the watch, at medium distance, shot on a close-up lens" plus whatever explanation depth you picked. system_prompt keeps the preserve-everything-except-the-camera framing, and description is a human-readable summary of what you asked for.

When to reach for it

V3 is the sweet spot of this family. V2 is too blunt the moment you want a specific angle; V4, V5 and V7 pile on transition modes, material presets and cinematography jargon that you don't need for everyday product and detail shots. If "45° angled view, close, close-up lens, English" covers your shot, this is the node.

Install + gotchas

It's part of the one-pack install - ComfyUI Manager, search "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

Then restart and find it under ArchAi3d/Qwen/Camera. You still need a Qwen-Image-Edit model (2509 or 2511) and its Qwen-VL CLIP on hand - none of these nodes generate anything without it. On a consumer GPU, use a GGUF or fp8 quantized build of the 20B model; the full bf16 weights are out of reach for most cards.

Same physics as V2: angles are directional, not measured. And remember the model can't do lens physics - "135mm bokeh" is a look, not a guarantee of compression, so judge results by feel. Left/right is picture-relative, and people in frame tend to get rotated instead of the camera.