ArchAi3D_Qwen_Interior_Focus

This node exists because of a toaster. The default preset is literally called kitchen_appliance_closeup, and the author's tooltip calls it "⭐ YOUR TOASTER SOLUTION!" - the pack's maker apparently got asked, repeatedly, how to make Qwen-Edit zoom in on a kitchen appliance and leave everything else alone. If that's your problem too, this is the node from the ArchAi3d Qwen pack that fixes it.

What you're really doing is writing a camera prompt. ArchAi3D_Qwen_Interior_Focus is a prompt-builder node for Qwen-Image-Edit (the "Qwen Edit 2509" the tooltips keep naming). It takes a handful of choices and hands you a ready-made instruction like "change the view to a face-level vantage point 1.5m to the front facing the toaster" - plus a research-flavored system prompt that tells the model it's an interior photographer who must preserve all furniture, lighting and spatial relationships. You don't type camera-speak; you pick presets.

How it works

The mechanism is the "vantage point" pattern from the pack's QWEN_PROMPT_GUIDE, which the author flags as the most reliable camera-change formula they tested. Feed it into the pack's own Qwen encoder (ArchAi3D_Qwen_Encoder_V2 or similar) with a loaded Qwen-Edit model, and the whole thing rewrites the frame. Because Qwen-Edit re-emits the whole image rather than patching a region, the node's whole job is to phrase the change so the rest of the room stays put - that's what the system prompt output is for.

The inputs that matter

Most of the dropdowns you can leave alone. The ones you'll actually touch:

• preset - 18 focus targets (kitchen appliances, furniture, fixtures, decor). The toaster one is the default for a reason; it works.
• scene_context - a sentence describing the room ("modern kitchen with white countertops"). The tooltip's right: it noticeably improves consistency.
• distance - 0.5 to 10 meters, default 1.5. Smaller = intimate closeup, larger = more room context. The author's research says distance-in-meters beats vague degrees for reliability.
• height - face_level is the standard; slightly_above reads as a countertop view, elevated as bird's-eye.
• auto_facing - keep it on. It bakes "facing the target" into the prompt so the subject stays centered.

Three outputs come out: prompt (wire to your encoder's user text), preset_description (handy for showing what you picked), and system_prompt (wire to the encoder's system_prompt input - this is the part that keeps the rest of the room from drifting into a pile of melted furniture).

Installing it

This is one node of the ComfyUI-ArchAi3d-Qwen pack by Amir Ferdos (ArchAi3d), an architect who builds interior-design tooling. Install the whole pack:

# ComfyUI Manager: search "ArchAi3d Qwen" and hit Install
cd ComfyUI/custom_nodes
git clone https://github.com/amir84ferdos/ComfyUI-ArchAi3d-Qwen.git
cd ComfyUI-ArchAi3d-Qwen
pip install -r requirements.txt   # mostly PyYAML; heavy deps are optional

Then restart ComfyUI. One honest gotcha: the node downloads no model, and the pack bundles none. You need a Qwen-Image-Edit checkpoint (2509 or 2511) loaded in ComfyUI - at 20B that realistically means a GGUF quantized build on consumer GPUs, and the encoder node needs the matching Qwen-VL clip and VAE.

Common issues

• Angle drifts. Qwen-Edit is directionally consistent but not geometrically exact; the tooltip's own research admits "angles may not be exact." Re-run with a fixed seed or nudge the distance.
• Whole-frame creep. Unedited pixels move because the model regenerates everything. That's why the system prompt output exists - skip it and your "focus" edit becomes a redesign.
• People in frame. The model may rotate the person instead of the camera. For people, use the pack's Person category nodes instead.

Worth reaching for when you want "hey, show me the toaster up close" without hand-authoring a camera prompt every time. That's a surprisingly common ask in real-estate and kitchen-design work.