ArchAi3D_Qwen_Interior_Navigation

Real-estate listings live on the walkthrough shot: camera drifts forward through a living room, past the kitchen, around the corner. ArchAi3D_Qwen_Interior_Navigation is the ArchAi3d Qwen pack's node for generating exactly that motion - and the default preset is enter_room, which the author notes is the popular one for exactly your "virtual tour" use case.

Like every camera node in this pack, it's a prompt builder for Qwen-Image-Edit. It doesn't move anything itself. You describe a movement, it hands you a prompt string that tells the model "move smoothly 3m forward through this modern living room with open kitchen," plus a system prompt that instructs the model to keep the architecture and furniture intact while the camera travels. The hard part of "navigation" with an edit model is that the room has to stay the same room as the camera moves - which is exactly what the generated system prompt is doing the heavy lifting on.

How it works

The internals are a movement matrix: a movement direction (forward/backward for depth, left/right for strafing, up/down for vertical travel), a distance in meters, a speed (slow = cinematic, normal, fast = dynamic), and an optional rotation for the look direction mid-move. The follow_wall rotation option is aimed at tracking shots where you want the wall to stay in frame while you dolly alongside it. Set a preset like enter_room and it fills most of these in for you.

The inputs that matter

• scene_context - describe the room. This is the single biggest lever on output quality; "modern living room with open kitchen" beats leaving it blank by a mile.
• distance - 0.5–10m, default 3. The tooltip's guidance is genuinely useful: a typical room wants 3–5m per move. Too small and the model has nothing to work with; too large and it starts inventing rooms.
• speed - slow is for cinematic walkthroughs, fast for dynamic fly-throughs. Worth matching to the pacing you'll use if you chain these into a video.
• rotation - none for a straight move, follow_wall when you're doing a tracking shot.

Outputs are the pack's standard trio: prompt (into your encoder's user text), preset_description, and system_prompt (into the encoder's system_prompt input - don't skip it on this node, it's what stops the furniture from re-arranging itself as the camera moves).

Installing it

It ships inside ComfyUI-ArchAi3d-Qwen (author: Amir Ferdos / ArchAi3d, an architect who makes AI interior-design tools). Install once, get all the nodes:

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

No models download - you supply the Qwen-Image-Edit checkpoint yourself (GGUF quantized is the realistic route on consumer VRAM; the 20B bf16 weights are a data-center toy).

Common issues

• The room mutates mid-move. Qwen-Edit re-emits the whole frame, so successive navigation steps drift - that sink migrates between frames. Keep scene_context specific and don't chain too many moves before regenerating.
• Direction semantics. Left/right in your prompt is picture-left/right, not the camera's or the subject's. The pack's research docs hammer this repeatedly because it's the #1 source of "it went the wrong way."
• This is still one frame. The node gives you one viewpoint move per pass. For actual walkthrough videos you chain frames - the roadmap mentions FLF video generation as future work, so today you assemble the sequence yourself.

If you're making listing-style interior tours on Qwen-Edit, this is the node that turns "move the camera" from a coin flip into a structured instruction set.