ArchAi3D_Qwen_Environment_Navigator

The focus-camera nodes answer "show me this thing from another angle." This one answers a different question: "walk me through the space." ArchAi3D_Qwen_Environment_Navigator is the pack's movement node - 14 navigation patterns that move and rotate the camera through a scene, which is how you build interior walkthroughs and cinematic establishing shots out of still-frame Qwen edits.

The panel

• preset - 14 patterns: interior walkthroughs (forward, pan right/left, strafe), landscape navigation (forward, rise, 360° pan), architectural (approach, circle, flyby), and cinematic (retreat, rise, descent). Pick one and the node fills in movement, distance, rotation and speed for you - or pick custom and drive everything yourself.
• scene_context - describe the space ("modern office hallway with glass walls", "dense forest with morning mist"). The node's own examples make the point that a good scene description is what keeps the walkthrough from inventing a different room.
• movement - forward, left/right strafe, up/down (drone-like vertical).
• distance (1–20m) - how far the camera travels. The tooltip's honest note: longer = more dramatic transition, i.e. more room for the model to drift.
• rotation - none, left/right (pan), up/down (tilt), combined with movement for fluid "move forward while rotating right" paths.
• speed - slow (deliberate), normal, fast (dynamic), smooth (cinematic). This changes the wording of the instruction, since Qwen reads "smoothly" differently from "quickly."
• maintain_focus + focus_subject - keep a subject in frame during movement for tracking shots. Enable it and name the subject ("the building", "the person").

Outputs

prompt (for the encoder), preset_description (what the preset actually does, empty in custom mode), and system_prompt - the per-style system prompt that keeps the model operating as a camera operator rather than a scene-rewriter.

How it works

Same foundation as the whole camera family: plain-language camera instructions, which Qwen-Edit 2509 responds to reliably because it understands distance and direction words. The navigation twist is combining movement and rotation into one sentence - "move smoothly 10m forward while rotating right" - and the presets are just curated versions of those sentences. The research insight baked in here (and it matches the pack's docs and the broader Qwen-Edit camera-LoRA scene) is that distance-based positioning like "10m forward" beats arbitrary degree rotations for reliability.

The honest take

The catch with navigation, which the pack is upfront about, is that a 10–20m camera move is a big change for an edit model to hold together - the further the camera travels, the more the model has to invent. Short hops (3–5m) per frame read as smooth walkthroughs; jumping the full 20m in one edit invites the scene to drift. If you're building frames for an FLF video or walkthrough, do multiple small navigations per step instead of one heroic move. And the known family physics apply: people in frame get moved instead of the camera, and left/right is picture-relative.

Install

Pack install, once:

cd ComfyUI/custom_nodes
git clone https://github.com/amir84ferdos/ComfyUI-ArchAi3d-Qwen.git
cd ComfyUI-ArchAi3d-Qwen && pip install -r requirements.txt

Or ComfyUI Manager → "ArchAi3d Qwen". Restart, then it's under ArchAi3d/Qwen/Camera. You need Qwen-Image-Edit weights (quantized on consumer cards) and one of the pack's encoders to turn the prompt into conditioning.