ArchViz Camera Prompt Builder

If you've ever written "slow dolly-in toward the kitchen island" into a Wan or LTX prompt and prayed the model actually read it, this node is the pack's answer to that prayer. ArchViz Camera Prompt Builder turns a few dropdowns into one consistent, carefully worded camera-motion sentence - no prompt engineering, no copying the same camera wording four different ways across a shot list.

It ships in comfyui_brick_tools, the free toolkit from Brick Visual (a real architectural-visualization studio - the pack is basically their in-house pipeline made public). Camera work is the whole game in archviz video, and this node is where that thinking lives.

How it works

Pick a movement_style - Linear, Orbit, Combined, or Static - and the node hands you only the action_type moves that belong to that style. Linear gets dollies, tracks, pans, tilts, booms; Orbit gets arcs and spiral orbits; Combined gets the cinematic stuff like the Dolly Zoom; Static gets locked-off framing. Choose a speed_modifier ("Slow and smooth", "Dynamic", and two others), and it stitches everything into a full sentence using templates like "slow and smooth forward dolly shot pushing directly toward the kitchen island, using a linear camera movement style…".

Two booleans are on by default and worth knowing:

• lock_target_subject keeps the camera aimed at whatever target_subject_preset you chose (building facade, staircase, living room space, courtyard, … 10 presets). Flip it off and the node writes "the architectural space" instead and tells the model to move freely.
• stability_reinforcement appends the anti-jitter clause: "maintaining absolute camera stability, perfectly smooth motion curves, and zero organic handheld shaking." That's the studio's fix for video models that default to wobble.

The output is a single CAMERA_PROMPT string. Wire it into your video model's text prompt or conditioning - Wan 2.2, LTX, Hunyuan, whatever reads natural-language camera descriptions.

The traps

• If you pick an action_type that doesn't belong to your movement_style, the node silently falls back to that style's first action rather than erroring. The dropdowns are linked, so you mostly can't trigger this - but know it exists.
• Leave target_subject at its default [Subject] and that literal text goes into your prompt. Fill it in or pick a preset.
• It's a prompt writer, not a guarantee. LTX is famously picky about camera prompts and may ignore elaborate ones entirely; this gives you consistent, well-formed input, and the model still decides what happens. Treat it as removing one variable, not removing all of them.

Install

Same as the rest of the pack - no models, no keys, no requirements file:

cd ComfyUI/custom_nodes
git clone https://github.com/brick-momen-sirri/comfyui_brick_tools

Restart ComfyUI and hard-refresh the browser (it ships frontend JS). Or use ComfyUI Manager and search comfyui_brick_tools. That's the whole install.

A niche tool, and an honest one: it writes one specific kind of prompt really well, and it's free. If you're doing camera-driven archviz video, that's worth having.