Rabbit Camera Perspective 🐰

What if picking a camera angle was as fun as spinning a toy? That's the pitch here. RabbitCameraSelector renders a little 3D rabbit inside the node, you drag it (or scroll to zoom), and it writes the exact camera prompt Fal's Multiple-Angles LoRA wants - something like <sks> front view eye-level shot medium shot - into a text widget, onto your clipboard, and out a STRING port. It's a visual prompt previewer built for Fal's Multi-View LoRA on Qwen-Image-Edit-2511, pitched squarely at portraits. The name is a lie in the best way: it calls no API, needs no key, and downloads nothing.

You'd reach for this for the same reason anyone builds camera-control nodes in the first place. Posing and framing a character used to mean an OpenPose skeleton plus a ControlNet checkpoint. In the Qwen-Image-Edit era it became a LoRA - camera framing turned into a prompt phrase, and the community's Multiple-Angles LoRAs blew up precisely because "new angles of a subject this well" was never a strength of edit models before. The hard part isn't the rabbit; it's that the LoRA cares about exact phrasing (8 view directions × 4 height angles × 3 shot sizes), and you don't want to memorize that while a generation is waiting on you. This node makes the syntax physical: drag, read the prompt, go.

How it works

The Python side is a pass-through, and I mean that literally. RabbitCameraSelector has one required input and its run function just returns it. Every bit of actual behavior lives in web/js/main.js, which registers a frontend extension that builds the UI when the node spawns. There's no Three.js, no WebGL, no external dependency - the rabbit is spheres and ellipsoids hand-projected onto a 2D canvas, painter's-algorithm sorted and backface-culled in vanilla JavaScript. It all runs in your browser, which is why this pack has zero Python dependencies and installs in seconds.

Three dropdowns cover the axes: Direction (front, front-right quarter, right side, back-right quarter, back, back-left quarter, left side, front-left quarter), Angle (low-angle, eye-level, elevated, high-angle), and Shot (close-up, medium shot, wide shot) - 96 combinations in total. Drag the rabbit and all three update at once (horizontal spin picks direction, vertical tilt picks angle, zoom picks shot size), and the rabbit re-orients to match if you touch a dropdown instead. Every change rewrites the prompt widget and fires it to your clipboard.

Inputs and outputs

prompt (STRING, multiline) - the one input, and it's really the node's display box. The JS writes the generated text into it, so you normally leave it alone. Its default is the friendly "Click or rotate the rabbit to generate prompt."
STRING output - the generated prompt. Wire it into your CLIP Text Encode / prompt conditioning for the Qwen-Image-Edit-2511 workflow, or anywhere else your prompt text needs to travel.

One gotcha on that input: it's a widget, and the node owns it. Type your own prompt in there, then drag the rabbit, and it gets clobbered. Use it as the readout it's meant to be.

Installing it

Two ways, both painless since there's nothing heavy:

ComfyUI Manager → search "Rabbit Camera Perspective" → Install → restart.
Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/SamLiu1000/ComfyUI_Rabbit-Camera-Perspective

Then restart ComfyUI. No pip requirements, no model files in the repo, no API key. The catch: the node does not include or apply the LoRA. For that <sks> token to mean anything, you need Fal's Multiple-Angles LoRA from Hugging Face (fal/Qwen-Image-Edit-2511-Multiple-Angles-LoRA), loaded onto Qwen-Image-Edit-2511 in your graph.

Common issues

The output does nothing. Almost always the LoRA isn't loaded or isn't attached to the Qwen-Image-Edit-2511 checkpoint. Without it, <sks> is just noise.
The rabbit doesn't appear. The whole UI is a frontend extension registered at load time. Hard-refresh (Ctrl/Cmd+Shift+R) after installing, and check the browser console - the pack's own history includes a "Failed to fetch" fix for the script import, so a stale cache can mimic a broken install.
Clipboard copy doesn't fire. navigator.clipboard wants localhost or HTTPS; on 127.0.0.1 it works, and elsewhere it degrades to showing the text in the node, which is cosmetic.

The node is new and barely surfaced in the community yet - 0 search impressions, one commit, no drama. But as a prompt-syntax crutch for the FAL multiple-angles LoRA, it does exactly one job and does it with a smile. If you're generating character turnarounds and want to stop guessing whether it's "front quarter" or "quarter front," this is a pleasant little upgrade to your workflow.