RedNode Camera Multi-Angle

You've generated a frame you like, and now you want the same scene from the side, or from up high. On Krea 2 that's genuinely hard - it's a text-to-image model, not an editor, so "now move the camera" just invents a new scene. The community's answer is a specific LoRA: fal's Qwen-Image-Edit-2511 Multiple-Angles, which takes a photo and the token sequence <sks> azimuth elevation distance and re-renders the picture from that viewpoint. RedNode Camera Multi-Angle writes that sequence for you, straight from the Camera Studio's geometry.

This is the node that makes your stage do double duty. The same Camera Studio that planned the original shot - subjects, camera position, lens - feeds its camera_json in here, and out comes the exact trigger phrase for the Multiple-Angles LoRA. No eyeballing "was that front-right or right-front?" The geometry is converted to the LoRA's own coordinate bands, so what you set on the map is what the edit does.

How it works

The LoRA only knows 96 viewpoints: 8 azimuths, 4 elevations, 3 distances. The node maps the studio camera's position and pitch onto those bands - an azimuth word for where around the subject the camera sits, an elevation word for the tilt (low-angle, eye-level, elevated, high-angle), and a distance word (close-up, medium, wide) computed from the actual field of view at your focal length. Then it assembles the phrase: <sks> front view eye-level shot medium shot.

Two settings show real attention to detail. right_means lets you flip which right the LoRA's "right side view" is - the author verified on a render strip that it's the viewer's right (we see the subject's left side), but you can flip it if your output comes out mirrored. And nudge adds plain-language steering - "rotate the camera a little more to the left" - when the camera sits between two of the LoRA's bands. Worth knowing: height nudges were tested and do nothing, so the node doesn't write them.

The inputs and outputs that matter

With camera_json wired, the three pickers (azimuth, elevation, distance) are ignored - the stage overrides them. Without it, they're the whole show, which is a fine manual fallback. trigger is the LoRA's token and you should keep it as <sks>. collapse_same matters once you use a camera path: it merges consecutive shots that land in the same bands, so you don't render the same viewpoint twice - with only 96 bands, a fine orbit would absolutely repeat itself. extra appends plain words after the camera prompt.

Outputs give you prompt plus the separate azimuth, elevation, distance, and degree values as lists - handy if you want to display or log what a shot mapped to - and a report string that says how many shots collapsed into how many distinct viewpoints.

Install and what it needs

Same pack, same install:

git clone https://github.com/RedNodeAI/ComfyUI-RedNodeStudio.git ComfyUI/custom_nodes/ComfyUI-RedNodeStudio

Restart, or search RedNode Studio in ComfyUI Manager. Then you need the Multiple-Angles LoRA for Qwen-Image-Edit-2511 yourself - it's not bundled. The pack ships an example workflow, RedNode_MultiAngle_QwenEdit2511.json, that shows the whole chain: Camera Studio geometry becomes the Multiple-Angles prompt, the edit LoRA re-angles the photo, and Krea 2 finishes the frame.

The honest caveats

The edit quality lives and dies with that LoRA, not this node - this node only writes the words. And because it's built on Qwen-Image-Edit, you're running the 20B edit model, which is a heavy ask on mid-range cards; budget for it. The other thing people miss: the prompt output is a list even for a single shot. That's the normal shape of things here, not a bug. If your re-angle comes out mirrored, check right_means before you blame the LoRA.