RedNode Prompt Frame

Word order is framing. Open a prompt with the person and the model gives you a close shot; put the room first and the camera pulls back - same words, different picture. And once a prompt is long enough to be useful, that order is the first thing to get lost. RedNode Prompt Frame exists because of exactly that: it splits your writing into Subject and Surroundings, then emits them in the order that sets the shot you actually asked for.

That's the whole mechanism, and it's worth sitting with because it explains why the node feels magic and isn't. It never rewrites your wording. It reorders it and changes the joining words, and it adds small scale cues at the wide end ("seen full length", "small in the distance"). The model's own training did the rest of the work; this node just stops you from fighting it.

The five steps

framing runs Portrait → Half body → Balanced → Full scene → Roomscale. The two tight steps lead with the subject, opening with "A close view of" or "A three-quarter view of". Balanced and the wider steps lead with the surroundings and place the subject inside them, adding the scale cue at the two widest.

Where people get burned is the wide end: at Full scene and Roomscale, a subject with nowhere to stand can drop out of the picture entirely. The fix is placement - "on the second bench", "at the water's edge", named against something already in your Surroundings. There are two dropdowns for quick placement (a where + a what, like "beside" + "the doorway"), but typing your own placement overrides both. The node will also tell you when you need it, because the second output is a notice line that does real work: it flags a wide framing with no placement, a style that lights its own scene and will fight your lighting dropdown, a prompt pushed past the 90–150 word working range, and a Portrait framing whose surroundings are long enough to pull the camera back on their own. Wire that notice into a report or a text display and the node starts coaching you.

The other controls, briefly

lighting is twenty setups written the way the model actually follows them - source, direction, what the light lands on. style is nine tested blocks that need no LoRA. brightness is a separate ramp from lighting, and zero means "say nothing". light_and_colour goes at the end and doesn't get rearranged. seed and resolve_wildcards behave exactly like the Prompt Box - wildcards and @keywords resolve in the finished text, seeded, and the box can pass them through instead. The four *_in inputs append external text (pair light_and_colour_in with RedNode Describe To Boxes), and framing_push restates the shot at the end of the prompt when a long prompt is talking over the middle of it.

Outputs: prompt (the finished STRING) and notice (the warnings line).

Installing it

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

Restart ComfyUI, or search RedNode Studio in ComfyUI Manager. No pip deps, no models. Python 3.10+.

The takeaway

Start at Balanced and pull wider only when you have a placement typed in. The panel previews what the node emits, so the thing you see on the canvas is the thing that renders - use that to verify the ordering before you burn a run. This is one of the few prompt tools where the mechanism (order as framing) is real enough that once you've used it, you'll start writing subject-last for wide shots in plain prompt nodes too.