System Message Video Presets (Legacy)

If you've used the pack's plain SystemMessagePresets node, you know the drill: a dropdown of tuned system prompts for image editing, ready to wire into an encoder. This is the video-flavored twin. Same dropdown, same ~80 presets - but the text it hands out has been rewritten for a video context before it reaches your encoder. It's for the video-edit models and pipelines where the system message needs to talk about motion, timelines, and shots rather than stills.

How it works

Identical interface to its sibling: pick preset, get a system_prompt STRING out, with the same escape_parentheses toggle (default off) for downstream parsers that choke on parentheses. The difference is invisible unless you read the code: before returning, the selected template runs through a converter that rewrites image-editing language into role-aware video guidance - "image" → "video" phrasing, composition instructions reframed around temporal consistency, and a guidance suffix appended if it isn't already there.

That matters more than it sounds. A system prompt written for single-image editing says things like "keep the composition and structure of the image same as reference." Fed verbatim to a video edit model, the model nods along and then has no instructions about motion, timing, or frame consistency - so you get a video that holds the composition but drifts in exactly the ways the prompt never covered. The video wrapper patches that gap so the same curated personas (photorealism, anime, analog film…) carry over to motion work.

When to use it over the stills version

Any pipeline where your "system prompt" slot belongs to a video model - LTX 2-class editors, MiniMax H3, and the video-conditioned encoders in this same pack. The style presets all work here, because an editing persona ("you convert editing requests into one concise instruction…") transfers fine to video; it's the framing that needed adjusting. If you're editing stills, use the plain SystemMessagePresets node instead - the video wrapper will rewrite words you didn't ask to change.

The usual gotchas

• It only hands you a STRING; the actual encoding happens downstream in whatever video encoder you wire it into.
• The preset list is the same huge style list as the stills version - F2_SYSTEM_MESSAGE is the sane default at the top.
• Don't feed video-wrapped prompts to a stills encoder and vice versa; the rewriting is tuned for its audience.

Installing it

Ships in silveroxides/ComfyUI-UtilsCollection. ComfyUI Manager: search ComfyUI-UtilsCollection, install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/silveroxides/ComfyUI-UtilsCollection
cd ComfyUI-UtilsCollection
pip install -r requirements.txt   # opencv-python, typing-extensions

Restart. No models, no downloads. Legacy alias of canonical UC_SystemMessageVideoPresets - same behavior, "(Legacy)" just keeps old workflow JSONs resolving.