FFMPEGA Effects Builder

The FFMPEGA Effects Builder is the pack's answer to a very specific frustration: you asked an AI agent to "make it cinematic" and it decided what that means. Fine for a first pass, but when you want to be the editor instead of the audience, this node is where you grab the controls. Pick effects from dropdowns, dial in numbers, and get a reproducible pipeline you can run forever without an LLM in sight.

It's not a standalone editor. Think of it as a configurator: you stack up to five effects, each with its own parameters, and it spits out a single JSON pipeline spec that the FFMPEG Agent node executes for you. That means you wire the Effects Builder's output into the Agent's pipeline_json input, set the Agent's llm_model to none (its default) and no_llm_mode to manual (also the default), and you've got deterministic, scriptable video editing that costs zero tokens and hallucinates nothing.

The inputs that matter

• preset - 27 dropdown options (Cinematic Look, Vintage/Retro, TikTok Ready, AI Upscale, Concat + Crossfade, and more). Pick one and it auto-fills the effect slots and params below, which you can then tweak. Set to none to build your own combo from scratch.
• effect_1 through effect_5 - the effects themselves, 209 of them organized by category (Visual, Temporal, Spatial, Audio, Encoding, Outcome). Chain order matters: effect 2 runs after effect 1, and so on. Each has a matching effect_X_params JSON field, e.g. {"radius": 5} for blur or {"factor": 2.0} for speed.
• raw_ffmpeg - a raw FFmpeg -vf filter string applied after the skill effects. This is the escape hatch for anything not in the 209. If you know your way around FFmpeg filter graphs, you can bolt on things the presets don't cover.
• sam3_target / sam3_effect - target an object instead of the whole frame. Set sam3_target to "face" (or whatever SAM3 should find), pick sam3_effect (blur, pixelate, remove, grayscale, highlight), and only that region gets the effect. This is how you blur one person in a crowd.

The only output is pipeline_json, which is also its whole job: feed it into the FFMPEG Agent's pipeline_json slot.

Installing it

You get this node as part of the ComfyUI-FFMPEGA pack, so install is the pack install: ComfyUI Manager → search ComfyUI-FFMPEGA → Install, or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/AEmotionStudio/ComfyUI-FFMPEGA.git
cd ComfyUI-FFMPEGA
pip install -r requirements.txt

Restart ComfyUI. FFMPEG must be on your PATH (ffmpeg -version should work). Nothing else is required - the core editing skills need zero model downloads. If you use a sam3_target, the ~300 MB SAM3 model auto-downloads on first use; flip allow_model_downloads off on the Agent node if you'd rather not.

Where people get burned

The effect_X_params fields are plain JSON - a missing comma or a stray quote silently resets things. Values are auto-clamped, so don't expect an error if you ask for a blur radius of 500; you'll get the max. Some effects need extra inputs - xfade and concat want a second clip, and that comes from the Agent's video_a slot, not from this node - leave them unconnected and the effect will fail or no-op. And while the README's advice applies pack-wide, it bites hardest here: one task per run. Chain five heavy effects on a long video and you'll watch your render time balloon and wonder which one is the culprit.

Reach for this node when you want edits you can save, reload, and trust - it's the difference between "the agent felt like it" and "here's exactly what happened."