arkennemasis fal · Sync React-1 · $10/min

Most lip-sync nodes answer one question: do the lips match the audio? React-1 asks a second one: does the performance match the line? Its selling point is emotion - you hand it a video, an audio track and a single word like angry, and it refines the facial expression, not just the mouth. Then it charges $10 a minute, which makes it the most expensive per-minute node in this pack, and caps you at fifteen seconds per request.

Both of those facts are in the node, not in marketing copy. Fifteen seconds, enforced:

video  - must be 15 seconds or shorter
audio  - must be 15 seconds or shorter

The 15-second cap is the whole workflow

If you're dubbing anything longer than a quarter of a minute, you're going to chunk it. Plan for that before you spend: split the clip and the audio into matching 15-second segments, run each through this node, and join the results. Get the split points on natural pauses or cuts - the pack's assembly chain will glue clips back together, but it can't hide a lip-sync seam in the middle of a word.

Cost, concretely: 15 seconds at $10 per minute is about $2.50 per segment. A one-minute talking head is four calls, about $10. That's the honest number, and it's also why you don't prototype on React-1 - settle the script and the voice first with the cheap speech nodes, then spend here.

The inputs that shape the performance

• video and audio - the clip and the track. Both capped at 15 seconds.
• emotion - a single word: happy, sad, angry, disgusted, surprised, neutral. Single-word only, per the field's own description. This is the reason to choose React-1 over Sync 2.
• model_mode - how much the model is allowed to touch. lips = lip sync only, minimal facial change. face = lip sync plus facial expressions, no head movement (the default). head = the full thing, expressions and natural talking-head motion. If your source clip is a calm locked-off shot and the output starts bobbing around, you left it on head.
• lipsync_mode - what happens when the two lengths disagree: cut_off (default), loop, bounce, silence, remap. Match the durations yourself and this stops mattering.
• temperature - 0 to 1, default 0.5, described as the expressiveness of the lip sync. Turn it down when the mouth is over-performing, up when it looks dead behind the eyes.

Outputs and the fal plumbing

video is the results as a ComfyUI VIDEO value - save it, dub it, join it. video_path is the file under output/fal/fal-ai_sync-lipsync_react-1/. info is the JSON receipt with the request id, cost estimate, saved files and fal's answer.

The shared fal behaviour applies: the key comes from FAL_KEY in the .env next to your launcher (no key widget exists on any node in this pack), the cost is estimated and checked against max_cost_usd before anything is uploaded, and max_concurrent (default 1) throttles paid calls across the run so a canvas of lip-sync segments doesn't fire four submissions at once. reuse_identical_run (default on) saves you from paying twice for an identical segment - worth checking before you re-run a batch you already paid for.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Hishamahmer/comfyui-arkennemasis

Restart ComfyUI, or search arkennemasis in ComfyUI-Manager. Only httpx is declared, for the pack's Codex nodes; the fal provider uses Python's standard library, so nothing heavy gets installed.

Where people get burned

• You fed it a 40-second clip. Pre-flight rejects it; nothing bills. Chunk it - counting in 15s segments from the start, not from where it feels natural.
• Mismatched durations. You'll get cut_off behaviour and a lip-synced clip that stops early. Set the audio length first, then the video.
• The face won't stop moving. model_mode on head when you wanted face.
• Looks worse than Sync 2 on a plain talking head. It can - React-1 is buying you expression, not fidelity. If the mouth is the only thing you care about, Sync Lipsync 2 costs $3 a minute instead of $10.
• The clip and the performance both look flat. emotion left at neutral and temperature at 0.5 is the safe setting, not the expressive one.