ElevenLabs SFX v3 and VEED Dubbing in ComfyUI: two audio upgrades

SFX v3, per ElevenLabs' docs. The new model (model_id eleven_text_to_sound_v3) follows prompts with several instructions more closely than v2, handles music and vocal sounds better, sounds more natural, and is less likely to clip on high sounds like chimes and whistles. That means one prompt like "football crowd cheering, a ball drops, crowd booing" can come back as a single sequenced effect instead of three generations you stitch together. It's in ElevenCreative (Sound Effects, Flows and Studio) and the API, with clips up to 30 seconds, seamless loops and 48 kHz WAV for non-looping effects. Note the docs call it limited access that may not be available in every region; accounts without it get an "Unknown model" error.

Prompting tips from the same docs. Describe the action, setting and mood rather than a category ("footsteps on gravel approaching slowly", not "footsteps"). Wording matters: their example says "a couple arguing in French" beat "a couple fighting in French". Exact counts like "triangle hit 4 times" aren't reliably followed. ElevenLabs itself says results vary by prompt, so run your usual prompts through v2 and v3 side by side before switching. Many of the glowing replies under the launch post read like a coordinated creator push, so we'd weigh the docs and your own ears over the hype.

VEED Dubbing in ComfyUI. A new partner node (merged Oct 27, announced by ComfyUI on Oct 29) translates the speech in a clip into another language, clones each speaker's voice and lip-syncs the result. Per the node code, clips can be up to about 10 minutes and need a visible face. The standard tier is cheaper with mono audio (its Spanish is Latin American, its Portuguese Brazilian); the Pro tier, per ComfyUI, keeps the background audio and handles multiple speakers, with a wider list of accents. Like other partner nodes it runs on Comfy API credits, with the price shown on the node. Only dub people who've agreed to have their voice cloned.

Our take: SFX v3 is the one to try first if you build sound beds for AI clips, because sequenced prompts cut out the stitching step. VEED Dubbing matters most for anyone already localising ads or explainers inside ComfyUI, since translate, voice and lip-sync now sit in one node after your edit.