How do I make a talking-avatar video in Higgsfield, and how many credits is one minute?

To make a talking avatar on Higgsfield, open Lipsync Studio or the AI Talking Avatar page, add a portrait photo and a script (or an audio track), pick a model and press Generate. One minute of talking video costs about 108 credits on the cheapest 720p route we could read: Lipsync Studio's Wan 2.5 Speak Fast showed 18 credits for a 10-second 720p take, and a minute is six takes. That's about $5.30 at the Plus plan's monthly rate. A Seedance 2.5 presenter at 720p runs about 420 credits a minute. Every number here was read on Higgsfield's own pages on November 3, 2026; we didn't generate a clip for this update, and retakes cost extra.

What is Higgsfield's AI talking avatar?

A tool that makes a photo speak your script, with the lips matched to the audio. Higgsfield describes it as making "a lifelike, lip-synced presenter from a photo and a script or audio." It sits on top of Higgsfield's video and voice models, so the same plan that makes the avatar also makes b-roll, product shots and other video.

What the avatar page lists, read November 3, 2026:

• Your face or a generated one. "Upload your own portrait photo, with consent, or generate a new face from a prompt on Higgsfield."
• Voices. Preset voices from Higgsfield Audio, ElevenLabs, or your own cloned voice; the page says the cloned voice is used "With your consent."
• Ready-made presenters for ads. "For product ads, Marketing Studio adds 100+ ready avatars."
• Several speakers. Eleven v4 dialogue, "up to 10 voices in one script", each lip-synced to its own face.
• Formats. "Reframe to 9:16, 1:1, or 16:9, and upscale up to 4K."
• Models. Cinema Studio (scenes up to 30 seconds), Seedance 2.0, Veo 3.1, Kling 3.0 and Wan 2.7, switchable per take.

How do I make a talking avatar video with Higgsfield?

Three steps on the website, in Higgsfield's own order. The fastest place to do it is Lipsync Studio, whose panel has exactly these fields: an image upload, "Audio text" or "Generate Audio", a prompt, then Model, Duration and Resolution above the Generate button.

Tips from Higgsfield's own walkthrough that save credits:

• Start from a neutral mouth. Higgsfield's blog says "Lips gently closed is not a style choice: the video model animates speech more cleanly from a neutral mouth than from a caught expression."
• Make the speech before the video. In its pipeline "the audio file is an input to the clip".
• Fix the first clip before making a series. "Fix anything here, before the series, not after."

Can I keep the same presenter across many videos?

Yes, by training an identity once. Higgsfield's Claude walkthrough (August 12, 2026) trains a Soul ID on a reference set and recommends "from 20 to 80 photos for the best character consistency", with training that "runs for about 10 minutes". After that, it says, "the character is callable by name in any prompt". Each new clip then needs only a new script, a start frame and a prompt. Higgsfield warns that "Re-attaching reference images per generation is where face drift comes from."

How many credits does one minute of talking avatar cost?

About 108 credits on the cheapest 720p route we read, about 84 if 480p is enough, and about 420 on Seedance 2.5. Higgsfield doesn't publish a per-minute price for avatars; it prices each generation by model, length and resolution and shows the cost on the Generate button. Most avatar models also cap one take at 10 or 15 seconds, so a minute is several takes cut together. Here is what we read on November 3, 2026, turned into a minute by our arithmetic:

What the minute figures leave out:

• Retakes. The table assumes every take works the first time. Budget for some redos; Higgsfield's walkthrough says to "Fix anything here, before the series, not after."
• The speech. If you type the script, the voice is generated first. Higgsfield's walkthrough priced Seed Audio speech at "about 1 credit" for a script around 10 seconds and "about 3.8" around 26 seconds, small next to the video.
• A one-time identity. If you train a Soul ID to keep the same presenter across videos, the walkthrough lists training at 25 credits, once.
• Editing. Six 10-second takes still have to be cut together in an editor; Higgsfield's avatar page says its exports reframe to "9:16, 1:1, or 16:9" and upscale "up to 4K".

What a plan buys, by our arithmetic: Plus ($59 a month for 1,200 credits) covers roughly 11 minutes of talking avatar on the 108-credit route, or just under 3 minutes on Seedance 2.5 at 720p. Higgsfield's pricing page itself describes Plus as "~ 80 Seedance 2.5 videos". Other plans: Starter $19 for 270 credits, Plus $47 before the 11.11 sale ($35 until Nov 14) a month billed yearly, Ultra $129 ($99 before the 11.11 sale ($65 until Nov 14) yearly) for 3,000. Our credits guide explains how credits are spent and reset.

What does Higgsfield's own walkthrough say a series costs?

Higgsfield's blog post on making a talking avatar in Claude (August 12, 2026) built a four-clip series with one presenter and put the total at "~400" credits, with a 14-second 720p Seedance 2.5 clip at "~91". It says "Almost all of the cost sits in two places: the talking clips and the one-time identity training."

Is a talking avatar free on Unlimited?

Not through Claude. Higgsfield's avatar page says "Unlimited models and Free Generations on plans are accessible only via higgsfield.ai and are not accessible on MCP/CLI, Canvas or Supercomputer." Its walkthrough repeats that "anything generated through MCP spends plan credits." On the website, whether a model counts as Unlimited depends on your plan; see Higgsfield Unlimited explained.

How many languages can the avatar speak?

Eighteen for dubbing, per Higgsfield's page on November 3, 2026: "AI Talking Avatar dubs the same avatar into 18 languages. Lips synced per language." Older pages, including an earlier version of this one, gave a higher number; the avatar page is the current source.

Can I use my own face or a client's face and voice?

Yes, with consent, and the terms put that duty on you. Section 5.3 of Higgsfield's Terms of Use (last updated October 23, 2026) says that if you upload photos or video containing someone's face, or recordings containing their voice, you promise "you have obtained all consents, releases, and permissions required under applicable law from each individual whose face or voice is included". It also says Higgsfield doesn't use that media to identify people.

In practice: your own face and voice are fine; a founder or spokesperson needs a written release; a celebrity, a customer from a review video or a stranger's photo is off limits. Higgsfield's walkthrough says voice cloning works from "a recording of 10 seconds to 3 minutes" and tells you to "clone only a voice that is yours or that you have permission to use." Record that sample somewhere quiet, on a decent microphone rather than a laptop's built-in one; our USB mic picks start with plug-and-play models.

Can I use a talking avatar in ads, or make them for clients?

Yes, under Higgsfield's terms, but check which tool fits the job first. Section 4.4 says Higgsfield doesn't claim ownership of outputs or restrict commercial use, and you "may transfer or sublicense your rights in Outputs to your clients". The full rights picture, including training and who carries the legal risk, is in can you use Higgsfield output commercially.

If you're an agency or brand making avatar ads for clients, compare before you commit: we put the per-30-second cost, avatar libraries, team seats and client-rights terms of the three main tools side by side in Higgsfield vs HeyGen vs Arcads for AI avatar and UGC ads. In short, that comparison found HeyGen cheapest per talking minute for explainers, Arcads built for testing many UGC-style variants, and Higgsfield costing more per talking second but covering product shots and b-roll on the same plan.

Do I have to say the video is AI?

Where the law requires it, yes. Higgsfield's section 5.5 says "Where required by applicable law, you will disclose that Output is artificially generated or manipulated", and section 6.4 says it may embed machine-readable provenance markings (C2PA-style), which it may make imperceptible. Ad platforms have their own labels for synthetic people, and presenting an AI actor as a real customer in a testimonial raises separate advertising rules; our HeyGen and Arcads comparison covers that question.

Who is a Higgsfield talking avatar for?

Higgsfield's page names "marketers, agencies, and DTC brands", course and training teams, founders who want a digital twin of themselves, and sales and support teams. It fits best when you need short presenter clips next to other video work on one plan: a product explainer with b-roll, a course intro, a localised version of one message.

It's a weaker fit for long talking-head lessons, where per-minute avatar tools cost less, and for anyone who wants a camera-free presenter without paying per clip. AI presenters also still struggle with long takes and big gestures; keep scripts short and shots simple. For a faceless channel built around a presenter, see how to start a faceless YouTube channel with AI.

What we couldn't verify

• Output quality. We didn't generate an avatar for this update, so we make no claim about how natural the lip sync looks.
• Cost on other models. We read credit costs only for Wan 2.5 Speak Fast and Wan 2.5 Speak in Lipsync Studio, Seedance 2.5 and the walkthrough's clip. Premium lip-sync models (Kling Avatars 2.0, Higgsfield Speak 2.0, Infinite Talk) couldn't be selected signed out, and we didn't price Veo 3.1, Kling 3.0, Wan 2.7 or Cinema Studio.
• Per-minute figures. Higgsfield prices per take, not per minute; our minute numbers multiply the take cost and assume the per-second rate holds, with no retakes.
• Which 18 languages. The page gives the number, not the list.
• Older language claims. Earlier Higgsfield pages and an earlier version of this article gave a higher language count; we now use the current page.

Comparing dedicated avatar tools too? The best AI avatar video generator prices a finished talking-head minute on HeyGen, Synthesia, Arcads and Higgsfield.

Lip-syncing a song rather than speech is a different job: the best AI music video generator prices the footage for a 3-minute song and the lip-sync options for sung shots.