You make one in Lipsync Studio, which sits under Video in the left nav, not inside the image or video generator. Upload a character image or a clip you already shot, type the line you want spoken, pick a model, pick a scene template, generate. That is the whole workflow.

The part that decides whether your clip is usable is the model you pick in step three. There are eight of them and they are not interchangeable.

If you want to test one before committing to a plan, a referral link carries up to 50% off for three hours: start with the discount applied. We may earn a commission if you subscribe through it, at no extra cost to you.

The eight models, and which one is for you

Higgsfield splits these into image-to-video, where your avatar comes from a still, and video-to-video, where the avatar is already moving. That is the first decision. A video-to-video model cannot fix a still, and an image-to-video model has nothing to lip sync onto.

ModelInputWhat it is actually good for
Google Veo 3.1Image to videoCinematic talking video, the closest to a produced spot
Kling 2.6 LipsyncImage to videoUp to 1080p with audio, the general-purpose pick
Wan 2.5 SpeakImage to video480p to 1080p with audio, cheaper resolution steps
Kling Avatars 2.0Image to videoTalking avatars on longer clips
Higgsfield Speak 2.0Image to videoPriority-queue speed, the fastest of the stills set
Infinite TalkImage to videoLong-form talking video, the one for a full monologue
Kling LipsyncVideo to videoLip sync onto footage you already have
Sync Lipsync 3Video to videoPrecise lip sync at up to 4K

Read the last two rows again if you shoot real footage. Most people assume a talking avatar tool means generating a fake person, then find the useful version is fixing the audio on a clip of themselves they already like. Sync Lipsync 3 is the only row in that table above 1080p.

Lip sync after the fact vs audio in the same pass

Lipsync Studio adds a mouth to a picture that already exists, driven by your audio after the render. Six separate video models do something different: they generate audio and picture in a single pass, so lip movement, room sound and physics align from the start. Higgsfield's docs call it native audio co-generation and list Seedance 2.5, Kling 3.0, Seedance 2.0, Flux 3, MiniMax Hailuo 3.0 and Wan 2.5, with Cinema Studio producing native audio across all versions.

If the face is close to camera and the line is short, Lipsync Studio is faster and cheaper. If you want a character talking in a room with sound that matches the space, generate it natively instead of fighting the mismatch.

The scene template does more work than your prompt

Each lip sync model ships with templates: General, Selfie, Podcast, Car Talking, and more. They set the framing and the look before your prompt gets a chance to matter, which is why the same prompt across two templates produces clips that look like two different tools.

Pick the template first, then write the line. A long cinematic prompt inside a Selfie template is how people conclude the model is bad when the frame was never going to hold the instruction.

Audio is a separate tool: Text to Speech, Voice Change and Translate. Translate re-voices a clip in 18 languages, and you can clone a voice from an MP3 or WAV upload.

What a talking clip costs per day

Higgsfield bills by credits, the cost of each generation appears on the Generate button before you confirm, and your plan sets a daily budget rather than a per-clip rate. Here is the derived number that matters for a posting schedule.

PlanPriceMonthly creditsCredits per dayCost per credit
Starter$19/mo2709$0.070
Plus (annual)$47/mo1,20040$0.039
Ultra (annual)$99/mo3,000100$0.033
Credits per day by plan

Monthly credits divided by 30, which is the honest way to read a plan you will spend down. Annual plans reset every 30 days rather than on your signup date, so the reset day drifts.

0501009Starter40Plus annual100Ultra annual
Data table
ItemValue
Starter9
Plus annual40
Ultra annual100

Nine credits a day on Starter is the trap in this category. A talking clip is a video generation with audio attached, and it is not the cheapest thing on the platform. If your answer to how many you need is one a day plus retries, Starter is not the plan.

The bottom line: decide first whether your avatar starts as a still or as footage, then pick the model, then check the credit number on the Generate button before you commit to a plan size.