Model · Talking presenters, Video

FantasyTalking

A free research model that makes a portrait photo talk: give it a picture and a voice recording and it produces a video of the person speaking with matching lips and head movement. Built on Wan 2.1; needs a capable graphics card and some technical setup.

Go to FantasyTalking’s website

At a glance

Open or closed: Fully open
Weights and code downloadable under a standard open licence
Code and weights under Apache 2.0; the Wan 2.1 base model is also Apache 2.0.
source · checked 12 Oct 2026 · checked on the maker's own page
Licence: Apache 2.0 (Permissive (e.g. Apache 2.0, MIT))
source · checked 12 Oct 2026 · checked on the maker's own page
Can I sell what I make?: Yes
Apache 2.0 allows commercial use; the repo makes no separate statement about outputs.
source · checked 12 Oct 2026 · checked, but the page was unclear or partial
Cost: Free option: Free to download and run yourself; free online demo on Hugging Face
Plans below.
How you use it: On your own computer, Through another service, Inside another tool
Weights on Hugging Face and ModelScope; Gradio demo Space on Hugging Face; merged into the ComfyUI Wan wrapper.
source · checked 12 Oct 2026 · checked on the maker's own page
Skill needed: Technical
Our judgement: Python install and command line, or ComfyUI.
source · checked 12 Oct 2026 · checked, but the page was unclear or partial
Made by: Fantasy-AMAP research team
Published by the Fantasy-AMAP GitHub organisation; the contact address suggests Alibaba (AMAP) but no company or country is stated.
source · checked 12 Oct 2026 · checked, but the page was unclear or partial
Status: Available
Code and weights released 28 Apr 2025; last listed update July 2025 (user guide). Research release, no ongoing roadmap stated.
source · checked 12 Oct 2026 · checked, but the page was unclear or partial

What you can make with it

A talking presenter, a talking version of a photo (main use)
A video, lips that match new audio

image to videoaudio to video

Inputs: one portrait image, an audio file and an optional text prompt to steer behaviour.

source · checked 12 Oct 2026 · checked on the maker's own page

Plans and prices

No paid plans listed.

source · checked 12 Oct 2026 · checked on the maker's own page

Rights and safety

Real faces and voices: Copies a real face (talking video from any portrait photo), no consent step
Animates whatever face you supply; no consent step in the open code.
source · checked 12 Oct 2026 · checked, but the page was unclear or partial
Limits: note: Benchmarks use 512x512, 81 frames (about 3-5 s depending on frame rate); maximum length not stated.
source · checked 12 Oct 2026 · inferred or from a secondary source; needs review

Models it uses

Wan 2.1

Where you can run it

Hugging Face, ComfyUI

Other options for “A talking presenter, a talking version of a photo”

Akool, D-ID, DeeVid AI, DomoAI, Dreamina, Dzine, EchoMimic, Elai.io, fal, Hallo2, Hedra, HeyGen, HunyuanVideo-Avatar, InfiniteTalk, Jogg AI, LivePortrait, Magic Hour, MuseTalk, OmniHuman, SadTalker, VEED, Vidnoz AI, Vozo, WaveSpeedAI

About this entry

Facts last checked 12 Oct 2026 against the maker’s own pages (each fact links to its source). Entry status: draft. Plain-English summary written by Plainly.