Talking video
A face, a voice sample and a script — and the face says it.
Before anything else
Never generate a real person's face, voice or likeness without their explicit permission. This tool clones a real voice from your sample and animates a real face from your picture. Use it on yourself, on somebody who has told you yes, or on a person who does not exist. Uploading a face or a voice you do not have permission to use breaks the rules here and, in many countries, the law.
Nothing sexual involving a minor, ever. No sexual imagery of a real person who has not agreed to it. No clip built to make somebody appear to say something they did not. Those end the account.
What this is
A face, a few seconds of someone's voice, and a line of script — and the face says the line in that voice.
Two models in one job. F5-TTS clones the sample and reads your script; it is zero-shot, so a few seconds of clean speech is the entire model of the voice. Then wav2vec2 turns that read into conditioning and Wan 2.2 S2V animates the face against it. The clip is as long as the read, so it never stops mid-word.
Leave the script empty and the uploaded audio is used exactly as recorded.
Running it yourself
This one needs a custom worker, for a reason worth knowing: the RunPod ComfyUI worker accepts image uploads only, so a WAV cannot arrive as a file. Our image adds four small nodes — audio in from base64, F5-TTS, frames-from-audio, audio out — and a corrected extra_model_paths.yaml, because the stock one does not map audio_encoders at all and wav2vec2 on a volume is invisible without it.
It is in the repo at gpu/comfyui-talk-worker/, about a 2 GB build because the weights stay on the volume. Models: wan2.2_s2v_14B_fp8_scaled (16.4 GB), wav2vec2_large_english_fp16 (0.6 GB), and the t2v 4-step LoRA — not the i2v pair, because S2V is a single expert. Sign up at RunPod to host it.
Not switched on here yet — the image has to be built and pushed first.
You give it
- A picture of the face
- A few seconds of clean speech as WAV — what the voice SOUNDS like, not what it says
- The line you want spoken
You get back
- An MP4 with the audio muxed in
- The generated read as its own WAV, so you can reuse it
Leave the script empty and the uploaded audio is used exactly as recorded — useful when you already have the take you want.