DLQ · REFERENCE TO RESULT

Create an avatar video

Photo + voice → Create video → Preview & download.

Saved projects

Select a project marked Video saved to watch or download its result. Inputs only means no output has been saved yet.

Start another test from current inputs

Experiments · saved work

Preview plays the saved output here. It does not change your inputs or use the video as a reference.

Bring your avatar to life

మీ photo, మాట్లాడాల్సిన voice ఇవ్వండి. Reference video ఉంటే జోడించండి. తర్వాత Create video.

1 · Choose your photo

or upload your own

PNG, JPG or WebP · up to 15 MB

No photo selected yet.

Choose a photo and add audio to continue.

Inputs are saved on this browser. Continuing does not start a render.

ADMIN · SAME PROJECT, SAME REFERENCE

Reference analysis

Measure the uploaded reference without leaving this project. Measurements are saved with this project's reference. Replacing the reference clears its analysis.

Select or upload a reference in Inputs first.

How this affects the output

Local SadTalker uses the photo and voice, and extracts head-pose and blink coefficients from an optional reference clip. The browser measurements below remain review diagnostics; the renderer uses its own compatible coefficients. Body/hand motion and non-human characters are not supported. Quality needs visual review.

Admin settings · engine & review targets

Local engine settings

SadTalker CPU · 256 px face/head preview · optional reference head pose and blinking (up to 25 MB).

Measured testing: a 1-second clip took about 7 minutes. Voice and original full-photo framing are not changed automatically.

External engine handoff

Technical engine setup →

Actual render measurements

Your video

Your saved photo and voice will be used here. You do not need to upload them again.

Available now: a short talking-face preview from one photo and your audio. An optional reference guides head movement and blinking. Body and hand movements are not transferred.

Ready when your photo and voice are attached.

Awaiting generated result.

Already have a video? Attach it for review

The upload above accepts an externally completed result; it does not generate a video.

4. Compare reference with new output

Open both videos together, run facial diagnostics, then review identity, lips, eyes, movement and detail. Different wording is not a same-speech lip-sync comparison.

In comparison, choose “Analyze faces”, review the five criteria, then “Save test record”. Readings and your decision return to this experiment.

5. Experiment statistics

Upload inputs to begin.

Face diagnostics are proxies, not certified identity or lip-sync accuracy. “Human reviewed” means a review was saved, not that the result passed.

6. Deliver to your website / client website

For a completed video: upload the approved MP4 to your hosting/storage, enter its public HTTPS URL, then copy the embed. Browser-local uploads cannot play on client websites.

Live AI assistant delivery · architecture, not connected

Client website widget → authenticated session → client knowledge/AI → speech and voice → real-time avatar stream. Required: server-side API keys, tenant isolation, permitted domains, short-lived sessions, usage limits and billing records. A generated MP4 is not an interactive assistant.

Client onboarding, billing and production hosting are not implemented in this local testing dashboard.

Ready. Start at step 1.