Deepfake avatar streaming uses your live camera to perform as a reference character. Prepare your OBS scenes and microphone, then use the studio below to start the AI picture and capture it for your broadcast. OBS handles the layout, sound and connection to your streaming service.
Sign up to try it with free credits.
Checking browser capabilities…
Checking live preview support…
YOUR BROADCAST / A CHARACTER-LED Q&A
Rehearse a creator Q&A about your next drawing: a small robot. Keep the viewer's question beside your character while you explain the shapes, announce a break, then return to discuss colors. Build the question and break scenes and check your microphone before starting AI. Add the moving character once that setup is ready.
Use Live Face Swap to prepare your character with your own camera. For deepfake avatar streaming, carry that picture into a broadcast scene you can inspect before viewers arrive.
Question + break scene + microphone → OBS
Arrange the text and check your voice with AI still off.
Camera → browser AI picture → OBS capture
Start AI when ready, crop the output and check it beside the question.
OBS → your streaming service
Connect your destination in OBS after the local test passes.
This workflow uses desktop window capture. It does not provide an OBS plugin, a dedicated Browser Source URL or a virtual camera. Capture support depends on your browser, operating system and OBS setup; check the actual OBS picture before starting a broadcast.
Notice how much space the comments and interface take around the character. For your own Q&A, reserve one area for the question and another for the host. A face that fills the browser preview may need a smaller, deliberate place in the broadcast. Judge the complete OBS composition, including the text, before you settle on a crop.
If you are still choosing the host's appearance, use the reference and performance checks on the AI Webcam Avatar page. Here, work with one chosen character and concentrate on the broadcast.
SETUP / KEEP AI OFF FOR NOW
Keep AI off while you build these two OBS scenes. If a session is already running, choose End in the studio before arranging the layout. Use the sample drawing question below, or replace it with a question you can answer. Leave room on the right for the character you will add later.
In OBS, use the + below Scenes to create a scene named Q&A. In Sources, add a Text source: Text (GDI+) on Windows or Text (FreeType 2) on macOS. Enter "What will you draw next?" and place it on the left. Choose a text color that is readable against the background and leave the right side empty for the captured character.
Check before continuing. Read the full question at the viewing size you expect your audience to use. If it is too small, enlarge the text or shorten your own question before adding the character. This text is placed in OBS; it is not generated by the studio.
Use Scenes + again to create a scene named Break. Add a new Text source saying Back shortly, with a readable color against the background. Leave browser, camera and display capture out of this scene. Select Q&A, then Break, and check that the output changes from the question to the break message.
Check before continuing. If Studio Mode is enabled, selecting a scene changes Preview first; use Transition to send it to Program. Check the actual output, not only the selected scene name. Test the microphone separately below: changing the picture does not necessarily mute a global audio source.
AUDIO / YOUR OWN VOICE
Test your voice in OBS while AI is still off. Record the sample question over the text-only scene and listen back. The AI picture on this page is muted; the studio's Record microphone setting controls clips saved here, not the OBS mixer. Add the moving character after the voice sounds clear.
Select your microphone in OBS Settings → Audio, or use a scene-specific Audio Input Capture source in Q&A. Avoid capturing the same microphone through both paths. With AI still off, start an OBS test recording and say "What will you draw next?". Stop and listen with headphones; a moving meter alone does not tell you whether the voice is clear.
Keep playback of recorded clips off while checking the microphone. If the voice echoes, inspect duplicate microphone sources, monitoring and desktop or application audio. Screen capture may also include application sound on some systems. Keep only the audio routes you intend to send to the audience.
PICTURE / THE RUNNING BROWSER WINDOW
With the scenes and microphone ready, prepare the browser capture for deepfake avatar streaming. Select the existing studio window before starting AI, then check the moving output once it starts. A URL pasted into an OBS Browser Source opens a separate browser context; it does not connect to this signed-in browser session.
Return to the Q&A scene in OBS. Add a capture source using the option for your operating system below and select the browser window containing this studio. Keep the studio tab selected and the window open at a fixed size. Make an initial crop around the output area; you will check it again after your own AI picture starts.
Choose the open browser window in the source properties. Verify that the captured image keeps moving while you operate OBS.
On supported macOS versions, choose Window Capture as the method and select the browser window. Allow the screen-recording permission requested by the system.
Check before continuing. Confirm that OBS captures the correct window, and resolve any empty capture or permission prompt now. The idle demonstration only helps locate the output area; it does not verify your own camera result. Do not use your physical webcam as the avatar source, because that would show the original picture.
Choose Face Swap, select a character or upload a reference, open the camera and sign in. Turn off Auto record if you only want the OBS rehearsal recording. Check the selected model's credit rate and your balance, then start AI. Wait for your own moving picture and make a small head movement; it should appear in both the browser and OBS.
Check before continuing. Deepfake avatar streaming uses credits while AI runs, including time spent adjusting OBS or showing the Break scene. If more setup work is needed, choose End in the studio and finish that work before restarting. Stopping an OBS recording alone does not stop AI usage.
View plans and credits before starting your session.
Select the captured source and use Transform → Edit Transform to crop away the browser chrome, camera input and studio controls. Resize the remaining picture in the scene, leaving the question readable beside it. Recheck the crop after scrolling, resizing the browser or changing its zoom.
Check before continuing. Inspect all four edges while the source is active. Keep the whole question "What will you draw next?" readable beside the host, with room for hair and shoulders and no camera-input panel or result controls in the crop. If that layout does not fit, adjust the source size before recording the rehearsal.
REHEARSAL / OPEN, BREAK, RETURN
Rehearse deepfake avatar streaming in an OBS recording before you broadcast. Keep the drawing question on screen for the opening and return. Speak the sample lines below, or adapt them to your own topic. The OBS recording should show the question, moving host, voice and scene transitions together; a studio-only clip cannot check that complete view.
Sample rehearsal script for a drawing Q&A. Speak these lines yourself; they are not generated speech or a recorded customer example.
Start an OBS recording with Q&A in the output. Read the question, glance briefly toward it, then face the camera as you give the first part of the answer.
Welcome! Today's question is: what will you draw next? Next time, I'll draw a small robot, starting with its head and simple shapes.
Check the recording. The whole question should stay readable while the host moves. Listen for the complete opening sentence and compare its timing with the moving face. Deepfake avatar streaming processes the picture separately from your microphone; judge a full sentence, not a paused mouth.
Say the transition line while Q&A is still in the output. Then switch the output to Break, using Transition if Studio Mode is enabled. Mute the microphone separately if it remains active and you want a silent break.
I'll take a short break, then explain how we'll choose the colors.
Check the recording. The recording should contain the end of the sentence before Back shortly appears. Confirm that the browser picture is absent and the break is silent as intended. This transition is a manual OBS action; the studio does not trigger it.
Return Q&A to the output, restore the microphone and check that the character is moving before speaking. Finish the same answer, then stop the OBS recording and watch all three parts together.
I'm back. We'll pick two main colors together and add one bright detail. Bring a color suggestion to the next stream.
Check the recording. Listen for I'm back without a clipped first word. The original drawing question should still be visible and the returning host should not cover it. If voice timing changes between the opening and return, resolve it before going live rather than assuming one fixed delay will fix both.
For deepfake avatar streaming, keep the break scene available during reference changes, account dialogs or connection problems. After the local rehearsal passes, connect your streaming service in OBS and check its preview or private test option before a public broadcast.
A DECISION / FIND WHERE IT CHANGED
When deepfake avatar streaming looks wrong, compare the browser output with the OBS recording. A moving avatar in the browser and a frozen picture in OBS point to different next steps than a stopped AI session. Find the first place where the picture or sound becomes wrong before changing the reference.
Check the capture source and selected window.
Keep the break scene visible. Re-select the correct window and check screen-capture permissions where applicable. A different character reference will not repair a capture source that is not receiving the browser picture.
Change the broadcast composition.
Move or shrink the captured area while keeping the host's face readable. The surrounding room is part of the AI picture; this workflow does not produce a transparent character layer. Recheck captions and source order in OBS.
Check the OBS audio path.
Confirm the selected microphone, mute state and duplicate sources. Compare the saved test with its moving face. If timing is inconsistent, resolve it before using a close-up talking character on air.
Switch to the break scene before changing the studio.
Check the session message and balance. End the session if needed, adjust the setup and test again off air. The break scene is something you create and select in OBS; this page does not switch it for you.
FINISH / TWO PLACES TO STOP
Deepfake avatar streaming uses separate AI and broadcast sessions. Stop each one explicitly when the Q&A ends, then save the recordings you want to keep.
Show the closing or break scene. Stop Streaming, and stop any OBS recording separately. Check that the broadcast has ended at your destination.
Return to the studio and choose End. Stopping OBS, hiding its capture source or switching scenes does not end the AI session. Check the studio status rather than assuming the browser stopped generating.
Save any studio clips before leaving the tab and locate your OBS recording separately. For the next session, reuse the scene layout but recheck the capture window, crop and microphone.
Keep the setup when the character, question and voice work together in the OBS test, and the break scene behaves as expected. If the browser size, display scale, audio device or scene layout changes, repeat that short check. A successful rehearsal is useful evidence for that setup, not a guarantee of an uninterrupted broadcast.
Prepare your character previewIf you only need a saved performance, the Webcam Video Recorder with Filter covers recording and trimming directly in the studio.
BEFORE YOU GO LIVE
Your live camera drives an AI character based on a reference image. This page produces the moving browser picture. The workflow described here uses desktop window capture in OBS to prepare that picture for a broadcast, with the microphone and streaming destination configured separately.
That does not connect OBS to the AI session already running in your browser. This page has no dedicated broadcast URL. The instructions use capture of the existing browser window, followed by cropping to the AI output. Test that path on your computer before broadcasting.
Enable it before starting AI if you want your voice in a clip recorded by the studio. For the OBS broadcast, select and test your microphone inside OBS. These are separate recording and audio paths, and the character effect does not convert your voice.
The OBS window-capture instructions on this page are for a desktop setup. A phone's browser preview or saved clip does not establish a mobile broadcasting connection. Use a desktop for this workflow, and verify its camera and capture behavior before going live.
The studio recorder, AI session and OBS broadcast are separate. A local clip limit does not define OBS's recording or streaming duration. The AI picture still depends on available credits, session limits and a working connection; plan a break scene instead of assuming it can run indefinitely.
Local camera setup is available before AI starts. Eligible new accounts receive trial credits; live generation uses available credits at the selected model's displayed rate. Time spent arranging OBS with AI running also uses credits. Check the balance and choose End when you finish generating.
Choosing an avatar focuses on the character's appearance and performance. This page focuses on the next task: getting the moving picture and your voice into a broadcast scene, testing the audience's view and ending both the broadcast and AI session deliberately.