One framework for generation (R2AV), editing (RV2AV), and animation (RA2V). Built by Tsinghua University and ByteDance with Syn-RoPE identity binding—delivering lip-synced, identity-consistent videos for portraits, dubbing, and multi-person scenes.
Generate with DreamID Omni
Upload an image and an audio clip, then add a prompt. DreamID Omni—unified human audio-video AI—generates identity-consistent talking video driven by voice and timing (R2AV).
Image *
Required. Use an image with a human subject, face or character.
Audio*
Required. Audio under 35 seconds. We’ll use its length for the video duration.
Prompt *
Required. Max 2000 characters. For scene, movements, camera, etc. Supports Chinese, English, Japanese, Korean, Spanish, Indonesian.
Requests are tied to your DreamID Omni account and will appear in your profile history once finished.
Preview
Demo clip · DreamID OmniEvery layer is engineered for semantic consistency and audio-visual alignment.
A holistic backbone orchestrating R2AV (Generation), RV2AV (Editing), and RA2V (Animation). Eliminates the friction of stitching incompatible models.
// TECH SPECS:Shared Latent Space ensures character identity, motion trajectory, and audio semantics are intrinsically aligned.
Proprietary rotary positional embeddings that solve referential ambiguity by rigidly binding identity tokens to specific spatial coordinates.
// TECH SPECS:Ensures Pixel-Perfect Identity Preservation, keeping faces and voices disentangled even in complex multi-subject scenes.
Next-gen Dual-stream Diffusion Transformer that performs bi-directional reasoning over audio and video signals simultaneously.
// TECH SPECS:Achieves Granular Lip-Sync, captures subtle micro-expressions, and maintains global illumination consistency.
Experience the DreamID Omni engine. Play the Source Audio to hear the raw input, then play the Generated Video to see the identity-consistent output.
"Today he receives the silver star for bravery and valor."
$Warm soft light, sub1 in black suit, white shirt. Serious, respectful tone.
"Nice work. Tell DCA, get a fire team."
$Dim industrial background. Middle-aged man, camo uniform, sweaty intense face.
"Really bad guy, someone who might be threatening girls with scissors or a knife."
$Long straight dark hair, grey shirt. Expression serious, concerned, focused.
"Increasingly powerful bursts of aggression, uh, persecution, anxiety."
$Dim room. Long blonde hair, looking down at screen. Furrowed brows, anxious.
"About you. About how you're changing."
$Outdoor blurred greenery. Man with shoulder-length hair, beige jacket. Intense gaze.
"Cash. He was supposed to come back the next day for his shirt. But get this..."
$Bright indoor. Light blue shirt, white tank top. Casual conversational tone.
The DreamID Omni engine unifies generation, editing, and animation into a single, streamlined process.
Upload Portrait or Video
Start by uploading a single portrait image (for animation).
Audio or Motion Reference
Upload your voice track (TTS/Recording). The engine extracts semantic motion and emotion cues.
Generate & Sync
The Symmetric DiT backbone fuses the source and driver, applying Syn-RoPE to lock identity and sync lips perfectly.
Everything you need to know about the product and billing. Can’t find the answer you’re looking for? Chat to our team.
Explore the DreamID-Omni engine, test your own assets, and build the next generation of human-centric experiences.