Runway Characters | Real-Time AI Avatars Reach SIGGRAPH
Runway Characters reached the Real-Time Live! stage at SIGGRAPH 2026 with a system that turns a single reference image into an interactive conversational character. Instead of relying on a prerecorded loop or a prepared 3D model, Characters generates each response frame by frame as the conversation happens, opening a more flexible path for animated hosts, digital companions, tutors, mascots, and other live visual experiences.
Runway demonstrates live AI characters without prepared animation
Characters begins with one image, which can show a person, animal, animated figure, brand mascot, fantasy creature, or abstract subject. The system reads the visual style directly from that reference and does not require separate fine-tuning or a style-specific adaptation before the character can begin responding.
This removes several assets normally required for an interactive avatar, including a complete 3D model, rig, animation library, and collection of prerecorded reactions. At SIGGRAPH, Runway demonstrated that process live by photographing a participant and turning the image into a conversational character within seconds.
Every frame is generated while the conversation continues
Rather than repeating short video segments to preserve the appearance of a character, the underlying model produces frames sequentially and streams them as they are generated. Runway reports an effective model time of 37 milliseconds per frame, supporting expressive HD video at 24 frames per second.
The server-side turnaround from the end of a user's speech to the beginning of the character's response is approximately 1.75 seconds in Runway's reported testing. Keeping that delay low is essential because even a visually convincing avatar can make a conversation feel unnatural when its reactions arrive too late.
Long conversations test more than visual consistency
During the SIGGRAPH demonstration, attendees held conversations lasting 30 minutes or longer without the character relying on a limited collection of looping reactions. That duration places additional pressure on facial consistency, lip synchronization, head movement, response timing, and the system's ability to maintain a believable visual presence.
For character designers, the reference image therefore becomes more than a static portrait. Facial readability, silhouette, mouth visibility, framing, and recognizable visual features can all influence how well the identity survives continuous movement and an extended interactive session.
Characters can connect visual identity with useful actions
The surrounding platform lets creators assign a custom voice, personality, knowledge, and tools to a character. It can also receive information from a camera or shared screen, use approved actions during a conversation, and be added to websites or applications through Runway's developer tools.
These capabilities extend the format beyond a talking portrait. A visual character could guide a product demonstration, provide design feedback, support an educational activity, host an interactive experience, or represent a brand while drawing its answers from an attached knowledge base.
Runway is exploring worlds, memory, and emotional control
The next research direction moves beyond characters that remain inside a fixed conversational frame. Runway is working toward navigable environments where users can move through a world and interact with a character in different ways, while text-based interaction and more direct control over emotion and expressivity are also being explored.
Other areas under consideration include multiple reference images, personal memory, individual objects, faster responses, and broader accessibility. Together, these ideas suggest a shift from isolated AI avatars toward persistent characters with a stronger visual identity and a more complete sense of place.
IMPORTANT: Navigable environments, expanded memory, multiple reference images, and additional emotional controls are research directions described by Runway, not confirmed finished features with announced release dates.{alertWarning}
Daisuki's Take: What This Means for Designers
The most meaningful change is that character design can now become part of a live interaction without first passing through a traditional animation pipeline. A strong reference image may serve as the visual foundation for a character that speaks, reacts, and remains active throughout a conversation.
This also gives designers new responsibilities. A character created for repeated interaction needs an identity that remains readable in motion, a voice that matches its appearance, and behavioral boundaries that support its intended role instead of allowing the experience to drift unpredictably.
We would approach Characters as a combination of visual design, interaction design, and editorial direction. The technology can animate the reference, but creators still need to define who the character is, how it should respond, which knowledge it can use, and whether its behavior remains appropriate during longer unscripted sessions.
Sources and Recommended Links
- Bringing Runway Characters to SIGGRAPH's Biggest Stage | Runway Official Site
- Building Runway Characters: Real-Time Conversational Video Agent from a Single Image | Runway Official Site
