How do you create an AI avatar of yourself for video content?

Create an AI avatar of yourself by uploading multi-angle photos of your face, generating a character reference sheet that locks your appearance, saving that identity to persistent project context, anchoring your voice with a face-referenced talking-head clip in Seedance 2.0, then generating scripted videos that all pull from that one locked identity.

Start by capturing your reference material: several clear photos of yourself — front, side, profile, plus tight face close-ups — in even lighting, with nothing in your hands. Close-up panels matter because small details (glasses, stubble, accessories) drift across shots if the model only ever sees you wide; documented character workflows explicitly include face and mid-angle close-ups for exactly this reason.

invideo is an agentic video creation tool with the current image and video models available inside it, so the whole pipeline below runs in one place.

1. Generate a character sheet from your photos. Upload your reference photos to the invideo agent and ask for a multi-angle character reference sheet. Documented productions lock a face with a 12-angle master sheet or a 4-angle turnaround plus close-ups, and a full sheet generates in under 3 minutes. For photorealism, Recraft renders faces with pores, lines, and stubble — the imperfections that make an AI face read as a real person — while Nano Banana and GPT-Image-2 handle the multi-angle sheet assembly; the invideo agent routes each step to the right model so you don't choose.

2. Save your identity to project context under a named key. Store the approved sheet as a named constant (e.g., HOST_ME) in the invideo agent's context tab. This is what makes it an avatar rather than a one-off image: every future generation pulls from the same locked identity, instead of you re-describing yourself each session — creators report losing about 20 minutes per session in tools without persistent memory. One 70-second production held the same character's appearance across every scene using only character sheets and agent context, no LoRA fine-tuning; expect roughly 5 generation attempts (~$9.78 in one documented case) to fully lock an identity you're happy with.

3. Anchor the voice to your face. Generate a talking-head clip in Seedance 2.0 with your face image attached as a reference — the model matches the voice to the face, producing a more consistent voice signature across generations than disembodied voiceover. Generate several voice samples, specify age, accent, and emotional tone, and select one; the invideo agent presents voice options per character for exactly this selection step.

4. Generate your videos. Give the invideo agent your script and let it dispatch talking-head shots of your locked avatar. For dialogue, provide the line plus scene context and skip per-second timestamps — timestamped dialogue prompts cause Seedance 2.0 to hallucinate filler lines and waste credits. For multi-clip pieces, Seedance 2.0 reference-to-video carries your character context from clip to clip, which is why it holds appearance better than start/end-frame methods.

5. Reuse the avatar for everything after. The locked identity compounds: one documented workflow defined a virtual instructor once and held visual and voice consistency across an entire course, collapsing a 3–6 month production timeline to 14 days. If generated faces come back over-sharp, a light blur-and-grain pass in post moves them closer to real footage.

Watch some of these to see what works for you:

it's always better for my experience to put a face so that see dance will recognize that face and try to match it as close as possible to something similar as far as voices go.

— a filmmaker documenting the face-anchor voice workflow