Here are 3 processes, depending on the model you choose:
Process 1: For audio-driven models and the OmniHuman model
Process 2: For the Arcads 1.0 model
Process 3: Clone option from a video
PROCESS 1 : Using Audio driven OR Omnihuman models :
Step 1 — Generate the actor image (choose a model)
Before you can make a “talking actor,” you need a strong single portrait image (clean face, good lighting, sharp details). In the image generator, pick one of these:
Nano Banana Pro
Best when you want very accurate, “smart” images (good real-world understanding), strong editing, and reliable text rendering in images
Seedream 4.5
Best when you need cinematic aesthetics, strong spatial reasoning, and especially consistent characters across multiple generations (great if you’re iterating a character look and want it to stay stable).
GPT Image 1.5
Best when you want very strong instruction-following and a tight prompt-to-image match, plus a solid edit workflow (generate + transform + edit). It’s a good “default” choice when you want the model to do exactly what you described.
Quick pick
Want “most controllable prompt fidelity”? → GPT Image 1.5
Want “best cinematic look + consistency across variations”? → Seedream 4.5
Want “smart editing + strong text/precision + all-around reliability”? → Nano Banana Pro
How to prompt the perfect actor image
Use a prompt that locks down identity + camera framing.
Prompt template
Subject: age range, ethnicity (optional), hairstyle, wardrobe
Framing: “front-facing medium close-up” / “head-and-shoulders”
Lighting: “soft key light, natural skin texture”
Background: “plain studio background” (keeps attention on the face)
Style: “photorealistic” (recommended for talking actors)
Example prompt
"Photorealistic head-and-shoulders portrait of a confident presenter, front-facing, neutral background, soft studio lighting, sharp focus on eyes, natural skin texture, 35mm lens look, minimal shadows, high detail"
Result
GPT Image ->
Nanobanana image ->
Seedream image ->
Tip: avoid heavy motion blur, extreme angles, hands covering face, or busy backgrounds—clean facial visibility makes the talking result look more believable.
Step 2 — Turn the image into a talking actor
Click the image you just generated.
Click Transform → Talking actor
Write your script (what the actor will say)
Pick a voice
Click Generate
Result ->
PROCESS 2 : USING ARCADS 1.0 MODEL
Click on Talking Actor section, then "Add actors"
Then click on Create Actor
Upload an image that you either generated with Arcads, or use one of yours! You can also prompt it straight from the platform!
You can add guidance on how the actor should speak and behave directly in the prompt.
For example:
“Make the actor talk with excitement and energy, looking directly at the camera, with a friendly and engaging tone.”
Once your instructions are ready, simply click Turn into Talking Actor to generate the video.
Then pick a voice of your choice, and click Pick Voice.
It will then generate a preview for you.
The actor is now ready to be used 🎉 Ensure the actor has completed the training process before it appears in the 'My Actors' folder. It usually takes 2-4 hours.
You can find it under Talking Actor → My Actors. Custom actors are private to your workspace and cannot be accessed by other users, ensuring privacy and security. If you encounter issues locating your actor, ensure the training process is complete or check the Custom Actors section of your workspace.
PROCESS 3: Using CLONE Option
This allows you to use a video as a reference for creating a custom actor. This is recommended if you want the Custom Actor to move more fluidly and with higher fidelity.
Strict requirements tied to the reference video to achieve best results:
Accepted file formats: .mp4 / .mov
Max file size is: 100MB
Minimum video reference length: 2 minutes
Speaking:
Speak continuously throughout the video
Say anything you like (for example, talk about your company or your product)
Full, uninterrupted speech is required for accurate voice and lip-sync training
Framing:
Face must be fully visible at all times
The mouth must never be covered
Keep a natural, stable framing
Avoid abrupt head turns or jerky movements
Expression and Tone:
Speak naturally
Use the tone you want your avatar to replicate (professional, casual, enthusiastic, etc.)