VEO 3.1 Is Out! Here's How to Make Realistic AI Avatars

I just created 20 ultra realistic AI avatars in 30 minutes. And no, I didn't use just Hey Jen. Watch what happens when you combine these three tools that nobody's talking about. Look, hey Jen is incredible. I've used it for months. But here's the thing. You hit a wall fast. Credit limits, expensive subscriptions, and the worst part, limited customization. You want to create different versions of your avatar for ads, different outfits, different backgrounds, maybe even prank your friends, and suddenly you're stuck paying for credit or locked into a plan that doesn't scale. Single tool thing can traps you. You're working inside one ecosystem, playing by their rules, hitting their pay walls. And if you're running a business or creating content at scale, that's a problem. You need infinite variations without infinite costs. So, I spent weeks figuring out a workaround. And what I discovered is that when you combine three specific tools in the right order, you unlock something Han can't give you alone. Unlimited hyperrealistic avatars that you control completely. Here's the

trinity. First, Nano Banana, Google's secret weapon that the internet is sleeping on. This thing creates character consistency like nothing else. You can transform your avatar into different outfits, locations, expressions, all while keeping the same face, same identity. Second chat GPT not just for writing anymore. I use it to engineer perfect prompts for nano banana and generate natural sounding scripts for Hey Jan that don't sound like a robot reading a manual. Third hen. This is where we bring everything to life. But instead of starting inside hey Jen and hitting their limits, we're feeding it pre-made customized avatars. We're bypassing the payw wall before we even walk through the door. The order matters. Nano Banana first for character creation variation. Chat GBT second for prompt alchemy and script polish. Hey genen third for the final voice and animation magic. Let me show you exactly how this works. Let's start with Nano

Banana. This is Google's Gemini 2. 5 flash image model and it's rated as one of the top image editing models in the world right now. What makes it insane is character consistency. You give it one reference image, describe what you want to change, and it keeps the person's identity locked while transforming everything out. And if you're wondering how to get this kind of reach for your product, most brands waste months and thousands with creators who either ghost project or deliver videos nobody actually watches. That's not how we work. In 10 business days, you get conversion focused content. Whether it's a full review, 10 shorts, or a PPC trailer that's ready to run, I reach over 5 million AI and software people, the ones who test tools, build products, and actually buy. 70 plus brands work with us. Recent campaigns are pulling median return and ad spend around 2. 8x. So, if you're tired of watching me promote your competitors, click the link below, answer 10 quick questions, and let's book your 10day sprint. Let me show you one more transformation. First, here's the process. Go to Google AI Studio. It's free. And access Nano Banana through the Gemini API. Upload a base image of the avatar you want to use. Could be a photo of yourself, a stock image, or even an AI generated face from MidJourney or Chat GPT. Doesn't matter. Nano Banana works with anything. Now, watch this. I'll upload my base avatar. Clean headsh shot. Neutral background. I type a simple prompt. Change the outfit to a black leather jacket. Urban street background. Confident expression. Hit generate. 15 seconds later, same face, completely different vibe. Let's do another professional business suit, corporate office setting, friendly smile. Generate. Done. Same person, new context. One more. Casual hoodie, coffee shop background, relaxed and approachable. You see the pattern? The face stays locked. The identity is consistent. But I can create infinite variations for different use cases. Ads, thumbnails, social posts, video avatars. This is the killer feature. Other tools like midjourney or chat GPT struggle with face consistency. You generate one image, try to generate another with the same face and it's slightly off. Different lighting, different angle, different bone structure. Nano Banana solves that. Pro tip, save every variation. You're going to feed these into Hen later. And having a library of pre-made avatar images means you never run out of content. I have 50 different versions of my avatar ready to go. Different outfits, different moods, different settings, all created in under an hour using Nano Banana. Now, speaking of AI generated content, here's something I've been testing lately that's honestly pretty wild. We're all creating stuff with AI now, right? Avatar scripts, images. But here's the problem. How do you know what's AI and what's real anymore? That's where Scippace AI detector comes in. And before you say another AI detector, hold on. This one's different and the data backs it up. Cispace just dropped benchmark and study where they tested their detector against GPT0 GPT Grammarly and Quailbot. They ran 4,000 samples, research papers, essays, news articles, summaries. And here's what stood out. Sispace hit 96. 2% F1 score on research papers. That's absurdly accurate. Why? Because it's not a generalpurpose tool. is trained specifically on academic content, citation styles, technical terminology, the structure of scholarly arguments. So if you are a student, researcher, or just someone who needs to verify if a document was AI written, this thing actually works. I tested it with a few essays I generated using chat GBT, then mixed in some human written sections. Cyspace caught the AI parts with precision, while other detectors gave me false positives left and right. And here's the bonus. It's not just about detection. Sci-pace has this feature called Scipace Agent, which automates research tasks, and they have lately released an update, agents gallery, where you can use ready-made agents based on your tasks. With SciEace, you'll get access to a suite of AI tools. If you're dealing with any kind of scholarly work, academic writing, or just want to verify authenticity, check out Sidespace AI detector. I will drop the link and discount codes for you in the description. Test it out with your own content. Worth the five minutes. All right, back to our unlimited avatar workflow. Let's move to step two. Now

let's talk about Chat GBT. Most people use it for writing blog posts or answering questions, but if you're smart, you use it as your creative co-pilot for AI workflows. Here's what I do. I open chat GPT and I say, "Generate five prompt variations for Nano Banana to create a professional business avatar in different settings. " Chat GPT spits out five detailed prompts. Boardroom, Zoom call, background, conference stage, office desk, networking event. I copy paste those directly into Nano Banana. Instant variety. But it gets better. I also use Chat GPT to write natural sound and voice scripts for Hey Jen. Here's the problem with most AI avatar videos. They sound robotic, stiff, overly formal. That's because people write like they're drafting an email, not like they're talking to a camera. So, I prompt Chad GBT like this. Write a 30-second script for a video ad promoting a productivity app. Casual tone, first person, conversational, like you're talking to a friend. Include a hook in the first 5 seconds. Chad GBT gives me something natural. I tweak it slightly and now I have a script that sounds human. I want to talk about something real quick because this workflow combining nano banana chat GBT, hey Jen, this took me weeks to figure out and honestly, it's just one of dozens of AI combos I've been testing and breaking down every single week. That's exactly why I built AM Master Pro. The big idea, you get one hub where everything you need, training, tools, updates, community lives in one place. I will drop the link in the description. Now, let me show you how to engineer perfect prompts for this workflow. I also use chat GPT to generate prompt templates that I can reuse. For example, avatar in location, wear and outfit, express an emotion, lighting style. I save that template, swap out the variables, and I've got 20 new prompts in 2 minutes. This is the prompt alchemy part. You're not just generating one-off ideas. You're building a system, a repeatable framework that speeds up your workflow and keeps your avatars consistent, creative, and on brand. All

right. Now, we bring it all together in Hey Gen. But here's the key difference from how most people use Hey Gen. We're not starting from scratch inside their platform. We're importing the custom avatars we already created with Nano Banana. All right, this is where it gets wild because hey Jen just integrated Google's PO3. 1 model, literally released yesterday, and the difference is insane. V3. 1 brings three massive upgrades to Hey Gen avatar. First, enhanced realism with true-to-life textures. Your avatars no longer look AI generated. The skin, texture, lighting, shadows, it all looks natural. We're talking broadcast quality output at 1080p. Second, native audio generation with perfect synchronization. V3. 1 generates audio and video together. So, lips sync is flawless. No more robotic mouth movements or weird delays. It understands natural conversation patterns, sound effects, ambient noise, the whole package. Third, superior character consistency across multiple scenes. This is huge. You can generate multiple videos with the same avatar. And VO3. 1 maintains visual continuity. Same face, same expression pattern, same style. It's like your avatar has muscle memory. Here's how it works. Go to Hey Gen. Click create instant avatar. Upload your nano banana character image. Hey Gen. Process it using VO3. 1 image to video capabilities. Now, the magic V3. 1 supports both horizontal 16-9 and vertical 916 aspect ratios. So, you can create one avatar and use it for YouTube, Tik Tok, Instagram reels, YouTube shorts, all optimized formats. Type your script or paste the one Chad GBT generated earlier. Select your voice. Hey, Jen. Voice library is solid. But here's a pro tip. If you have a custom voice clone from 11 Labs, you can upload it here for even more personalization. Look at this. On the left, old hijen without VO3. 1. On the right, same character with VO3. 1 enabled. The texture quality, the lighting, the natural motion. It's night and day. And because VO3. 1 generates up to 60 seconds of continuous footage, you can create longer, more dynamic videos without stitching clips together. The model handles scene extensions automatically, maintaining visual and audio consistency throughout. Here's three different videos I generated with the same nano banana avatar powered by VO3. 1. Notice how the character stays consistent. Same facial structure, same style, same vibe. That's V3. 1's. Improve prompt adherence and character consistency at work. This is what makes the workflow unlimited. And because we pre-generated the avatars with Nano Banana, I'm bypassing Hen's biggest bottleneck, credit costs for avatar creation. I'm spending credits only on video rendering, which is way cheaper.

All right, look. I've walked you through the perfect workflow, but let me be real with you for a second. This is not going to work perfectly the first time. It never does. So, let me save you about 3 hours of frustration by showing you the five issues I ran into when I was figuring this out and exactly how to solve them. Problem number one, Nano Banana just gave you garbage. So, you type in your perfect prompt. photorealistic woman, brown hair, blue eyes, professional, and what comes back? Some weird distorted face that looks like it went through a blender. I've been there. Here's what's happening. Nano Banana gets confused when your prompt has too many details or when those details contradict each other. Like, if you ask for realistic but also artistic style in the same prompt, it's going to freak out. The fix: strip it down. Start super simple. Young woman, neutral expression, front-facing, natural lighting. Get that working first. Then layer in details one at a time. And if it's still giving you trouble, use a reference image. Upload a photo that's close to what you want, then ask Nano Banana to match that vibe. Works way better. Problem number two. Chad GPT wrote a prompt that doesn't work. This one got me so many times. You feed Chad GPT your avatar description. It spits out this beautiful detailed prompt. You paste it into nano banana and the result is nothing like what you asked for. Here's the thing. Chat GBT doesn't actually know how Nano Banana's current prompt syntax works. It's guessing based on general AI image generation patterns. So sometimes it'll add parameters that Nano Banana ignores or use phrasing that confuses it. Solution: Use Chad GBT as a starting point, but then refine it yourself. If the first result from Nano Banana is 70% there, go back to chat GBT and say, "Keep the face and hair exactly the same, but change the background to X, that targeted correction works way better than asking it to rewrite the whole thing. " Problem number three, hey Genen won't accept your avatar. You've got your perfect avatar image from Nano Banana. You're excited. You upload it to Hey Gen and it just rejects it. File not supported or image quality too low. Super frustrating. Usually, this is a resolution or quality issue. Hey, Gen needs high-res images. Ideally, at least 1024 B124 with a clear, sharp face. If your nano banana output is too small or pixelated, hey Jen's going to say no. Quick fix, run your image through an upscaler first, works almost every time. Problem number four, your avatars look too AI. You know that uncanny valley feeling when you look at an avatar and something just feels off. Maybe the skin is too perfect or the colors are weirdly oversaturated or the eyes look a little dead. This happens because Nano Banana, like most AI image generators, defaults to this hyper polished aesthetic that doesn't feel human. The fix is in your prompt. Add phrases like photorealistic natural imperfections, film grain, slide, skin texture. That tells nano banana to dial back the perfection and make it look more real. Also avoid words like flawless or perfect. That actually makes it worse. Problem number five, lip sync and hijen is off. You generate your Avatar video and the lips are just not matching the audio. Maybe there's a delay or the mouth movements look robotic. This one's usually on the audio side. If your script is too fast or if the audio file has background noise, Hun's lips sync engine struggles. What works? Keep your script shorter, like 30 seconds max per clip. Speak at a normal pace in your audio recording and make sure your audio file is clean. No echo, no static. If it's still off, try re-uploading the audio in higher quality. Wave instead of MP3 helps. Let's hit some advanced tricks. First

blending images. Nano Banana lets you reference multiple images at once. If you want your avatar to have a specific hairstyle from one image and an outfit from another, upload both and describe what you want to merge. It'll combine them intelligently. Second, style transfers. Want your avatar in an anime style? Cyberpunk aesthetic watercolor painting look? Nano banana handles style transformations while keeping the face consistent. Just add in the style of X to your prompt. Third, create an avatar families. If you're building a brand with multiple characters, think explainer videos with different personas. Use Nano Banana to create a cast. Keep the prompts similar for visual cohesion, but vary the details. Now you've got a team of avatars that look like they belong together. Fourth, voice cloning in Hen. If you want your avatars to sound exactly like you, record 2 minutes of yourself speaking, upload it to Hunen's voice cloning feature, and now every avatar can use your real voice. This adds a layer of authenticity that default AI voices can't match. Now, common mistakes to avoid. First, don't overprompt Nano Banana. Keep your descriptions clear and concise. If you write a paragraph of instructions, the model gets confused and the output gets muddy. Stick to one, two sentences max. Second, don't skip script editing and chat GBT. Even though chat GBT writes well, you need to add your personal touch. Read the script out loud. Tweak the phrasing. Make it sound like you. That's what makes the avatar believable. Third, don't upload lowresolution images to Hey Gen. Nano Banana outputs high quality images. Make sure you're downloading the full resolution before uploading. Blurry avatars kill credibility. Look at this.

10 different avatars, all generated in the last hour. Different outfits, different settings, different vibes. I can use these for ads, content, client work, or just mess with my friends. And you can do the same starting today. And if you want the full prompt templates, plus weekly workflows like this, check out AMS or Pro link in the description. This workflow alone took me weeks. You got it in 12 minutes. Imagine what else is inside. Thanks for watching. Now go create something insane with your new unlimited avatar setup. See you in the next one.