AI Influencer in Hebrew · The Complete Guide to Creating UGC Videos That Hold and Sell Products.

Imagine a girl holding your product, looking at the camera, and talking about it in Hebrew with a lot of emotion, as if she is a friend of yours who has to tell you about a new trend. Now imagine that this girl does not exist. She does not exist anywhere in the world. Artificial intelligence invented her through a single prompt, planted your product in her hand, and made her talk. This is exactly what we did in this bonus meetup of the AI Master Club, live, step by step. This thing only started working in Hebrew recently, so Amichai scheduled a spontaneous meetup: instead of watching this in some Reel on TikTok, learn to do it yourself, from start to finish. Do not get hung up on the word influencer. What you will learn here is a complete workflow: creating a high-level realistic character, placing a real product inside the image (even in Hebrew on the packaging), emotional talking animation, and chaining clips into a long video. You can take this for product advertising, client videos, content for your business, and endless other directions. All with Google tools and the 20 dollar Gemini subscription.

What you will know how to do at the end of the guide

• Create a high-level realistic image of an AI influencer who does not exist
• Use a Gem that receives an image and returns the exact prompt for a similar creation
• Place a real product inside the image, in the hand of the character, with a single prompt
• Animate the character into a video speaking Hebrew with emotion and body language
• Chain clips beyond the 8 second limit, into a video of a minute or more
• Edit and connect everything in CapCut, or directly inside Flow
• Configure correctly to avoid burning credits, and work with Nano Banana Pro for free
• Bypass Hebrew glitches with ElevenLabs and lipsync
• All the prompts and questions that came up in the meetup

The static image of the video: every day of work in AI

Step 1 · Creating the character with Nano Banana Pro

Prompt structure for creating a realistic character (UGC)

Photorealistic UGC selfie shot on a phone front camera. A young woman taking a casual selfie. [Subject] describe her: hair, face, expression, age. [Clothing & pose] what she wears, how her body is positioned (sitting / standing). [Environment] where she is (home kitchen, bedroom), what's visible in the background. [Style & lighting] daytime / golden hour / indoor light, strong or soft. Make it look real and slightly imperfect: natural skin texture, not-perfect lighting, a little everyday mess in the background, like a real phone photo. Aspect ratio 9:16.

Step 2 · Placing the product in the character's hand

The two-frame technique: to create movement, we prepare several images of the character in different states (a frame holding coffee, a frame holding the product, a frame returning to the phone). Later we will define a first frame and a last frame for the AI, and it will fill in everything in between as a video. Always keep the same background and the same proportion between frames, and change only what is needed (angle, body position, what is in the hand).

Step 3 · Moving to Flow and animating to video

Animation prompt: influencer speaking Hebrew with emotion

A UGC-style video. A 20-year-old woman speaks naturally in Hebrew, as if she is gossiping with her friends about an exciting new trend she just discovered. She speaks with a lot of emotion, natural facial expressions and hand gestures. She can really see the product she is talking about and is genuinely excited. At some point she brings the [product] closer to the camera to show it off. Keep her face and voice consistent. Make it feel authentic and spontaneous, like a real phone selfie video.

Step 4 · Chaining clips beyond 8 seconds

Step 5 · Connecting and editing in CapCut

The limitations, so you do not get disappointed and waste credits: it almost never turns out perfect the first time. Sometimes there are distortions in hands, jumps, or glitches. The secret is to try again and again and check which body position works best, it is trial and error. In Hebrew there are still speech glitches (in English it is almost perfect). And note: the more prompts you add on the same image in Nano Banana, the quality might drop, so open a new chat and re-enter the original image.

Saving credits · The settings you must know

AI Master Club opens the door for you: a live meetup every week, recordings, and a community. 47 NIS per month, cancel anytime.

Hebrew glitches · The professional solution

Want your own character, and not a fictional one? You can. The standard tool for this is HeyGen: you feed a video of yourself talking for a few minutes, and it clones you into a digital avatar that you can feed text and it will talk. Excellent for short segments (intro for TikTok, Reel). The tip: make quick cuts and do not always be on the face, let people see the visual of what you are talking about. Most viewers do not notice the nuances you see in front of the mirror every morning. We will get to the HeyGen meetup separately.

All the tools from the meetup

Creating and editing images. Activate the Nano Banana Pro model, create a character, and place a product. The basic subscription provides credits

The studio that animates images to video with VO 3.1, and connects clips. Enter via Google Labs with a Gemini account

Editing and connecting clips into a long video. Free, in the browser or app. Work in 9:16 aspect ratio

Creating almost perfect Hebrew speech and voice cloning, to bypass glitches. Download soundtrack and return with lipsync

Source of inspiration for characters. Take a picture and feed it to a Gem that returns the exact prompt for a similar creation

Questions that came up in the meetup

Currently VO 3.1 is still not perfect in Hebrew. The professional solution: create the speech in ElevenLabs (including voice cloning), download the soundtrack, and use a lipsync tool to synchronize it to the video. In English, by the way, it works amazingly without noticing. This is expected to improve significantly within a few months.

Flow is designed to work with Google tools: VO 3.1 for animation, and Nano Banana Pro for unlimited image generation (free). Kling is an excellent external video model, in the top tier together with VO, but without Hebrew. You can create segments with it, connect them, and add Hebrew speech via ElevenLabs and lipsync.

In terms of software, VO 3.1 and Kling are the best in the world, there is no superior scientific tool. They were simply trained on many more videos of people talking, so they are stronger there, it is a statistical matter. But it is very worthwhile to try: we saw impressive videos even of a fetus in the womb. Give a first frame and a last frame of the process, and let it fill in the middle.

Yes. The standard tool is HeyGen, which clones you into a digital avatar from a short video. Excellent for short segments (intro, Reel). HeyGen itself has already started integrating VO technology. For a long one-hour format I would not recommend it, but for short social content it passes wonderfully.

Make sure you activated the Pro model before running Nano Banana, and that you clicked the banana icon. Sometimes the model just loses its mind and refuses. The simplest solution: open a new chat and enter the same request from the beginning. Do not stubbornly butt heads on the same chat.

As long as you have a still image of the product, it works. If the product does not exist at all, you can generate a visualization for it with Nano Banana Pro. Important: the input must be a still image, not a video. If you have a video of the product, stop it on a frame, remove the background, and input the image.

And what in the next meetup

The full recording of the meetup, along with the written guide, all prompts and links, is available to club members. Every meetup is saved with an organized guide page that can be scanned in a minute and copied and pasted. Try the method, play with your products, and you will discover that this is just scratching the surface.

Frequently Asked Questions

How do you avoid glitches in Hebrew? It ruins the point.

Currently VO 3.1 is still not perfect in Hebrew. The professional solution: create the speech in ElevenLabs (including voice cloning), download the soundtrack, and use a lipsync tool to synchronize it to the video. In English, by the way, it works amazingly without noticing. This is expected to improve significantly within a few months.

Is Flow only for creating a long video, or can you work with Kling?

Flow is designed to work with Google tools: VO 3.1 for animation, and Nano Banana Pro for unlimited image generation (free). Kling is an excellent external video model, in the top tier together with VO, but without Hebrew. You can create segments with it, connect them, and add Hebrew speech via ElevenLabs and lipsync.

Are the models biased towards people? I need videos of scientific processes.

In terms of software, VO 3.1 and Kling are the best in the world, and no scientific tool is better. They were simply trained on many more videos of people talking, which is why they are stronger there, as it is a statistical matter. Still, it is very worth trying, as we have seen impressive videos even of a fetus in the womb. Provide a first frame and a last frame of the process, and let the tool complete the middle.

Can I use my own image instead of a fictional character?

Yes. The standard tool is HeyGen, which replicates you into a digital twin from a short video. It is excellent for short segments like intros or reels. HeyGen itself has already started integrating VO technology. For a long, one-hour format I would not recommend it, but for short social content, it works great.

Nano Banana tells me 'I do not edit images'. What am I missing?

Make sure you activated the Pro model before running Nano Banana, and that you clicked the banana icon. Sometimes the model simply acts up and refuses. The simplest solution is to open a new chat and enter the same request from the beginning. Do not stubbornly insist on the same chat.

What if the product is new and does not exist in the world yet?

As long as you have a still image of the product, it works. If the product does not exist at all, you can generate a rendering of it using Nano Banana Pro. Important: the input must be a still image, not a video. If you have a video of the product, pause it on a frame, remove the background, and input the image.