AI UGC ads
The video argues that AI UGC can hold your product convincingly if you fix the inputs and the workflow instead of fighting the model. A good reference-image creator on a top video model solves the product-holding and realism problems. Native-format ads written in your customers' own language make the videos convert, and libraries of formats and angles plus a Claude–Max Fusion MCP link turn that into a repeatable system that produces 10–20 videos a week.
The video names three problems with AI UGC: the creator holds the product awkwardly and doesn't match the brand, the speech and look read as obviously AI, and one video takes hours when a brand needs 10 to 20 every week. It promises to solve all three through a three-level ladder: beginner, intermediate and expert.
As proof that the approach works, the video shows ad results and points to the heavy spend behind them. Its reasoning is that a brand wouldn't keep spending that much unless the ads were making a profit.
The video says Sora 2 Pro is the best model for realistic AI UGC, but that its API expires in September. It therefore urges viewers to make as many videos as they can right now.
The ladder moves from a single clip of someone holding the product (beginner), to high-performing content built on native formats and customer research (intermediate), to a proven, repeatable system built on libraries and tool integration (expert). The expert end state is a stocked library of ads that can be downloaded and handed to an editor without the founder spending hours.
The video bases its intermediate level on how people use Meta, TikTok and Facebook. They scroll while relaxing, not to learn (that is YouTube) and not to buy (that is the store), so the ad's job is capturing attention and diverting it to the product. That is why the video says ads should look like native organic content rather than like ads.
The video leaves the expert-level system only outlined and says building it is beyond the scope of this video. It offers a follow-up video if viewers ask in the comments, and it promotes its school community of ecom brand owners with a weekly ad-review call.
All-models platform with Sora 2 Pro briefed like a real creator — The video recommends one platform that hosts all video models (Max Fusion, or File or Kia on a budget) and names Sora 2 Pro as the best model for realistic AI UGC, which it says uses a given product perfectly. Apply: Open the platform, select Sora 2 Pro, add a picture of your product, and describe the video as you would to a real creator, for example a TikTok get-ready-with-me routine where the creator applies the foundation to her face.
Pinterest reference-image creator build — A way to create a new, realistic-looking creator by starting from a real Pinterest photo of the right person in the right setting. The video combines that photo with the product image in Nano Banana Pro, then uses the result as the starting image for the video model. Apply: Search Pinterest for a specific person in the product's setting (e.g. 'UGC makeup table creator') and keep clicking images until you find a good example. Drag it into Nano Banana Pro with your product image, have the AI put the product somewhere, prompt 'change her face so it looks like XYZ', then use the result in Sora 2 Pro, Seed Dance or Kling.
Native format plus customer-language creative strategy — A three-part process for ads that convert: find a native content format that already performs well, do deep customer research, and rewrite the format's words in language your customers actually use. Apply: Research a specific avatar on Reddit, TikTok and similar places. Swap a generic caption like 'Watch me apply this makeup' for avatar-specific language, such as 'When you have 5 minutes to apply your foundation before your work husband arrives' for busy single working moms.
Format-to-prompt pipeline (TikTok content search, content analyzer, Claude) — A workflow inside Max Fusion that finds proven native formats on TikTok, turns the video into a prompt with the content analyzer, and has Claude adapt that prompt to your product and customer research. Apply: Right-click in Max Fusion, open TikTok content search, add a proven format to a board and use the content analyzer to get a prompt. Paste it into Claude with an instruction to change it for your product and avatar using your research document and an angle, then take the updated prompt back to Max Fusion and add a picture of your product.
Format library and winning-angle library — The expert-level foundation: one library of high-performing viral formats and their variations, and a second library of your brand's proven winning angles, so angles can be applied to different formats. Apply: Spend one day a week finding as many high-performing viral formats and variations as you can, log them somewhere you can access at will, and keep a separate record of your proven winning angles to apply to those formats.
MCP link between Max Fusion and Claude — An MCP connects an AI like Claude with the software you are using. In the video it connects Max Fusion to Claude so ads are built from inside Claude rather than by typing prompts on a screen. Apply: Connect Max Fusion to Claude and add your format ideas and angles to a Claude project. Write and generate inside Claude chat, run many chats at once, and collect a library of ads to download and give to your editor.
The video says that prompting a random creator holding your product as an image gives a plasticky result, even when you add 'realistic iPhone shot'. It treats realism as a problem of the reference image, not of prompt wording.
The reference image is chosen by matching a specific person in a specific location to how the product is used. For a foundation, that means someone at a makeup table in their room.
The same reference-based creator image can be fed into Sora 2 Pro, Seed Dance or Kling. The video shows three different results from it, so the creator build is separate from the choice of video model.
The video's prompt advice is to brief the video model as you would brief a real creator. Its examples are a TikTok get-ready-with-me routine and a taste test of different drinks where the creator likes the one from the image.
The video separates the jobs of format and words. The native format earns attention, and conversion depends on swapping generic captions for avatar-specific language found by researching customers on Reddit, TikTok and similar places.
The video splits the expert-level libraries into formats and angles. Proven winning angles can then be applied to many different formats, and the video recommends setting aside one day a week for hunting formats.
The video turns viral videos into prompts with the content analyzer, then has Claude rewrite those prompts for your product, avatar, research document and angle. This avoids writing prompts from scratch.
The video says its MCP setup lets many chats run at once inside Claude, so a library of finished ads builds up without the founder doing the work by hand.
«you spend hours to make one video when you need 10 to 20 every single week to even scale your brand and afford a bigger team»
— 00:26
«a brand wouldn't do that if it wasn't actively making them a profit»
— 00:52
«the API for Sora 2 Pro is going to expire IN SEPTEMBER.»
— 01:34
«describe the video as if you're describing it to a real creator»
— 01:48
«you want to go on Pinterest and you want to search a specific person in a specific location where you want them to use a product»
— 02:34
«what you have to realize is that you are in the game of capturing attention and diverting it to your product or to your service»
— 04:15
«When you have 5 minutes to apply your foundation before your work husband arrives.»
— 05:11
«spend 1 day a week just going through and finding as many high-performing viral formats as you can»
— 06:54
«An MCP just allows you to connect an AI like Claude with the software that you're using.»
— 07:23
Reception
Viewers were warmly positive, praising the humor and the practical value of the video and asking about a follow-up and the paid community. Much of the rest is repeated promotional spam for TalkPix, Tagshop AI and similar tools, and there was some price and budget concern.
A compact three-level walkthrough that treats AI UGC product-holding as a problem of input quality and workflow rather than prompting. Its detail is strongest at the beginner reference-image workflow, and the expert-level system is only sketched and deferred to a possible future video.

08:17