Unnamed company
Make ad variants with AI: create an image first, then turn it into video
Used for clients; not compared with creator videos
Unknownpaid-social
What they did
Freelancers make ~99% of client ads with AI. Use Higgsfield as a model aggregator (one subscription, credits). Generate the scene as an image first (GPT Image 2 for consistent text; Nano Banana, weaker on fine print) until it looks authentic, then use it as the reference frame for a video model (Seedance 2 for authentic UGC style, Veo 3, Kling) - far cheaper than prompting video directly. Once a format is proven (e.g. text wall: two scenes + app demo in the middle + text block on top), encode it as a node-based workflow in Weavy or Flora with variables to generate hundreds of iterations. Rationale is control over variables and future self-learning loops, not cost savings; creators on retainer can still be cheaper today.
What happened
Used in production for clients; no performance comparison vs creator content given.
“I wouldn't do it from a cost savings perspective, but to me, if we can do it in AI, that means variables become very controllable.”
“And then you use that as the reference image within the video model because it's going to be much easier to get it right with an image model and it's much cheaper.”
“we always want to create text walls that consist of two scenes, an app demo in the middle, and it's going to be text block on top. You can pretty much automate that in something like Weevy.”
Related topics
Get tactics like this every Monday
New tactics from the week's founder interviews, each linked to where it was said.
More from this episode
- Say 'app' or show a demo so the App Store isn't a surpriseMost winning app ads do this; installs suffer otherwise
- Moongate: Remake one winning ad ~400 times with new hooks and use casesWell over $1M spent on the concept
- Run short videos with a wall of text in the audience's own pain wordsOne of the main working formats; no numbers
- Feed 10-20 Reddit threads to Claude to write ads in your audience's wordsHis standard process; no numbers
- Picnic: Mix app demo clips into everyday videos, like a chore list for ADHD
- Nibble: Run ads that look like normal posts; let a web page do the selling
- Blinkist: Send Meta ad clicks to your blog instead of the App StoreClaims to be one of the first to scale this way; no numbers
- Compare cost per trial with trial-to-paid rate for each age groupAges 25-34 cost the most; older users looked under-served
- Give ads that deliver differently, like mostly Facebook feed, their own ad set
- Split ads into groups by format and message, each with its own budget
- On a small budget, put all ads in one group and pick winners weekly
- Have every ad team member interview one user a weekTaught the team most of how to write ad copy; no numbers
- Post on LinkedIn 2-3 times a week
- Fix the store page, onboarding and paywall alongside paid ads