Most AI ad tools have one answer to every brief: generate it. That works well for a talking-head UGC ad and badly for an iMessage conversation, a pricing card or a screenshot of an app, because those formats are made of text and text is the thing image and video models are worst at. Letters warp, words get invented, and a mockup with slightly wrong proportions is obvious to anyone who uses the app it is imitating.
So these 13 formats are built the other way round. Each is a self-contained HTML page whose animation is a pure function of time; a headless browser steps it frame by frame and ffmpeg assembles the result. Nothing is generated. The copy is real DOM text, the iOS chrome is pixel-accurate, and the same inputs produce the same file every time.
AirDrop product carousel
An iOS AirDrop share sheet springing up over a blurred product backdrop, then a carousel of product cards moving through the window with the AirDrop chime.
Read the format → Video · deterministicApp UI tour
Real app screenshots rendered inside a CSS iPhone mockup — rounded body, metallic rim, dynamic island, screen glare — floating on a soft brand-coloured stage as the tour moves through screens.
Read the format → Video · deterministicApple Notes reveal
An iPhone note typing itself out — a bold title first, then short confessional or listicle lines beneath it, with quiet keystroke ticks, closing on a branded card.
Read the format → Carousel · deterministicBranded carousel
A designed multi-slide carousel — a cover, a run of one-point content slides, and a close — rendered as square PNGs for Meta carousel ads, LinkedIn document ads and Instagram.
Read the format → Video · deterministicChatGPT chat-reveal
A ChatGPT conversation rendered at the app's native geometry — a question typed into the composer with the keyboard up, then the answer streaming back, ending on your brand.
Read the format → Video · deterministiciMessage notification cascade
A stack of iOS notification banners springing in one after another over a blurred backdrop, each pushing the stack upward.
Read the format → Video · deterministiciMessage chat-reveal
A pixel-accurate iMessage conversation that types itself out — bubbles arriving with the real send and receive sounds, a typing indicator, and the product appearing as a genuine URL-preview rich link rather than a pasted image.
Read the format → Video · deterministicKinetic type
Rhythmic typographic motion design — short phrases landing word by word on full-bleed brand cards, a skewed accent slab wiping each cut, with product shots punctuating the rhythm.
Read the format → Video · deterministicMyth-vs-fact explainer
A voiceover-led kinetic piece: a calm narrator reads myth-and-fact pairs, each myth line getting a red strike sweep before the fact lands, with word-level highlighting synced to the read.
Read the format → Video · deterministicPhoto-grid collage
A moodboard that assembles itself — real product and lifestyle photos dropping in one tile at a time with a soft bounce, then paying off as a completed grid under a brand lockup.
Read the format → Image · deterministicNative-style image ad
A still native-style image ad in one of three sub-styles: a frozen iMessage thread at its best moment, an Apple Notes note, or a designed brand promo card.
Read the format → Video · deterministicValue-prop motion ad
A bold hook sticker followed by three to five punchy benefit claims animating over rotating product shots, alternating brand-colour blocks with photo beats.
Read the format → Video · deterministicCinematic vignette
A cinematic single-hero product film built from one real photograph — a slow push, layered depth, film grain, a brand-tinted glow, and short copy lines dissolving through the lower third.
Read the format →How this template renders
Nothing here is generated by an image or video model. The format is a self-contained HTML page whose animation is a pure function of time — no CSS keyframes, no timers, no randomness — and the renderer steps that page frame by frame in headless Chrome, screenshots each frame, and assembles them with ffmpeg into an H.264 file that ends on a crossfade into your brand end card.
Two things follow from that. The render is deterministic: the same inputs produce the same file, so a format that looked right yesterday looks right today. And every word on screen is real browser-rendered text, which means the copy is exact and stays sharp at any resolution. The whole thing costs a couple of credits rather than a video generation, which is what makes these formats practical to run at volume.
What they cost
A couple of credits, because there is no model in the loop — the cost is browser CPU and ffmpeg. That is the practical difference: a video-model render is priced per second of output, so testing ten variations of a hook is a real budget decision. Testing ten variations of a template ad is not.
The rest of the catalog
These sit alongside 110 creative recipes that are model-rendered — UGC talking-heads, product films, viral product formats — and 18 explainer visual styles.
Driving it from an agent
Every format on this site is reachable from an AI agent as well as the web studio. Hermoso ships an MCP server, so Claude, Claude Code, Cursor, Codex and any other MCP client can call plan_ad to draft the creative and render_ad to produce it, name a format as the recipe, and get a finished file back — with the same brand grounding and the same composited-text guarantees. The MCP and CLI overview covers the setup, and the generated tool reference is the complete list.
Render a native-format ad in a couple of credits — the free plan needs no card.
Start free → See pricing