Native-UI static ad for DawnBands, a short text-message thread (co-parent blames her, teen texts from bed, sister vent) rendered deterministically and set inside a Dawn brand frame with the band, root cause, mechanism and pill. The thread is a scene of the blame on her, never a review.

Lane: statics · not built yet: 1 (see below) · source: /Users/ayden/.openclaw/workspace/skills/format-ui-mockup

UI mockup — sources

Reference: davidaistar HEa4AVv-ths, "How I Create AI Image Ads For E-com" (analysed in ~/research/davidaistar/analysis/)

What his UI templates do, in order (paraphrased): 1. Research: ~10 minutes in Reddit forums, Facebook groups and competitor reviews for the customer's own phrasing. 2. Inputs: product photo with the background removed, brand name, one-line product description, a funnel-stage template (Starpop.ai on Nano Banana Pro). 3. Gen 1 improvises: leave the extra-details field blank, render 4 variations, let the model pick layout and copy. 4. Gen 2 edits: take the best composition and run a narrow edit that swaps in the exact text and fixes the product scale. 5. BOF set: testimonial card; iMessage thread (contact name, a few bubbles, the product photo sent inline, the whole ad is the conversation); tweet (a testimonial plus a product shot); offer. He files all three UI ones under testimonial / social proof for buyers who already know the product. 6. Claims (his, unverified): UI mockups "lower ad defense"; the Gemini pass attributes "lower CTR from text density, higher BOF conversion" to the UI ads, while the transcript guide attaches that same CTR tradeoff to "three reasons why". Treat both as claims.

Keep: the vessel (a conversation she reads as content), real customer phrasing, and his separation of composition from exact text. We take that one step further: no model touches the text at all. Change: - Content. His thread is a testimonial with the product sent in the chat, and testimonial statics are Dawn's 5 weakest pairings (MAP.md §2B). Ours is the blame scene with no product or result in it. - Funnel. His UI templates are BOF (product-aware). Dawn's gate is problem-aware, so the product zone outside the thread does the teaching (root cause + mechanism). - Rendering. His is AI composited; ours is deterministic HTML, because image models garble UI text and his own test showed forced exact text degrades the render. - His tweet: not built (feed-inside-a-feed, Fish 9-25 logic; Fish's go needed).

davidaistar statics diff (analysis/diff_statics.md)

Gap row: "No UI-mockup templates (iMessage thread / fake Tweet) anywhere in the 20-template library." Proposed D16 iMessage-thread "w/ a mom's verbatim VOC line as the message". This skill is that D16, reasoned against our receipts (references/d16-proposal.json).
The diff also warns: don't resurrect fake-scarcity UI vessels (call screen, closing-down letter) without the compliance check. Those stay banned here.
REPORT.md §2.11: map his funnel templates onto recipe B; Dawn is problem-aware, so mid-funnel.

Our receipts (fake-UI on Dawn)

Thiswakesthem2 (template D08, eyebrow-headline-cloud): $3,776 / 164 / $23.02 CPA, best static CPA in the account. The ledger autopsy called the "clickable article-UI look" the winning variable. Later ruling (learnings.json, "not_working"): "Fake-UI as a scalable FORMAT, 0-for-2 post-autopsy… Thiswakesthem2's $23 CPA was execution-specific, not format-carried."
sniper-p1p2-2026-07-20, P1-A fake-UI statics (a1-alt, a2, a3): review card / search bar / star rating, message constant. $48.66 total, 0 purchases. The ledger's Batch #19 line ($259.46 / 2p / $130 CPA, "review card, iMessage, portal") is the WHOLE sniper batch; both purchases came from b2 (the blame-removal arm), not fake-UI. Its iMessage label doesn't match the batch's own test design.
m1-retroproof-fakeui-2026-07-25: m1a attendance portal $20.23 / 0; m1b review card $0.18; m1c sister text thread $0.49 ("the 6:45am rant thread went quiet": a before/after result story). Shared ad set → Meta put the spend on one ad; the thread was never read.
Attendance-shame statics: $121 / 0 across two batches (learnings.json) → school threads are reserve only.
Blame/lecture/villain framings ("you're-not-lazy", "stop being the villain", lecturing the teen) died in every batch → the accuser in a thread is reasonable and wrong, and the ad never lectures her.

Doctrine this format had to reconcile

skills/dawn-statics/SKILL.md eight locks + Fish 9-30 rulings + 10-04 "Next format lane" (text-message overlay on a real band photo).
skills/native-image-gen/SKILL.md: POST-SCREENSHOT BAN (9-25, FB/Reddit post) and the Notes-app ban (9-30). In natives, "screenshot/feed vessels are out entirely". This format is statics-only for that reason.
static-templates/README.md "Not templated": Notes-app lists (Neurolyfe) and call-screen vessels are banned.
corrections-archive-2026-10-03: no attribution on static quotes (supersedes lock 3's forum credit line).
feedback_native_image_sourcing_corrections 8(d): a generic light-mode HTML Notes render "doesnt look enough like the notes app"; copy a REAL iOS ref. Its ref path (output/dawn-sep25-adhd-iterations-images/notes-ref-rd/) is no longer on disk.
MAP.md §2B: recipe B (brand voice to her, burden, exhaustion, guarantee/science authority; testimonial = weakest).

starpop.ai articles (written versions, read 10-06 from `~/research/davidaistar/starpop/txt/`)

how-to-create-product-photos-with-ai: the source photo decides the output (sharp at 100% zoom, soft even light, full resolution) and a composited element's shadow must follow the scene's light direction or it reads as pasted. Adopted for Variant B: the shot spec sent to Fish and the card-shadow rule (SKILL §7, render-spec Variant B). His scene-prompt template and marketplace export table don't apply (the band is never generated; statics are 1:1 2048).
ai-static-ads-that-convert, starpop-static-ads-guide: his BOF set is testimonial headline + offer; "quick win: paste a strong testimonial onto a visual that already works". Not adopted: testimonial content in the thread (Dawn's 5 weakest pairings, MAP §2B) and BOF framing (Dawn is problem-aware). Reusing a paying static's image as the Variant B background is a possible later one-variable test, but the D-grid winners carry their own text, so it needs Fish's call first.
ai-ad-formats-guide: synthetic content must not mislead viewers about who said something or what happened. Supports §12 "stitched thread": no caption or primary text may claim the exchange was real.
create-product-mockups-with-ai: a tool roundup (template generators, AI UI frames, packaging mockups). Nothing for this lane: our thread is a deterministic HTML render, and the band ships in a pouch that is never mocked up.