Mirra October ads: review board

Static 9:16 · ugly lo-fi · remake of Mirra's own past meme · CN only · click-to-WhatsApp · ad_id mir-ad-20261008-001

• Generated woman, not the woman in the winner photo (her asset status couldn't be confirmed, so the PRD's alternative is used). Jev first judged v1 too close to the winner's woman; v2 changed hair, top, phone and room, and Jev passed it.
• Hands at full size: 2 (phone hand + waistband hand), each joined to an arm. Phone covers the face. No skin.
• Image audit (G7, ugly-ad mode): NOTES 8.2. Its 'R88 hard fail' line is a known checker bug (it fires on the word 'R88' in its own text); its 4:5-only and bottom-logo rules don't fit this PRD (1:1/9:16 sizes, Mirra logo top-right).
• Second sizes (per the PRD) come after your GO on this master.

Your call: GO on this master? Then I make the second size and put both on the chain.

Static 9:16 · ugly lo-fi · native iOS Messages · CN only · click-to-WhatsApp · ad_id mir-ad-20261008-002

• The photo bubble is Mirra's real dish photo, whole (not cropped), with iMessage-style rounded corners. A real iOS status bar replaced a blurred band; the pill sits clear of the input bar.
• Kept the native iOS colours on purpose (Jev: the native look is what makes it work).
• Image audit (G7, ugly-ad mode): NOTES 8.4. Its 'R88 hard fail' line is a known checker bug (it fires on the word 'R88' in its own text); its 4:5-only and bottom-logo rules don't fit this PRD (1:1/9:16 sizes, Mirra logo top-right).
• Second sizes (per the PRD) come after your GO on this master.
• G7 also calls Mirra's own laksa photo 'studio-styled'; it is the real menu photo, left as is.

Your call: GO on this master? Then I make the second size and put both on the chain.

Static 1:1 · ugly lo-fi · CN only · click-to-WhatsApp · ad_id mir-ad-20261008-003

• The bento is codex's drawing of the real Japanese Furikake Rice (dish 83); it matches the real photo (lotus chips, furikake rice, spinach with chilli, seaweed tofu, both sauce cups); only the small empty lid cup is missing.
• Hand at full size: 1, joined to the arm. The red marker became Mirra pink (red is banned on paid).
• Image audit (G7, ugly-ad mode): NOTES 8.4. Its 'R88 hard fail' line is a known checker bug (it fires on the word 'R88' in its own text); its 4:5-only and bottom-logo rules don't fit this PRD (1:1/9:16 sizes, Mirra logo top-right).
• Second sizes (per the PRD) come after your GO on this master.

Your call: GO on this master? Then I make the second size and put both on the chain.

Static 1:1 · ugly lo-fi · two-panel meme · CN only · click-to-WhatsApp · ad_id mir-ad-20261008-004

• All five dishes are Mirra's real photos, whole: Japanese Furikake Rice on top; Sweet & Sour Hericium Rice, Konjac Asam Laksa and two more below. No gift item shown.
• Cream ground instead of white and no sparkles (Mirra's paid rules). The cats are newly drawn look-alikes, not the meme photos.
• Image audit (G7, ugly-ad mode): REJECT 7.4. Its 'R88 hard fail' line is a known checker bug (it fires on the word 'R88' in its own text); its 4:5-only and bottom-logo rules don't fit this PRD (1:1/9:16 sizes, Mirra logo top-right).
• Second sizes (per the PRD) come after your GO on this master.
• Open G7 points I left for you: it says the right side has dead space and the bottom grid is smaller than the top photo; it also calls a garnish in Mirra's own dish photo (Teriyaki Mushroom Asada Burrito Bowl) 'an invented ingredient' — that is the real photo.

Your call: GO on this master? Then I make the second size and put both on the chain.

Static 1:1 · ugly lo-fi · CN only · click-to-WhatsApp · ad_id mir-ad-20261008-005

• The closed boxes and the open bento are codex's drawing of the real Mirra box and the real Sweet & Sour Hericium Rice (dish 53).
• Remade in one fresh pass after the image audit found codex's sleeve print misspelt ('Smilal'), a fridge brand name, red ticks and, after repeated edits, waxy food. Now: sleeves show only the MIRRA wordmark, no brand names in the room, pink ticks, food drawn once.
• Image audit (G7, ugly-ad mode): NOTES 8.5. Its 'R88 hard fail' line is a known checker bug (it fires on the word 'R88' in its own text); its 4:5-only and bottom-logo rules don't fit this PRD (1:1/9:16 sizes, Mirra logo top-right).
• Second sizes (per the PRD) come after your GO on this master.

Your call: GO on this master? Then I make the second size and put both on the chain.

Static 4:5 · ugly lo-fi · CN only · click-to-WhatsApp · ad_id mir-ad-20261008-006

• The printed photo is Mirra's real dish photo, whole.
• The notice and both tape strips sit on one door only (taped across the seam it would tear when the doors open).
• The photo slot is about 3% narrower than the photo, so a sliver comes off each side of the real photo (the dish stays whole).
• Image audit (G7, ugly-ad mode): NOTES 8.4. Its 'R88 hard fail' line is a known checker bug (it fires on the word 'R88' in its own text); its 4:5-only and bottom-logo rules don't fit this PRD (1:1/9:16 sizes, Mirra logo top-right).
• Second sizes (per the PRD) come after your GO on this master.

Your call: GO on this master? Then I make the second size and put both on the chain.

Static 9:16 · ugly lo-fi · CN only · click-to-WhatsApp · ad_id mir-ad-20261008-007

• Fixed after the image audit: removed a red-and-white object at the bottom right, full-width Chinese comma on the sign, cream sign mounted on the posts, warmer sky, crisper trees.
• Image audit (G7, ugly-ad mode): REJECT 6.6. Its 'R88 hard fail' line is a known checker bug (it fires on the word 'R88' in its own text); its 4:5-only and bottom-logo rules don't fit this PRD (1:1/9:16 sizes, Mirra logo top-right).
• Second sizes (per the PRD) come after your GO on this master.
• Open G7 points I left for you: it calls the warm sky 'golden hour' (Jev chose the warm grade over the cold blue), says the steel frame is very blue (that is the reference billboard's frame), and says the headline is not fully in perspective.

Your call: GO on this master? Then I make the second size and put both on the chain.

Static 4:5 · click-to-WhatsApp · skeleton: Factor fibre grid · ad_id mir-ad-20261007-005

• Cells left to right, top to bottom: Mapo Tofu Rice · Jap Chae · Dry Classic Curry Konjac Noodle · Spicy Fusilli Pasta · Ginger Beancurd Seaweed Rice · Mei Cai Fried Rice.
• The comma after 六个午餐 came out as a Latin comma, not the Chinese one.
• 8 Oct: the MIRRA wordmark moved in to 70 px from the right edge, as the logo spec says (it sat 16 px from the edge). Nothing else changed.

Your call: Keep the cut-off bottom row (the reference's look), or make both rows full height? And may I give the six photos one shared warm grade so they match?

Static 4:5 · click-to-WhatsApp · skeleton: Planthood offer card · ad_id mir-ad-20261007-004

• The 低于 500 卡 (under 500 cal) pill stays: your standing rule is a pill on every ad.
• The comma after 500 卡 still comes out as a Latin comma; codex ignores the request.
• 8 Oct: the MIRRA wordmark moved in to 70 px from the right edge, as the logo spec says (it sat 16 px from the edge). Nothing else changed.

Your call: The true pixel edit that keeps the real photo (about US$0.17) was refused: the OpenAI API account has no credits left (nothing was charged). Top up the OpenAI credits and I'll run it, or approve v2c with the redrawn food.

Static 9:16 1080x1920 (+ a 4:5 crop) · built on Mirra's own past winner S19 · illustration, not a customer photo · ad_id mir-ad-20261007-007

• No price on this ad (per its brief).
• The same illustrated woman in both panels; the change shows in her outfit, posture and lunch, never her body.
• 8 Oct: the MIRRA wordmark moved in to 70 px from the right edge, as the logo spec says (it sat 16 px from the edge). Nothing else changed.

Your call: OK as is? (Earlier version v2 was rejected for the pill sitting under the Reels buttons; v3 fixes that.)

Carousel · 4 cards 1:1 · skeleton: Eatology · every food photo is Mirra's own, placed pixel for pixel · ad_id mir-ad-20261007-008

• At thumbnail size the shiitake on card 2 and the tempeh on card 4 could read as meat; the dish tags help.
• 8 Oct: the MIRRA wordmark moved in to 70 px from the right edge, as the logo spec says (it sat 16 px from the edge). Nothing else changed.

Your call: OK as is, or swap card 3's photo for one with a clear top-right corner? And is the tempeh photo fine under its dish name?

Single image 1:1 · card 1 without the counter, dots and swipe prompt · ad_id mir-ad-20261007-009

• 8 Oct: the MIRRA wordmark moved in to 70 px from the right edge, as the logo spec says (it sat 16 px from the edge). Nothing else changed.

Carousel · 4 cards 1:1 · skeleton: DayDay Catering · every food photo is Mirra's own, placed pixel for pixel · ad_id mir-ad-20261007-010

• No dish names on these cards (brief: tag-none).
• Card 4 uses the brief's fallback photo; the real hands-opening-the-bento still isn't extracted yet.
• 8 Oct: the MIRRA wordmark moved in to 70 px from the right edge, as the logo spec says (it sat 16 px from the edge). Nothing else changed.

Your call: OK as is? (The brief caption still says RM17++ in two places; I'll change it to RM19++ before upload.)

Single image 1:1 · exactly card 1 (the test compares carousel vs single image) · ad_id mir-ad-20261007-011

• 8 Oct: the MIRRA wordmark moved in to 70 px from the right edge, as the logo spec says (it sat 16 px from the edge). Nothing else changed.

R7F · at home, bedroom full-length mirror · face hidden (phone over the face / framed below the chin) · ad_id mir-ad-20261007-023

• Made with codex ($0): both halves are drawn in ONE image so the woman, outfit, room and camera match, then split at the white gap. The prompt locks the same trousers, one watch on the same wrist, a plain tee and the same skin in both halves, and asks for an obvious difference.
• Hands checked at full size: exactly 2 in each half, each joined to its arm. No skin at the waist. Same trousers and watch in both halves.
• Face hidden by framing (cropped at the chin).
• Jev: picked over the other seed (0.73/0.56). G7: 7.0.
• G7's only flags are visible at 2x: a thin light sliver behind one belt loop and a soft edge where the thumb pinches the waistband.
• G7 also marks every pair down for having no pill, logo or ad canvas; that is expected, because these stills only feed the reel.

Your call: Approve these stills? On your GO they go to ba-inserts/, and I'll show you the Wan motion dry-run and cost before firing (paid).

R7F · at home, bedroom full-length mirror · face visible (blurred or barred in the reel edit) · ad_id mir-ad-20261007-024

• Made with codex ($0): both halves are drawn in ONE image so the woman, outfit, room and camera match, then split at the white gap. The prompt locks the same trousers, one watch on the same wrist, a plain tee and the same skin in both halves, and asks for an obvious difference.
• Hands checked at full size: exactly 2 in each half, each joined to its arm. No skin at the waist. Same trousers and watch in both halves.
• Face visible on purpose (version B); it gets blurred or barred in the reel edit, per your 6 Oct rule. G7 marks B down for showing the face.
• Jev: show, low confidence (0.31/0.43). G7: 6.4, no hand or continuity defect.
• G7 says the 'before' waistband shows few strain creases and the cheek is slightly smooth; it also notes the face is the same in both halves, which the continuity rule asks for.
• G7 also marks every pair down for having no pill, logo or ad canvas; that is expected, because these stills only feed the reel.

Your call: Approve these stills? On your GO they go to ba-inserts/, and I'll show you the Wan motion dry-run and cost before firing (paid).

R4F · at home, plain living-room wall (4:5, for the reel's 9:10 top panel) · face hidden (phone over the face / framed below the chin) · ad_id mir-ad-20261007-017

• Made with codex ($0): both halves are drawn in ONE image so the woman, outfit, room and camera match, then split at the white gap. The prompt locks the same trousers, one watch on the same wrist, a plain tee and the same skin in both halves, and asks for an obvious difference.
• Hands checked at full size: exactly 2 in each half, each joined to its arm. No skin at the waist. Same trousers and watch in both halves.
• Face hidden by framing (cropped at the chin).
• Jev: pass. G7: NOTES 8.3, no AI artifacts.
• G7 also marks every pair down for having no pill, logo or ad canvas; that is expected, because these stills only feed the reel.

Your call: Approve these stills? On your GO they go to ba-inserts/, and I'll show you the Wan motion dry-run and cost before firing (paid).

R4F · at home, plain living-room wall (4:5, for the reel's 9:10 top panel) · face visible (blurred or barred in the reel edit) · ad_id mir-ad-20261007-018

• Made with codex ($0): both halves are drawn in ONE image so the woman, outfit, room and camera match, then split at the white gap. The prompt locks the same trousers, one watch on the same wrist, a plain tee and the same skin in both halves, and asks for an obvious difference.
• Hands checked at full size: exactly 2 in each half, each joined to its arm. No skin at the waist. Same trousers and watch in both halves.
• Face visible on purpose (version B); it gets blurred or barred in the reel edit, per your 6 Oct rule. G7 marks B down for showing the face.
• Jev: show (0.62/0.67, both orders). G7: 4.8.
• G7 says the dark inner waistband inside the gap can look like a phone at thumbnail size (at 2x it is clearly the waistband), the button twists a bit far, and the 'after' face is a little smoother.
• G7 also marks every pair down for having no pill, logo or ad canvas; that is expected, because these stills only feed the reel.

Your call: Approve these stills? On your GO they go to ba-inserts/, and I'll show you the Wan motion dry-run and cost before firing (paid).

R11 · at home, hallway mirror selfie · face hidden (phone over the face / framed below the chin) · ad_id mir-ad-20261007-028

• Made with codex ($0): both halves are drawn in ONE image so the woman, outfit, room and camera match, then split at the white gap. The prompt locks the same trousers, one watch on the same wrist, a plain tee and the same skin in both halves, and asks for an obvious difference.
• Hands checked at full size: exactly 2 in each half, each joined to its arm. No skin at the waist. Same trousers and watch in both halves.
• Face hidden by framing (cropped at the chin).
• Jev: picked. G7: NOTES 8.2, no AI artifacts.
• G7 also marks every pair down for having no pill, logo or ad canvas; that is expected, because these stills only feed the reel.

Your call: Approve these stills for the R11 reel? (The R11 reel itself still needs your GO.)

R11 · at home, hallway mirror selfie · face visible (blurred or barred in the reel edit) · ad_id mir-ad-20261007-029

• Made with codex ($0): both halves are drawn in ONE image so the woman, outfit, room and camera match, then split at the white gap. The prompt locks the same trousers, one watch on the same wrist, a plain tee and the same skin in both halves, and asks for an obvious difference.
• Hands checked at full size: exactly 2 in each half, each joined to its arm. No skin at the waist. Same trousers and watch in both halves.
• Face visible on purpose (version B); it gets blurred or barred in the reel edit, per your 6 Oct rule. G7 marks B down for showing the face.
• Jev: picked. G7: 7.4, no AI artifacts.
• G7 also marks every pair down for having no pill, logo or ad canvas; that is expected, because these stills only feed the reel.

Your call: Approve these stills for the R11 reel? (The R11 reel itself still needs your GO.)

Reel 9:16 · 30 s · Wan 3.0 · the real closed Mirra box is the only reference · ad_id mir-ad-20261005-001

• Changed for your RM19++ ruling: the friends line no longer says 'as low as seventeen-something'; it now says 'order with friends or colleagues and it's cheaper' (Jev agreed in both orders).

Your call: GO to fire? One paid fire, about US$3.00 list (Higgsfield Wan, US$0.10/s).

Reel 9:16 · 30 s · Wan 3.0 · the real closed Mirra box is the only reference · ad_id mir-ad-20261005-002

Your call: GO to fire? One paid fire, about US$3.00 list (Higgsfield Wan, US$0.10/s).

Reel 9:16 · 7 s hook · Wan 3.0 · the real closed Mirra box is the only reference · ad_id mir-ad-20261005-003

• Fires only after P1 is approved: its body is P1's take.

Your call: GO to fire? One paid fire, about US$0.70 list (Higgsfield Wan, US$0.10/s).

Reel 9:16 · one 26 s Wan take + real dish photos and end card in the edit · generated women · ad_id mir-ad-20261007-026

• The lid lifts toward the lens and the edit cuts to the real Mirra photo; open food is never generated.
• Pace 1.5–3.0 syllables per second, every line checked.

Your call: GO to fire? One paid fire, about US$2.60 list.

Reel 9:16 · one 26 s Wan take + real dish photos and end card in the edit · generated women · ad_id mir-ad-20261007-027

• Her colleague's bite is silent: no taste verdict from a generated person.

Your call: One line to confirm: in 'I switched to Mirra' a generated woman speaks as a Mirra user (no result claimed). The signed brief depends on it. OK to keep? Then GO to fire (about US$2.60).

Reel 9:16 · 30 s · narrator only: the generated woman never speaks · uses the R11 before/after stills above · ad_id mir-ad-20261007-028 (A) · mir-ad-20261007-029 (B)

• Wan fires: B1 opening from each version's before still (3 s each), one shared tapau + closed-box take (10 s), the before→after shot for A and B (4 s each). The change happens while her palm covers the lens, so nothing morphs on screen.
• Narrator voice is a separate step (ElevenLabs or Wan voice).

Your call: Approve the R11 stills above, then GO for the five Wan fires (about US$2.40 total)?

Reel 9:16 · ~42 s · a 15 s Wan skit (generated women) + real order-surge, kitchen and scoop footage + narrator · ad_id mir-ad-20261007-012

• Your rule: generated people never speak as customers. Jev scored the two held lines 0.70–0.76 as testimonial-like (control lines 0.05–0.23).

Your call: Fire R1 as written (US$1.50), or the version without the two held lines (US$1.10)?