A GCC business sets up once: business details, logo, brand colors, and the posts and posters it already has. Ishhar learns the brand from them, then each week either suggests the campaign or takes the owner's own offer. It generates the videos, posts and posters, schedules them, and publishes to Instagram, Facebook, YouTube, X and WhatsApp by itself. A weekly marketing operation that runs on autopilot.
3 of 4 kept. Layout, photo style and caption rhythm learned from what already worked.
Evenings in Dubai drop under 30°C this week. Lead with sunset karak on the terrace.
Your August combo post got 3.1× your usual reach. Bring it back for weekday mornings.
2 December is six weeks out. Start a teaser series in the brand colors from 1 November.
0 9 * * 4Every Thursday 09:00, draft next week's campaign*/5 * * * *Every 5 minutes, publish posts that are due0 20 * * 3Wednesday 20:00, remind the owner if approval is pendingWalkthrough of the full product with a fictional cafe. Steps 1 to 3 run today in Phase 1. Connected publishing ships in Phase 3, and the cron-driven autopilot in Phase 5. Click any step to jump to it.
Name in EN and AR, category, branches, opening hours, logo and brand colors. Done once.
Owner, onceUpload previous posts and posters. The good ones become style references; weak ones are skipped with a reason.
Owner, oncePalette, fonts, tone, forbidden words, CTA style and what performed. Every campaign inherits it.
IshharThree suggestions each week from season, regional calendar and past results. Or the owner types their own offer.
Owner picksScript, storyboard, voice, style. Four checkboxes. Nothing spends money before this.
Owner approvesReel, posts, stories, posters, captions EN + AR. One 9:16 master becomes 1:1, 4:5 and 16:9 with logo and subtitles. Every call logged to the cost ledger.
IshharDays, times and platforms set once. Cron jobs draft next week's campaign and queue every post.
Owner, onceA dispatcher posts to Instagram, Facebook, YouTube, X and WhatsApp when each slot comes due. Then the loop starts again.
IshharThe moat is steps 02-03, 05 and 07. Models are rented and commoditizing. What a competitor cannot copy quickly is a brand that gets learned from the business's own history, a cost gate that keeps margins intact, and a schedule that keeps running whether or not the owner logs in.
The full brief-to-pack pipeline, multi-tenant on Supabase with RLS, orchestrated by n8n, assembled with ffmpeg.
Pay to use from day one. No manual billing, no DM outreach, no free builds. The product sells itself or it does not sell.
The pack stops being a download. Connect accounts once, then publish Reels, posts, Shorts and tweets with captions and hashtags already attached.
Lead times: Meta App Review and Business Verification take weeks, filed in P2. YouTube default quota allows roughly 6 uploads per day per project; request the increase early. X API posting is a paid tier, so its monthly fee goes into COGS.
The module that opens supermarkets, hypermarkets and every business that still prints. Generated layouts, manual control over every element, print-ready output.
A business sets its schedule once and stops thinking about marketing. Cron jobs draft each week's campaign ahead of time, hold it for approval, and publish every post on time without anyone logging in.
Insights flow back from every published post. Ishhar learns what works for each business and feeds it into next week's suggestions. This is where "AI marketing employee" stops being a tagline.
Illustrative mock with a fictional market. In the editor, every tile, price and headline is a selectable element.
No business advertises more often than a supermarket. New offers every Thursday, dozens of SKUs, posters in every aisle, flyers in every apartment building, and a designer or agency invoice behind all of it. Ishhar turns that whole cycle into one Excel upload.
One Excel or CSV: product, price, old price, unit, photo. 40 SKUs land in the campaign in seconds.
Grid posters, a hero offer Reel, shelf talkers per SKU, WhatsApp catalog message, screen loop, flyer PDF.
Manager swaps two photos, fixes one price, approves. Branch addresses and phone numbers stamp per location.
Reel and posts publish on schedule, WhatsApp broadcast goes out, print PDFs head to the print shop for the weekend rush.
Why this vertical is a margin machine: posters and flyers cost cents to produce since the template engine and ffmpeg do the work, not paid video APIs. A hypermarket on Business tier consumes mostly near-zero-COGS assets while paying for the replacement of a designer, an agency and three days of coordination every single week.
| Model | Role | Unit cost | Why this one |
|---|---|---|---|
| Seedance 2.0ByteDance, via host APIs | Video default | ~$0.08-0.12 /s | Ranked first on the Artificial Analysis video arena for both text-to-video and image-to-video. Native audio, strong multi-shot consistency. Growth and Business tiers. |
| Wan 2.7Alibaba, open weights | Budget video | ~$0.05 /s | Best open model, Apache 2.0, native audio, 1080p. Starter tier and the free watermarked campaign. Self-hosting becomes an option at scale. |
| Kling 3.0Kuaishou | Spokesperson | ~$0.10-0.17 /s | Strongest character consistency, multi-shot with up to 6 cuts, 4K. For a recurring branded spokesperson appearing week after week. Business tier. |
| Gemini Omni FlashGoogle | Draft previews | cents /clip | 720p, 10-second clips at a fraction of flagship cost. Cheap storyboard-stage previews so customers approve the look before the real render spends money. |
| Veo 3.1Google DeepMind | Premium | ~$0.15-0.75 /s | Best cinematic quality and lip-sync. Enterprise and premium one-off work only, never a default. |
| Seedream 5 ProByteDance | Image default | ~$0.035-0.07 /img | Wins current design-led comparisons: composition, restraint, ad and editorial imagery. Default for social stills and poster background art. |
| GPT Image 2OpenAI | Text-in-image | ~$0.03-0.08 /img | Best text rendering of the current field. For hero creatives where an English headline is baked into the art itself. |
| Qwen-ImageAlibaba | Arabic in-image | cheap /img | Candidate for Arabic baked-in typography, worth the half-day test given Alibaba's MENA focus. Poster prices never rely on it: those are real text layers. |
| Qwen3-MaxAlibaba, DashScope intl | Marketing brain | <$0.05 /campaign | Strongest Arabic marketing copy of the tested set. Also writes the weekly campaign suggestions and reads uploaded past posts for style. Claude as fallback for English-heavy premium accounts. |
| ElevenLabs Flashv2.5 | Voice | ~$0.03 /campaign | WhatsApp voice notes and standalone audio. In-video narration comes from the video model's native audio track. |
Every provider sits behind a config value, not an abstraction layer. When a model is dethroned or retired, and this market just proved one will be within months, the swap is one value and one prompt template. The cost ledger shows exactly what the change does to margin within a week.
A Dubai social agency retainer runs AED 3,000 to 10,000 per month. Ishhar is priced as a department replacement, never compared to a $12 design tool.
Extra campaign at AED 39. 60-second video costs 2 credits since it doubles video COGS. One-off pack at $39-49 for non-subscribers, the trial-to-paid bridge.
Posters, flyers and shelf talkers render in the template engine for cents. Supermarket-heavy accounts are the highest-margin accounts on the platform.
Video is 80 to 90 percent of COGS. Host pricing for Seedance varies several-fold, so the ledger re-prices margin weekly with real numbers, not estimates.
First product in the portfolio whose distribution depends on nobody's network. You pay, you use. Acquisition is general marketing, and the product is its own best advertisement.
Meta App Review and Business Verification gate Instagram publishing and take weeks. Mitigation: file during Phase 2; download-pack remains a full fallback.
Reve was the image default in July and shut its API in August. Mitigation: providers as config values, text as text layers, cost ledger re-pricing margin weekly.
The model only works if businesses run campaign 5 and 6, not just campaign 1. Mitigation: autopilot and the weekly poster cycle are the retention products; measure week-4 repeat rate from day one.
One embarrassing Arabic headline costs a customer. Mitigation: Qwen3-Max default, real-text poster typography, approval gate before anything renders, publishes or prints.
A scheduled post that goes out with a stale price or on a day of mourning hurts the brand. Mitigation: approval window before spend, a pause-all switch, and calendar-aware blackout dates.