Saved Flowwow 1.5M ₽ on AI ad production

Flowwow was running ads the old way — live actors, product shoots, a production team. For the holiday flight they decided to test whether new methods (AI video generation) could acquire audiences while cutting production cost and resources.

This was a strategic experiment: measure the effectiveness of AI content in performance and media placements, and form hypotheses for scaling the format in 2026.

Three concrete A/B test hypotheses: 1. Can atmospheric AI spots replace UGC actor footage in effectiveness — until now all UGC was phone-shot with live actors 2. Can AI video become a budget alternative to classic product shooting 3. Do "hooks" in the first seconds change CTR and engagement

This is what's rarely discussed about AI projects, but it was the main difficulty.

Flowwow has a clear emotional brand idea: one person taps "place order," another — possibly in a different city or country — gets the message "you have a gift." Flowwow is a conductor of feelings between people. Anticipation of wonder, warmth, care — these aren't just nice words, they're a technical brief for the atmosphere of every shot.

Delivering that through AI generation is harder than producing a pretty picture. AI defaults to sterile-perfect, surreal, plastic. We needed living, lived-in, warm — where you want to stay. That's why every frame was composed not as a product shot but as an atmospheric scene with a story inside.

The brand's visual metaphor — a flow of feeling between sender and recipient — was embodied through concrete imagery and worked as a visual language of emotion

Work was done in tandem with the brand's in-house designers. Every visual concept was approved by the Flowwow team at each stage — against their guidelines. This isn't a "client said ok" — it's a professional brand check by people who know it from the inside.

Seven banned competitor colors in one brand book. Every frame must be dominated by the brand coral (#FF7663). At the same time, competitor colors are forbidden: yellow (Yandex Eda), pink (Samokat), red and yellow (Yandex Market), blue (Yandex Lavka), green (VkusVill), light blue (OZON), purple (Wildberries). AI doesn't get marketing logic — it can easily paint a background blue and the frame visually works for a competitor. Every result was checked manually against the guidelines.

13 product categories — each with its own visual logic. Live and artificial fir trees, tabletop firs, wreaths, nobilis arrangements, winter bouquets, holiday decor, ornaments, garlands, candles, cakes, cupcakes, gingerbread, gift sets. Each category has its own texture, motion in frame, way of presenting. This required visual literacy, not just technical execution.

Brand objects in every scene. Flowwow's branded boxes and bags, coral ribbons, ceramic mugs, toppers, cards — all of it had to fit organically into every spot. AI doesn't understand proportions or how each object functions in reality.

Hooks were a separate creative and technical task. Each of the 9 holiday spots had 3–5 hook variants in the brief. First you had to pick the one organic to the specific scene without breaking its atmosphere. Then realize it technically — the table sets itself, tea pours itself, the tree grows from the floor, the wreath instantly appears on the door. Each hook requires precise control of start and end animation frames — a fundamentally different task than just generating a scene. Plus brand consistency in every scene and animation.

Platform types with different technical requirements. Yandex / VK / Smart TV, performance networks, social media — each with its own formats, weight limits, requirements for subtitles, disclaimers, logos, safe zones

13 ideas at the start, 442 generations in process. Each spot started as a draft concept and got refined through iterations. First storyboards, then static scenes as generated stills — the client and designers see composition, color, brand elements before video generation begins. Only after approval — the final animation. This creates a clear process and control at every production stage.

Different tools for different tasks. Most AI video attempts fail right here — they take one tool and try to close everything with it. As of late 2025, no single generator handled the full spectrum of tasks at the required level. We built a pipeline from Midjourney, Nanobanana, Kling and several other tools — one is better at holding atmosphere and style, another is more precise with texture and detail, a third is more stable at animating objects

13 unique spots — 10 holiday and 4 image-driven without tying to a news hook.

442 generations in process, final files structured by placement and format, ready to upload to ad accounts.

Three parallel A/B tests in one project: AI vs UGC, AI vs product shoot, with-hook vs without-hook.

Approved by Flowwow's in-house designers. Spots accepted, budgets running in live ad placements.

Cost: 400,000 ₽. Timeline: 1 month. Alternative with classic production: 1.5–3M ₽ and 2–3 months