Summary of "Про креативы без иллюзий: поиск идей, тесты и производство с AI"

Speakers

Dima — host, founder of Along.
Tatyana Shestakova — Head of Creative Production & Marketing Operations at a major Health & Fitness brand; angel investor and advisor in AI/creative startups.

Creative strategy for startups and new apps
Testing and iteration frameworks (what to test, how many creatives, budgets, cadence)
AI-assisted production pipelines and tooling
When to use paid channels vs organic; validating product–market fit with ads
Creator / UGC vs AI-generated talent debate
Forecasts for 2026 (agents, personalization, creative roles)

Do social research: Reddit, TikTok, X, forums, search trends (Google/other engines) to learn language, pain points, and search volume.
Dissect competitors’ creatives — identify messages, visuals, and iterations they made; do not copy 1:1.
Build a reference library (Figma, Notion, FeedBoard or similar) to collect organic content, inspiration, and reusable components.
Combine disparate inspirations (like LEGO): mix visuals, copy and formats to create fresh concepts.
Revisit old formats/cyclical trends — some ad styles resurface successfully.

Start with 5–6 different concepts; if constrained, a minimum of 3.
For each concept create ~3–5 variations.
Let the platform (e.g., Meta) prioritize creatives within concept bundles to surface winners.
Budget guidance:
- $50–$100 per creative minimum for proper testing.
- $10k+ total testing budget gives more reliable early signals.
Cadence and expectations:
- Weekly testing cycles if budget allows.
- Aim for ~20% of tests to “pass” (show promise); keep 2–3% as full winners for scale.
Keep creative testing spend a small share of campaign spend: ~10% advisable; 20–30% acceptable early.

Minimize variables when iterating: change messaging while keeping visuals constant (or change visuals and keep messaging) to isolate drivers.
After isolating the working element, iterate on identity, voice, early seconds (first 3s), or add avatars/social hooks.
Avoid trivial changes (e.g., button color) — they rarely move the needle under modern platform algorithms.
Document the process: idea → test → iterate → document → new batch.

Rule of thumb: automate mass iterations and pre-processing; human assembly/refinement remains needed for high quality.
Typical pipeline:
1. Generate references.
2. Batch-generate visuals/videos via multiple models.
3. Pre-process outputs.
4. Human edit/assemble final creatives (editors or tools like CapCut).
Tools and roles mentioned (names may differ in transcripts):
- Wavy UI / VUI — pipeline orchestration & multi-model generation.
- “Nanban” / “Nanbanana” — identity/visual generation models.
- Hicksfield — platforms for combining models, creating avatars and motion control.
- Video SDKs / agents — generate video creatives; some offer SDKs for integration.
- Perplexity — trend/semantic parsing and social research automation.
- FeedBoard — store organic video/static inspiration.
- Figma, Notion — shared knowledge base and reference libraries.
- CapCut — editing.
Use good prompts to produce many visual variants quickly; reserve humans for final curation.

Performance depends more on message, relevance, format, and targeting than on whether talent is AI or human.
Use AI for low-cost identity/appearance testing; once a winning persona is found, hire real creators (UGC) to match that profile for authenticity and scale.
Expect a resurgence of handcrafted UGC and mixed-media aesthetics — human authenticity will be valued.
In identity-sensitive verticals (e.g., health & fitness), matching model to audience (race, gender, age, body type) matters a lot.

Google / YouTube Shorts: prefer voiceover, subtitles, product demos; users expect human/demo elements.
TikTok: organic style, platform-specific patterns; reach is unpredictable and often delivers lower long-term conversion/subscription quality.
Meta: more systematizable (audience optimization); creative diversity is key to resilience.
Always check platform policies (especially for health and fintech) to avoid disapprovals or account action.
Organic “TikTok farms” can drive reach and installs but often produce low-quality traffic, refunds, or App Store penalties — treat organic viral tactics as experimental supplements, not replacements for controlled paid testing.

Test demand with ads driving to funnels (web funnels, one-page sites, Stripe checkout, analytics) — no need for fake App Store pages.
Use no-code / rapid builds to create fast MVPs and test conversion before building full product.
Use top-of-funnel metrics and funnel completion events to decide whether to invest in product development.
Be mindful of fraud, refunds and platform responses if traffic quality is poor.

Intellectual property: direct copying of competitor creatives can trigger complaints, suspensions or bans. Avoid copying 1:1.
Negotiate with creators; don’t accept first quoted price.
Store and document everything: playbooks, test results, creative knowledge base.
Expect winners to burn out (creative fatigue) — continuous iteration is necessary.

More AI agents will manage production and performance marketing; creative marketers who orchestrate agents will be core roles.
Production costs will continue to fall, enabling more tests and deeper personalization at scale.
Personalization and identity-driven creatives will become more important — expect more AI-driven avatar/identity testing plus selective humanization (UGC).
Growth of creative automation & SDKs that can produce entire video creatives on demand; human curation and quality control remain critical.

Competitor creative dissection: study message, visual, iterations; use competitor language but create distinct visuals.
Test matrix template: 3–6 concepts × 3–5 variations per concept; weekly cycles; $50–$100 per creative minimum.
Iteration playbook: isolate variable (message vs visual) → iterate identity → add social hook / first 3s tweak → document.
Automated creative pipeline: scrape references (Perplexity / subreddit parsing) → generate variations (multi-model pipeline) → human edit/assemble → upload.
MVP validation via ads: run ads to web funnels / one-pagers, track funnel metrics, then build product if signals are positive.

Concepts to start: 5–6 (minimum 3 if constrained).
Variations per concept: 3–5.
Spend per creative test: $50–$100 minimum; $100+ recommended when scaling winners.
Expected pass rate: ~20% showing promise; 2–3% scalable winners.
Creative testing budget as % of campaign budget: start ~10%; 20–30% acceptable early.

Don’t copy creatives 1:1; dissect and speak the user’s language.
Store references centrally (Figma / Notion / FeedBoard) and formalize a playbook.
Use AI to accelerate iterations and test identities; convert best-performing AI personas to real creators for authenticity.
Prefer controlled paid experiments (Meta, Google Search/YouTube, ASO) for reliable validation; use TikTok/organic as supplementary channels but expect lower-quality conversions.
Protect against IP risk and negotiate with creators.

Dima — host, founder of Along.
Tatyana Shestakova — Head of Creative Production & Marketing Operations (major Health & Fitness brand); angel investor and advisor.