Batch AI Creatives — UGC Ad Testing & Prompt Template Library

Why Batch Beats One-Off Prompting

First posts look great; by week three style drifts and output slows. Performance accounts feel it first: the same “winning” static runs until CTR collapses (creative fatigue). Batch production fixes that with templates + AI Chat + AI Image + score—the same loop as Optimize Then Generate, scaled across a content calendar.

Benefits:

• Faster weekly output (5–20 assets per session)
• Visual consistency for brand and ad accounts
• Lower cost per creative vs bespoke prompting every time
• Use batch text-to-image for cheap screening, then let CTR decide what graduates to 2K

Agentic batch loop: template × variant × score

Compress the four layers into an operator-ready cycle:

Theme (weekly mood) → Templates (5–10 bases) → Chat variants (3 per template)
→ 1K screen → 2K finals + voice → publish & score → promote gold templates

Same shape as an agentic marketing workflow: Chat plans, Image/Voice produce, you curate. Humans touch CTA copy and compliance on the top 30%—not every image from scratch.

Minimum metrics set

Retire bottom 20% of templates monthly to limit creative fatigue across the account. If the same frame runs two weeks and CTR slides, swap the template before you raise budget.

The Batch System (4 Layers)

Layer 1 — Campaign theme

Define one theme per week or launch:

• Product drop / SKU focus
• Seasonal (back to school, holiday gifting)
• Evergreen pain point ("sleep", "hydration", "productivity")

All templates inherit the same mood keywords so feeds feel cohesive.

Layer 2 — Prompt template library (5–10 bases)

Each template is a fill-in-the-blank optimized prompt stored in Prompt Library, archived with a naming convention (see below):

Gold prompt templates (fill-in-the-blank)

These five bases cover the most common social hooks. Swap only bracketed fields—keep structure fixed so A/B ad testing stays comparable.

1. Studio premium

[PRODUCT] on a clean studio surface, soft side light with a subtle rim light,
premium commercial photography look, seamless light-gray or off-white backdrop,
product centered and sharp, packaging and material detail readable,
[MOOD] mood, 9:16 vertical, no text, no watermark, no face in frame.

2. Ingredient / material close-up

Extreme close-up of [INGREDIENT OR MATERIAL] from [PRODUCT], macro photography,
shallow depth of field, real texture and micro-detail visible, soft natural light,
background blurred into [COLOR TONE], scroll-stopping composition,
[ASPECT RATIO], no text in image.

3. Meme-native

[PRODUCT] in [everyday absurd or contrast situation], casual phone-snapshot UGC look,
light noise and natural exposure, feels like a real social feed photo before captions,
[MOOD] relatable or ironic vibe, 9:16, product recognizable, no text in image.

4. Before / after glow

Same framing, split left-right or top-bottom: [PAIN STATE] before [PRODUCT]
versus [IMPROVED STATE] after, matching light and camera angle,
believable and not over-retouched, [MOOD] mood, [ASPECT RATIO],
no text and no arrow overlays in the image.

5. Lifestyle scene

[PRODUCT] naturally placed in [LIFESTYLE SETTING], authentic ambient light,
lived-in atmosphere, product sharp in the foreground, soft background clutter that does not compete,
[MOOD] mood, [ASPECT RATIO], no face or hands-only, no text in image.

For a UGC handheld variant, append to the lifestyle template: UGC handheld phone photo style, natural window light.

Layer 3 — Optimize and variant generation

For each template:

1. AI Chat — paste template + SKU details → 3 style variants (0.5 credits each)
2. AI Image — cover + 2 alternates at 1K (nano-banana-fast) for screening; promote winners to 2K (nano-banana-2)
3. AI Voice — 10–20s ad reads from best stills—see AI Voice Workflow

Target 3–5 usable assets per template before moving to the next.

Channel notes (Meta Ads vs TikTok vs organic)

Ship compliance mains (white background) and ad lifestyle variants from separate prompts—mixing them in one Ads Manager campaign often confuses learning. Log template ID when you upload so fatigue analysis maps back to Prompt Library entries.

Supported image models (only):

Credits: Chat 0.5 / reply; Image 1K = 3, 2K = 4, 4K = 8; Voice ~1 credit / 500 characters.

Credit planning example (one template, one SKU):

Adjust counts to your plan; draft at 1K, finalize only winners. Do not default to 4K before CTR data.

Layer 4 — Score and recycle

Track per asset:

• Hook retention (short-form with voice)
• CTR (static ads)
• Saves/shares (organic)

Promote top prompts to "gold templates". Retire bottom 20% monthly to fight creative fatigue.

Prompt Library naming convention

Batch ops need searchable IDs—not "new template 3". Use:

sm_[style]_[angle]_[version]

Examples:

Suggested tags:

Note platform, safe-zone rules, and last CTR in the library entry. New teammates pull by ID instead of copying prompts from chat history.

Platform pixels & safe zones

Lock aspect ratio before export; keep text and logos clear of edges and system UI.

For faceless shorts, draft VO in AI Chat, pair product stills from AI Image with AI Voice, then assemble captions in your editor—ideal for accounts that need a steady Reels calendar without on-camera talent.

UGC Ad Batch Pattern

Batch UGC-style ads like this:

1. AI Chat — define 3 persona angles (skeptic, enthusiast, busy parent)
2. Generate one image prompt per persona—handheld light, kitchen/bathroom setting (reuse sm_ugc_handheld_v3)
3. AI Voice — 15–30s read per persona (~35–75 words each)
4. Export variants with different opening hooks (same body prompt)

Reuse Ecommerce Image Optimization prompts for hero product stills first, then drop them into social templates for UGC ad testing.

End-to-end weekly case: 3 SKUs × 5 templates

Run one full week of batch AI creatives + A/B testing on a skincare line (serum / SPF / gift set):

Output & credit budget

Budget control: 3× 1K per template first; only promote "shippable" frames to 2K.

Schedule

CTR decision rules (copy as-is)

1. Promote to gold: 7-day CTR ≥ account average × 1.2 and retention/saves hold → tag gold, scale next week
2. Hold & retest: CTR within ±10% of average → swap hook copy only; keep visual structure
3. Retire or rebuild: CTR ＜ average × 0.8, or two-week decline → tag retired, pull from primary rotation

Templates that usually get promoted

Ingredient close-up and before/after glow often win feed CTR; meme-native is volatile but strong for organic resonance; studio premium is steadier with mid CTR—keep one version for brand and one for ads. Trust your account data over taste.

Weekly Calendar Example (compact)

Anti-Patterns

• 50 unique prompts with no shared structure—style drift guaranteed
• Skipping AI Chat refinement to "save time"—re-roll costs more credits
• Publishing every variant—curate top 30% only
• Finalizing at 4K before CTR data—waste credits on losers
• Running one template too long—creative fatigue erodes early wins

FAQ

How many templates do I need to start?
Five strong templates beat twenty weak ones. The five fill-in bases above are a complete starter set.

Images only or voice too?
Both. Batch stills for feed and TikTok covers, then add AI Voice for ads and faceless shorts.

Can teams share templates?
Store sm_* IDs in Prompt Library with platform / mood / SKU tags, or maintain a shared doc with fields; link these guides for onboarding.

When should I use Nano Banana 2?
Screen with nano-banana-fast; after composition and offer clarity check out, finalize with nano-banana-2 (and cl / 4k-cl / pro variants)—do not spend credits on unvalidated creatives.

Related Guides

• Optimize Prompt Then Generate
• AI Voice Workflow
• AI Poster Design Workflow