Tạo hàng loạt
8 ảnh nhất quán cho chiến dịch.
In AI Image 2, “batch” is not pressing send repeatedly—it is keeping character, style, and layout logic coherent under one brief. AI Chat Thinking mode publicly supports up to roughly 8 related images per request (UI may vary)—ideal for poster series, comic panels, and e-commerce scene sets.
This tutorial stays entirely inside AI Chat chat—no coding APIs.
When batching is worth it
| Scenario | Count | Value |
|---|---|---|
| Social weekly schedule | 5–8 | Unified visual language |
| Comics / storyboards | 4–8 | Character consistency |
| E-commerce SKU scenes | 6–10 | One product, many lifestyles |
| A/B hero visuals | 3–4 | Same copy, different palettes |
If images are unrelated, generate singles—no forced batching.
Prep: one “series brief sheet”
| # | Use | Ratio | Copy diff | Notes |
|---|---|---|---|---|
| 1 | Feed header | 16:9 | Slogan A | |
| 2 | Square post | 1:1 | No text | |
| … |
Reference this sheet inside the batch prompt so the model plans rhythm.
Core prompt template (series campaign)
Enable Thinking mode. Generate 6 coherent social images for "Summer Iced Drink" campaign, unified 4:5 aspect.
[Brand] Brand colors #00A8E8 and white, fresh realistic photography, no garbled text.
[Series plan]
1. Product close-up with ice bubbles
2. Beach handheld drink
3. Friends toasting
4. Convenience store cooler display
5. Subtitle-style slogan image, copy "One Sip of Summer"
6. Promo end card, copy "Second cup half price"
[Consistency] Same glass cup shape, same liquid color; if people appear, same short-haired East Asian woman.
Label each image bottom with tiny index 1/6–6/6 for selection.After sending, do not interrupt—canceling during Thinking wastes a long wait.
Storyboard / comic batch
Thinking mode, Japanese manga 4-panel vertical, story "late for the subway",
same male lead: black hair, glasses, blue shirt, khaki pants.
Dialogue under each panel (4 lines in quotes),
clear lines, consistent outfit between panels, progressive expressions.If panel 3 changes the face, use region edit on that panel only, or follow-up: “Regenerate only panel 3, character locked to…”
E-commerce: one product, many scenes
Thinking mode, white wireless earbuds, generate 8 lifestyle 1:1 images:
office desk, commute subway, gym, café, outdoor run,
airplane tray, rainy window, minimal white hero.
Consistent product form, natural light, commercial photography.After selection, follow up on weak scenes only—do not rerun the whole set.
Post-batch curation (10 minutes)
- First cut: delete garbled text, extra fingers, product distortion.
- Second pairing: map numbers 1–6 into channel slots.
- Unified post: add logo, legal lines, real prices in PS/Canva (never ship AI numbers raw).
- Archive: folder name
2026-618-campaign-v2-approved.
Speed tips
| Tip | Detail |
|---|---|
| Brand block first | 80% of series text repeats—copy-paste brand block |
| Instant then Thinking | Draft composition in Instant; finalize in Thinking to save quota |
| Split batches | Over 8 images: second prompt starts with “Continue previous batch character and palette” |
| Reference upload | Upload approved hero: “Batch scenes using this product appearance” |
Common failures and fixes
| Symptom | Cause | Fix |
|---|---|---|
| 8 images style jump | Brief too scattered | Add “same photographer, same grade” |
| Only first 3 good | Model fatigue | Split into two batches |
| Text degrades per image | Copy too long | Only 1–2 images with text, rest text-free |
| Character not same person | Vague description | Specify hair, outfit, accessories |
Cost mindset vs singles
Batch waits longer once, but total rounds are fewer than “20 single retries.” Real ops cost is often curation and copy edits—batching concentrates effort on curation, not luck.
Next steps
- Image editing — fix one image in a series
- Social posters — multi-platform size extension
- Prompt engineering — strengthen BRICK template
When you reliably deliver 6 usable, style-unified images, you have small-campaign AI production capability.