Which AI model should I use to write my Black Friday email sequence?
I've tested this exact scenario three times now, most recently in October when I helped a Shopify store owner draft a five-email Black Friday sequence. She sells handmade leather bags—totes, crossbody styles, wallets—price range $65 to $180. She had a list of 4,200 subscribers and wanted to send one email per day from the Monday before Thanksgiving through Cyber Monday.
Claude Sonnet 4.5 wrote the best emails. I gave it her brand voice guidelines (a two-page doc she'd used for her last copywriter), three past emails that performed well, and a prompt that said "write five emails for a Black Friday sale, 25 per cent off everything, free shipping over $100, each email should give a different reason to buy and none should sound desperate". It came back with subject lines, preview text, body copy and a P.S. for each one. The tone was warm, specific, and confident. I didn't have to rewrite anything except one joke that didn't land.
Gemini Flash is faster and cheaper—about one-third the cost per token—but the output felt generic. It used phrases like "don't miss out" and "limited time only" in every email, and the subject lines were interchangeable. You'd spend the savings on editing time.
DeepSeek V3 gave me the most interesting ideas—one email was structured as a packing list for a weekend trip, another as a gift-giving decision tree—but the execution was uneven. Some sentences were too clever, some too flat. If you've got a strong editorial eye and you want creative angles, it's worth trying. If you need something you can send with minimal edits, stick with Claude.
For a five-email sequence at around 300 words each, you're looking at roughly $0.40 to $0.60 in API costs with Claude Sonnet 4.5 on Kryotta. That's nothing compared to a copywriter's day rate, and you can regenerate sections until the voice is right.
Can AI actually generate product images that won't get my Instagram ads rejected?
Yes, but you need to know what triggers the review filters. I've made this mistake more than once.
Stable Image Core and SD 3.5 Large both produce clean, high-resolution images that look professional. The problem is Meta's ad review system flags anything with distorted hands, faces that sit in the uncanny valley, text that's slightly warped, or backgrounds with spatial inconsistencies. If you're generating lifestyle shots—product on a table, product held by a model, product in a styled room—you'll hit those issues constantly.
What works: isolated product shots on a solid or gradient background, flat lays from directly overhead, close-ups with shallow depth of field where the background is intentionally blurred into abstraction. I made twenty-four images for a candle brand in September using Stable Image Core on Kryotta—white candles on a pale grey surface, shot from above, soft shadows—and all twenty-four cleared review on the first try. No hands, no faces, no architecture.
What doesn't: any image with a human figure, even if it's blurred or cropped. Any background with windows, doorways, or furniture where the perspective looks slightly wrong. Any image with text overlays generated by the model itself.
If you need lifestyle shots with models, hire a photographer or use stock. If you need fifty variations of your product on different surfaces in different light, AI will save you a week and $800. One image costs about $0.04 to generate with Stable Image Core. You'll make ten to get three keepers, so budget $0.40 per final asset.
How much will it cost me to automate my entire Cyber Monday workflow with AI?
I'm going to give you real numbers based on a mid-sized Shopify store doing about $40,000 in revenue over the five-day Black Friday to Cyber Monday stretch. This is what you'd actually use AI for, and what it costs.
Email sequence (five emails, 300 words each, written with Claude Sonnet 4.5): $0.50
SMS messages (three short texts, 50 words each, written with Claude Haiku 4.5): $0.05
Product images (twenty variations for ads, Stable Image Core): $8.00
Social captions (fifteen Instagram and TikTok captions, 100 words each, Claude Sonnet 4.5): $0.30
Ad copy (ten Facebook ad variations, 75 words each, Claude Sonnet 4.5): $0.25
Video ads (three 5-second product spin clips, Luma Ray 2): $3.00
Total: about $12.10 in AI costs. You'll spend more time than money. Writing the prompts, reviewing the output, editing what doesn't work, uploading assets, setting up campaigns—that's the real cost. Budget two full days if you're doing this yourself, or four to six hours if you're delegating the execution and just handling creative direction.
The alternative is hiring it out. A freelance copywriter charges $500 to $1,200 for a Black Friday email sequence. A product photographer charges $150 to $400 for a half-day shoot. A video editor charges $75 to $150 per finished ad. You'll spend $1,500 to $3,000 to get the same assets, and you'll wait longer.
Should I start with email or product images when I'm prepping for Black Friday?
Email. Always email.
Product images matter, but if your list is warm and your offer is clear, people will buy even if your photos are just decent. I've seen Etsy sellers do $8,000 in Black Friday revenue with iPhone shots on a white tablecloth. I've never seen anyone do that with beautiful images and a poorly written email.
Start by writing your sequence. Five emails is the standard: one announcement (Monday before Thanksgiving), one reminder (Wednesday), one "sale is live" (Black Friday morning), one "still time" (Saturday), one "last chance" (Cyber Monday). Each email should give a different reason to buy—gift ideas, bestsellers, new arrivals, bundles, final hours—and the tone should shift from excited to urgent without ever sounding panicked.
Once the emails are drafted, you'll know what images you need. If email three talks about your bestselling tote bag in camel and black, generate those colourways. If email four pitches a gift set, shoot or generate that bundle. Don't make twenty-four product images and then try to write emails around them. It doesn't work.
The other reason to start with email: you can test the copy before Black Friday. Send a preview to a small segment of your list in early November and see what happens. If the open rate is low, rewrite the subject lines. If the click rate is weak, sharpen the call to action. You can't test product images the same way—they either work in the ad platform or they don't, and you won't know until you spend money.
Which AI video tool should I use for TikTok Shop ads, and will they actually convert?
I've used both Luma Ray 2 and Nova Reel for short product ads. Luma gives you more control and better quality. Nova Reel is faster and cheaper.
For TikTok Shop, you want five to seven seconds of product movement—a slow spin, a close-up pan, a hand placing the product on a surface—with no cuts and no text baked into the video. TikTok's algorithm favours native-looking content, which means anything that looks like a traditional ad gets suppressed. AI video tools can make clips that feel organic if you keep the prompt simple.
Luma Ray 2 works well for this. I made three ads for a small brand selling enamel pins in October. The prompt was "slow 360-degree rotation of a round enamel pin on a wooden surface, natural light, shallow depth of field, no hands, no text". Each video took about four minutes to generate and cost $1.00. Two of the three were usable. The third had a weird flicker in the background that I couldn't fix, so I scrapped it.
Nova Reel is faster—under two minutes per video—and costs about $0.40, but the motion is less smooth and the lighting can look flat. If you're making ten variations to test, Nova Reel makes sense. If you're making three hero ads, use Luma.
Will they convert? I don't have a clean answer. The pin brand ran all three videos as TikTok Shop ads with a $200 total budget and got a 2.1 per cent conversion rate, which is decent but not exceptional. The client said the AI videos performed about the same as the iPhone videos she'd shot herself. The advantage of AI isn't that it converts better—it's that you can make more variations in less time, test faster, and find what works without spending a day on a shoot.
What's the one Black Friday task I should automate with AI this year if I only have time for one?
Customer service replies during the sale. Not product images, not emails, not ads—replies.
Black Friday weekend is when your inbox and DMs explode. People ask about shipping cutoffs, return policies, gift wrapping, size charts, whether the sale applies to pre-orders, whether they can use two discount codes, whether you ship to APO addresses. Most of these questions have the same answers, but you can't ignore them because an unanswered question is a lost sale.
I set up a simple system for a Shopify store owner last November using Claude Haiku 4.5 on Kryotta. We made a one-page doc with answers to the fifteen most common questions—shipping times, return window, how the discount code works, product care instructions—and I wrote a prompt that said "you are a customer service rep for [brand name]. Answer this question using the information below. Be helpful and warm. Keep it under 75 words. If the question isn't covered, say you'll check and get back to them within two hours". She copied every email and Instagram DM into Kryotta, pasted the response, tweaked it if needed, and sent it. She answered 160 messages over four days and spent maybe three hours total.
Claude Haiku 4.5 costs almost nothing—around $0.01 per reply—and it's fast. You're not automating the sending, you're automating the drafting, which means you stay in control but you don't burn out writing the same answer forty times. If you've got a VA or a team member handling customer service, give them this system. If you're doing it yourself, it's the difference between spending Black Friday weekend chained to your inbox and actually sleeping.
Questions people ask
Can I use AI-generated images in Facebook ads without disclosing they're AI? Meta doesn't require disclosure for product images, but if your ad gets flagged in review, being upfront in your appeal can help. The bigger risk is rejection for quality issues—distorted details, weird shadows—not for being AI-made.
How do I know if my Black Friday email copy sounds like AI? Read it out loud. If you hit three sentences in a row that start the same way, or if you see phrases like "don't miss out" and "limited time" in every paragraph, rewrite. Real people vary their rhythm and don't repeat themselves.
Should I generate all my product images at once or make them as I need them? Make them as you need them. You'll waste time and money generating images you never use. Write your emails and ads first, then generate only the images those campaigns require.
What happens if TikTok rejects my AI-generated video ad? You get a reason code—usually "low quality" or "misleading content"—and you can appeal or upload a new version. I've had better luck uploading a completely different video than arguing with the review team.
I've spent the last two months helping small brands and freelancers figure out what AI can actually do for Black Friday prep, and the answer is: quite a lot, as long as you're realistic about what it's good at. It won't replace your creative judgement, but it'll handle the repetitive work that eats your time. If you want to test any of the workflows I've described here, Kryotta gives you access to Claude, Gemini, Stability AI, Luma and the rest in one workspace. Start a free account and see what fits your workflow.
