B

ecom | saas | @imdialed

[B @JuggMotion]

@nazloveslife open dms pls

[B @JuggMotion]

Hiring VAs

Need to be skilled editing videos, generating/cleaning voiceovers, and image -> video generation

Will pay for API credits + train you on our internal workflow.

Comment below.

[B @JuggMotion]

@mikefutia clone

[B @JuggMotion]

@cartiernoahh peep messages

[B @JuggMotion]

If you don’t like reading

https://t.co/ZOoB5p0VhH

[B @JuggMotion]

Time to expose how to build off that to make AI generated content

any tool, anytime, all inside of Claude.

Go to https://t.co/cuSxgKbAqU and sign up. It’s an API aggregator that gives you one billing account across Seedance 2.0, Sora 2 Pro, Veo, Wan, Kling — all the top video models. No monthly subscription. Runs on credits. Way cheaper than running each model’s API directly or paying Higgs Field.

Inside https://t.co/uQI8LsCT5y → API Keys → Create New Key. Name it per brand the same way you did the Google one.

Copy the key. Don’t paste it anywhere shared.

Tell Claude to build the skill.

In a fresh chat, give it the skill creator.
Paste your https://t.co/uQI8LsCT5y key. Tell Claude:

“Build a skill that runs image-to-video generation through https://t.co/uQI8LsCT5y’s Seedance 2.0 model. The images will be generated using our imagen skill. Handle errors, retry failed generations, support first-frame and last-frame inputs, stitch the output clips together.”

Save it as a skill. “Save this as a skill named seedance-image-to-video.”
Now you have video generation in your library, scoped per project, available across every chat.

Text-to-video is unpredictable. Same prompt twice gives you different actors, different framing. Useless for ad consistency.

Image-to-video gives you control. You generate the still images first using the imagen skill from the last tweet. Each generated image is a keyframe.

Seedance animates between them. Actor locked. Framing locked. Product packaging locked. You control every frame, the AI just animates the in-between.

The killer feature: first-frame + last-frame chaining.

Seedance accepts a first frame URL and a last frame URL per clip. Which means clip A’s last frame = clip B’s first frame. Smooth transitions. No jump cuts. The actor doesn’t teleport across the cut.

Without this, multi-clip AI video looks like a slideshow. With it, it looks like an actual ad.

Here’s workflow to get going:
https://t.co/OmNXSzkQeL a winning source ad. Twitter bookmarks, BrandSearch, swipe files. Save the .mp4.

2.Shot-by-shot analysis. Drop the .mp4 into Claude. Have it use the Gemini API to transcript at 0.5-second intervals. Extract the structural DNA — hook architecture, animation style, pacing, voiceover script.

3.Adapt to your avatar. Pull from your Claude project knowledge (Schwartz, DTC masterclass, band patterns). Generate 5 shot-by-shot concepts. Pick the strongest.

4.Script iteration. Kill the AI voice. Specific numbers instead of “most.” Phrases your avatar would actually say to herself. Read every line out loud — if it sounds like a brand commercial, rewrite it.

5.Generate keyframes via the imagen skill from the last tweet. Character lock (“Linda — pale Irish skin tone, late 30s, expressive eyes, claymation style”) referenced in every keyframe so she’s recognizable across the entire video. Real product packshot dropped in as reference so AI doesn’t hallucinate the label. ~30 keyframes for a 70-second video.

6.Generate clips via Seedance. First frame + last frame + motion prompt per clip. ~$1.80 per clip. Claude shows the cost preview before it runs.

7.Stitch + review. Claude auto-concatenates the clips. You review for character continuity drift, frame artifacts, transition errors. Reroll what broke.

https://t.co/v2GTlH6LCD layer. Don’t trust Seedance for voiceover — the visual model is solid, the voice synthesis is iffy. Re-run the clips with audio: false. Take the muted video to ElevenLabs. Generate the voiceover at real-creator quality. Stitch the audio over the video.

Jugg.

[B @JuggMotion]

@finityx 10/10 until the p&l comes back

[B @JuggMotion]

If a customer reaches out and you don't respond, you've started a clock. That clock ends in one of three ways:

1. They forget about it (rare)
2. They request a refund (common)
3. They file a chargeback (worst case)

The chargeback is the worst. Chargebacks cost more than the refund itself — you lose the sale, you pay processor fees, AND your chargeback rate increases.

We run heavy MRR, have 13 day average shipping, and run fairly aggro DR ads. Here's how we run customer support to still not get chargebacks.

Even if the response isn't positive — "yeah, I'm sorry your order is delayed, it'll be there in 3-4 business days" — it's a response. Customers need to feel spoken to.

Here's the analogy I drill into my team. Think about your girl. Would you rather her ghost you or argue with you? Neither's great. But if she's arguing with you, communicating with you, there's still some thread — respect, reciprocity, hope. You know she's not out doing something else with somebody else. If she's ghosting you, you have no idea what she's doing.

Customer support is the same. Treat them like a relationship because that's what it is. They're the bloodline.

Here's how I hire CS reps:

Before the SOP matters, you need the person.
Where I find them:

Upwork (my default — fast turnaround, lots of options)

https://t.co/zdAytbyDIz (cheaper rates, more vetting needed)

Your network — even hiring a friend or sister works for this role. It's not that hard.

What to look for:
Responds quickly (test this in the application phase — if they take a day to reply to your message, they'll take 2 days to reply to your customers)

Kind energy in their writing

Good English — non-negotiable

Experience with other DTC brands is a plus — they've seen similar SOPs and won't need every micro-detail explained

When to hire: the moment you start dreading checking your support inbox. If you're spending 30+ minutes a day on emails, hire. The compression on your brain is costing you more revenue than the $100–$200/week the hire costs.

When to scale beyond one CS person: once you're past roughly 50 tickets/day. Promote your most experienced one to a managerial role handling past tickets and training, then bring on new inbound handlers underneath them. Customer support is a department, not a single hire.

Here's high level structure of my SOP:
The full document is 9 sections. The framework matters more than the wall of text:

Mission and Loyalty Rules — the philosophical north star (covered below)

Tools and Workflow — help desk, tag taxonomy, snooze rules, store admin permissions, subscription management, supplier fulfillment channel

Daily Routine — the exact morning sequence (payouts → disputes → help desk → supplier channel → ads platform → comments)

Refund Saver Decision Tree — how to refund without bleeding money, soft-door framework for buyback

Chargeback Response — how to win every chargeback, threat response, withdraw pursuit script, evidence packet structure

Comment Management by Page Type — different page types get different response styles

Common Scenarios — pre-built replies for every recurring question (where is the product made, ingredients, shipping, returns)

Escalation Protocols — what gets escalated to founder immediately, what waits, how to ping so it actually gets read

Daily KPI Form — what the CS team submits at end of shift

Quick word on Section 8 — escalations. There's one item that gets escalated immediately, no exception: a customer mentioning self-harm, suicide, or mental health concerns. The CS team has a specific protocol for these conversations and the founder is looped in immediately so it gets handled as humans, not as a brand. This isn't optional and it doesn't go in the batch queue. It's the most serious section of the SOP and the one I take most seriously.

Other immediate escalations: serious physical reactions, BBB / FTC / viral social threats, multiple customers reporting issues with the same batch, negative comments that could expose brand-level issues, any new chargeback that appears.

Here's move important parts of it:
1. The customer-first loyalty rule (with 3 exceptions)
The CS SOP literally says: "Your loyalty is to the customer, above the business."

I put this in writing because of myself. When money's coming in, it's hard to keep the focus on the customer.

The natural pull is toward "what's good for the business." But your CS team is in the trenches with customers all day — they need a clear north star.

There are exactly three exceptions where you protect the business over the customer:
- Suspected fraud
- Repeat refund requests on the same profile
- Anything requiring founder approval

Other than those three: customer first, always.

This rule has saved me more money than any other policy. CS teams that operate from "customer first" prevent way more chargebacks than CS teams that operate from "protect the business."

2. The chargeback math + win-rate bonus
Chargebacks cost roughly $25–$30 per dispute filed (varies by platform). They also raise your chargeback rate, which you cannot afford to let slip past 0.3%.

Your CS team's job is to win every chargeback.

To incentivize this, the SOP includes a bonus structure: $25 per won chargeback. The math:

If they win, I'm out $25 (the bonus)

If they lose, I'm out the original transaction + processor fees + chargeback rate damage

Net: paying $25 for a win is a steal. The bonus IS the cost of the win, plus it incentivizes the CS team to hustle to win them.

3. The threat response philosophy

When a customer threatens to dispute with their bank — DON'T fight them. Give them the money.

The SOP literally says: "If they threaten to dispute with their bank, refund immediately."

This is counterintuitive to most operators who feel "they shouldn't get away with this." Doesn't matter.

The money's already gone — the only question is whether you take additional damage on top. Take the L, prevent the worse L.

4. The withdraw pursuit framework

When a chargeback IS already filed, your CS team's job is to convince the customer to withdraw it.

The framework:
- Acknowledge the dispute (no defensiveness)
- Confirm a full refund has been issued
- Explain why withdrawing helps both parties
- Add an incentive — gift card to their account once withdrawn

The gift card incentive is the multiplier. Most customers won't take the time to withdraw without a reason. The gift card gives them one. We're testing this script right now — out of three pursuits last week, two customers withdrew. The other one hasn't responded yet, so I think we're just gonna call them lol.

If gift cards stop converting, the next test is physical gifts — flowers, branded merch, whatever drives the withdraw rate up. The cost of the gift is always less than the cost of the chargeback.

5. Page-type comment response templates

If you run advertorials or third-party-style ad pages, comment management is critical. Comments visible on your ads either reinforce the angle or break it.

The SOP has different response templates for:
- Brand pages (direct product, brand voice)
- Editorial pages (third-party feel, neutral)
- Personal pages (founder-style, intimate)
- Expert pages (authority voice)

If your CS team responds to a comment on an editorial page using brand-page voice, the page's authenticity gets blown. CTR drops. Conversion drops.

This single framework — different response styles by page type — has measurable impact on advertorial performance. The exact templates per page type are in the gatekept version.

6. Marketing-savvy responses to brand-vulnerable questions

Some questions hit harder than others. The classic example for supplement brands: "Where is your product made?"

If components come from a region customers react to negatively — China, for instance — the SOP doesn't tell the CS team to dodge or lie. It pivots.

Instead of leading with country-of-origin, the script highlights:
- Facility certifications (FDA cGMP, ISO, etc.)
- Batch testing protocols
- Ingredient-level sourcing ("the creatine is sourced from X, the magnesium from Y, the L-theanine from Z")

The customer gets a real, specific answer that builds trust without pointing at a single country that triggers a knee-jerk negative reaction. Be smart. You're still a marketer.

The SOP handles dozens of these brand-vulnerable common questions — quality concerns, manufacturing details, ingredient sourcing, pricing pushback.

Here's what I currently use as my tech-stack, kinda just because I'm too lazy to swap and its been doing well. Any suggestions lmk might swap eventually.

Help desk: Reamaze (recommended — flexible, paid). Alternatives: Brandify, Brandwise, Gmail + a Google Sheet (what I used for the longest time)

Comment management: Meta business suite (did do CommentGuard at first)

Subscription management: Loop (what we use).

Tracking app: Parcel Panel/Will

Chargeback automation: Chargeflow
Alternative: Disputely, Disputifier, etc.

Knowledge base: Custom GPT trained on the SOP + brand-specific info, so the CS team can ask the bot before pinging me

Jugg

[B @JuggMotion]

aight iceman was fire

back to tweeting

[B @JuggMotion]

2 day hiatus while I process these albums

[B @JuggMotion]

figured it out.

[B @JuggMotion]

How do I connect these mega glasses to Claude I need to launch more ads

[B @JuggMotion]

How do I connect these mega glasses to Claude I need to launch more ads

[B @JuggMotion]

For the guys who don't feel like reading

https://t.co/wS3LVztLYd

[B @JuggMotion]

https://t.co/swBU4CJyVF

[B @JuggMotion]

Just got off a call with someone doing 200 orders/day

About to run completely out of stock

Because they ordered 5k units, thinking it would last a month

It won't even last 2 weeks

And I see this shit constantly

The math everyone fucks up

You're not tracking the right number

"I do 200 orders a day" doesn't tell you anything about inventory

What matters is units leaving your warehouse per day

Pull your Shopify reports right now

- Look at items per order
- Multiple that by orders a day
- Then forecast on this if you were to grow at a reasonable amount over the next 14-20 days

When you start moving over to actually placing stock, production times, 3pl, etc.

You place the order thinking "great, I'm covered"

It may take them 2 ½ weeks before it arrives at your 3PL

Another 10 days before any of it actually arrives to customers

Then if you have a good supplier they’ll have materials ready for you, so that 2 ½ weeks may cut down to 4-5 days or so (this is all in China)

During those 22 days you sold 100% of your OLD inventory

So when your new 5k order is ready to ship, you actually have 0 units left

This is why brands randomly go "out of stock" when their ads are finally working

They did the math wrong 2 weeks ago and don't realize it until it's too late

I've personally lost everything because I didn’t take into account forecasting of inventory and the cycle of it all

So make sure you’re taking the math serious so this product that is finally starting to scale, can continue to scale 10 months from now and youre trustpilot isn’t a warzone

[B @JuggMotion]

on the beach

drinkin a miami vice

larpin like im working

really im just slavin claude

jugg

[B @JuggMotion]

my speed at which I can transmute an idea into action with my growing m army of VAs is directly correlated to how much I spam on twitter

[B @JuggMotion]

wake up to five star reviews on product

be jumping around in my underwear n shi

big ass smile on my face

like damn yall rlly love me huh

[B @JuggMotion]

everyone in my city think I’m a drug dealer but my product is not FDA approved yet so idk what they talkin abt

[B @JuggMotion]

https://t.co/4pbLy4dTHI

Full video of me doing this inside

[B @JuggMotion]

Time to expose how to build off that to make AI generated content

any tool, anytime, all inside of Claude.

Go to https://t.co/cuSxgKbAqU and sign up. It’s an API aggregator that gives you one billing account across Seedance 2.0, Sora 2 Pro, Veo, Wan, Kling — all the top video models. No monthly subscription. Runs on credits. Way cheaper than running each model’s API directly or paying Higgs Field.

Inside https://t.co/uQI8LsCT5y → API Keys → Create New Key. Name it per brand the same way you did the Google one.

Copy the key. Don’t paste it anywhere shared.

Tell Claude to build the skill.

In a fresh chat, give it the skill creator.
Paste your https://t.co/uQI8LsCT5y key. Tell Claude:

“Build a skill that runs image-to-video generation through https://t.co/uQI8LsCT5y’s Seedance 2.0 model. The images will be generated using our imagen skill. Handle errors, retry failed generations, support first-frame and last-frame inputs, stitch the output clips together.”

Save it as a skill. “Save this as a skill named seedance-image-to-video.”
Now you have video generation in your library, scoped per project, available across every chat.

Text-to-video is unpredictable. Same prompt twice gives you different actors, different framing. Useless for ad consistency.

Image-to-video gives you control. You generate the still images first using the imagen skill from the last tweet. Each generated image is a keyframe.

Seedance animates between them. Actor locked. Framing locked. Product packaging locked. You control every frame, the AI just animates the in-between.

The killer feature: first-frame + last-frame chaining.

Seedance accepts a first frame URL and a last frame URL per clip. Which means clip A’s last frame = clip B’s first frame. Smooth transitions. No jump cuts. The actor doesn’t teleport across the cut.

Without this, multi-clip AI video looks like a slideshow. With it, it looks like an actual ad.

Here’s workflow to get going:
https://t.co/OmNXSzkQeL a winning source ad. Twitter bookmarks, BrandSearch, swipe files. Save the .mp4.

2.Shot-by-shot analysis. Drop the .mp4 into Claude. Have it use the Gemini API to transcript at 0.5-second intervals. Extract the structural DNA — hook architecture, animation style, pacing, voiceover script.

3.Adapt to your avatar. Pull from your Claude project knowledge (Schwartz, DTC masterclass, band patterns). Generate 5 shot-by-shot concepts. Pick the strongest.

4.Script iteration. Kill the AI voice. Specific numbers instead of “most.” Phrases your avatar would actually say to herself. Read every line out loud — if it sounds like a brand commercial, rewrite it.

5.Generate keyframes via the imagen skill from the last tweet. Character lock (“Linda — pale Irish skin tone, late 30s, expressive eyes, claymation style”) referenced in every keyframe so she’s recognizable across the entire video. Real product packshot dropped in as reference so AI doesn’t hallucinate the label. ~30 keyframes for a 70-second video.

6.Generate clips via Seedance. First frame + last frame + motion prompt per clip. ~$1.80 per clip. Claude shows the cost preview before it runs.

7.Stitch + review. Claude auto-concatenates the clips. You review for character continuity drift, frame artifacts, transition errors. Reroll what broke.

https://t.co/v2GTlH6LCD layer. Don’t trust Seedance for voiceover — the visual model is solid, the voice synthesis is iffy. Re-run the clips with audio: false. Take the muted video to ElevenLabs. Generate the voiceover at real-creator quality. Stitch the audio over the video.

Jugg.

[B @JuggMotion]

Go to Google AI Studio. Scroll to the bottom of the homepage.

Click Get API Key → Create API Key.

Name it specifically for this use case — I name mine per brand or per project (e.g. “Imagen Test - [Brand]”).

Reason: API keys can leak, and if one does, you want to know exactly which workflow it belonged to so you can revoke it without taking everything else down.

Copy the key. Don’t paste it anywhere shared.

Tell Claude to build the skill.

In a fresh Claude chat, give it the skill creator (or tell Claude to build a reusable skill that takes image prompts as input and outputs generated images via the API).

Paste your API key. Tell Claude:
Set up the skill. Test it on a basic prompt

Claude will:
1.Look up available image gen models on Google AI Studio
2.Find Nano Banana 2 (Google’s current top image model — the one you actually want)
3.Write the skill code wrapping the API call
4. Run the test prompt against the API
5.Output the test image inline

If Google Drive is connected, Claude saves the output there too. If Co-work is set up to write to local files, the image lands directly in your computer’s file system.

Step 3 — Save as a Claude skill.
Once the test works, tell Claude: “Save this as a skill named [whatever].” I name mine “Imagen Gen Skill - [Brand]” so they’re scoped per project.

Now the skill is in your library. Available across all your chats, on desktop AND browser. Generate images by name-dropping the skill in any future chat.

Using this you can now run more automations through Claude such as a Seedance, king, advertorials, product pages via shopify mcp, and even mass generate native ads

will drop the sauce on each of these once I land