We are two physicists who would rather write a script than record ourselves talking to a camera. For months, that preference felt like a liability. Every marketing guide said the same thing: show your face, be authentic, post three times a day, engage in the comments. We did none of it. We were too busy building NerdSip to become content creators. Then we automated the entire thing, and it outperformed everything we had ever done manually.
350,000 views in two weeks. Zero face reveals. Zero hours spent editing video. All of it running on autopilot while we wrote code and fixed bugs.
This is the exact system we built, why it works, and what we learned along the way.
Why We Were Terrified of Marketing
Let's be honest about something. Most developers are not afraid of hard problems. We will happily spend a weekend debugging a race condition in a serverless database. We will refactor an entire navigation stack because the state management felt slightly wrong. These things energize us.
Marketing does the opposite.
Marketing feels arbitrary. You post something, it gets 14 views. You post something nearly identical the next day, it gets 3,000. There is no stack trace. There is no debugger. The feedback loop is noisy and delayed, and the rules change every time a platform updates its algorithm.
We launched NerdSip in March 2026. The app was solid. The learning experience was good. Users who found it loved it. The problem was that almost nobody was finding it. Our App Store listing existed in a vacuum. We had a beautiful landing page that received roughly the same traffic as a personal blog about sourdough starters.
We tried the obvious things. We wrote blog posts. We posted on Reddit. We sent cold emails to education bloggers. Each effort produced a small, temporary spike followed by a return to baseline. The fundamental issue was scale: as a two-person team, we could not produce enough content to matter on platforms that reward volume above all else.
TikTok was the platform we kept hearing about. Short-form video. Massive organic reach. The algorithm rewards content quality over follower count, which theoretically levels the playing field for small creators. But every TikTok strategy we read about required exactly the thing we did not want to do: record ourselves.
Then we discovered the slideshow format.
The Slideshow Insight That Changed Everything
TikTok is not just a video platform anymore. Photo carousels, where users swipe through a series of images, have become one of the highest-engagement formats on the platform. Educational content performs especially well as slideshows because each swipe delivers a new piece of information. The viewer feels like they are learning something with every frame.
This was the insight that unlocked everything for us. Slideshows do not require a camera. They do not require your face. They require images and text. Both of which can be generated by AI.
We looked at what was already working in the educational slideshow niche. Science facts. Psychology insights. "Things you didn't know about..." posts. The successful accounts all shared a pattern: visually striking images paired with concise, curiosity-driven text overlays. Six to ten slides per post. A hook in the caption. Relevant hashtags.
Every single component of that formula was automatable.
Building the Pipeline: OpenClaw for Strategy
The first problem was content strategy. What topics should we post about? What captions should we write? What hooks would stop someone mid-scroll?
We already used OpenClaw for marketing research at NerdSip. It is an AI agent designed specifically for marketing workflows: competitive analysis, content research, audience analysis, and content generation. We pointed it at TikTok's educational niche and asked it to do what it does best.
Here is what OpenClaw handles in our pipeline:
- Topic generation. We feed it a broad category (neuroscience, space, psychology, physics, biology) and it generates 20 to 30 specific post topics per batch. Not generic topics. Specific, curiosity-triggering angles. "Why your brain deletes memories while you sleep" instead of "How memory works."
- Caption writing. Each topic gets a TikTok caption with a hook in the first line, educational value in the middle, and a call to action at the end. OpenClaw understands the platform's tone: conversational, slightly breathless, but grounded in real science.
- Hashtag strategy. It generates a mix of high-volume and niche hashtags for each post, calibrated to the specific topic rather than the same generic set pasted everywhere.
- Content calendar. It schedules topics across categories to avoid posting three neuroscience posts in a row. Variety keeps the audience engaged and signals to TikTok's algorithm that the account covers a breadth of content.
One OpenClaw session gives us two weeks of content. Topics, captions, hashtags, scheduling. All of it lands in a structured JSON file that feeds the rest of the pipeline.
The Secret Weapon: Panoramic AI Images
This is the part that makes people stop scrolling.
We use Nano Banana 2, Google's image generation model available through AI Studio (its full technical name is Gemini 3.1 Flash Image Preview). It generates photorealistic images from text prompts. Good ones. The kind of images that make people pause and say, "Wait, that's AI?"
But here is the trick. We do not generate standard portrait images for each slide. We generate a single 4K ultrawide panoramic image in 21:9 aspect ratio. One massive, continuous scene.
Then we split it into six equal vertical strips.
Each strip becomes one slide in the TikTok carousel. When the viewer swipes through, they are moving across a single panoramic scene. The visual continuity is hypnotic. Each frame reveals the next piece of a larger picture. It feels intentional and cinematic in a way that six unrelated images never could.
Why the Panorama Technique Works So Well
There are three reasons this technique outperforms standard slideshows.
First, it rewards swiping. Each swipe reveals a new section of the panorama. The viewer's brain wants to see the complete image, so they keep swiping. TikTok's algorithm interprets this engagement (time on post, swipe-through rate) as a strong signal that the content is valuable. More engagement means more distribution.
Second, it looks expensive. A panoramic image split into frames looks like a professional photography project. It does not look like a low-effort AI post. The production value signals quality, even though the entire image was generated in under 30 seconds.
Third, it creates a narrative. A panoramic scene of the deep ocean, or the surface of Mars, or the inside of a neuron, tells a visual story. Each slide is a chapter. The text overlay on each frame adds the educational content. The combination of visual narrative and factual content is far more engaging than text on a plain background.
The Technical Details
For those who want to replicate this, here is the exact process.
We call the Nano Banana 2 model through Google AI Studio's API. The prompt follows a specific template:
"A photorealistic 4K ultrawide panoramic image (21:9 aspect ratio) depicting [scene description]. Continuous composition from left to right. Cinematic lighting. Rich detail throughout the entire width. No text overlays. No borders."
The scene description comes from the topic that OpenClaw generated. For a post about deep-sea bioluminescence, the scene might be: "the abyssal zone of the ocean, moving from dark crushing depths on the left through increasingly bioluminescent creatures toward a cluster of glowing jellyfish on the right."
The model returns a single high-resolution image. Our script then uses sharp (a Node.js image processing library) to slice it into six equal vertical segments. Each segment is saved as a separate PNG in portrait orientation. We add text overlays programmatically using the caption content from OpenClaw. The final output is a folder with six numbered images ready to upload.
The entire generation process for one post takes about 45 seconds. Image generation is the bottleneck; the slicing and text overlay are nearly instant.
What the Numbers Actually Look Like
We started posting on March 17, 2026. Here are the real numbers from the first two weeks.
- Posts published: 24 (roughly 2 per day, scheduled)
- Total views: 350,000
- Total likes: 5,000
- Followers gained: 450 (from zero)
- Top-performing post: 100,000 views ("Earth in 250 Million Years")
- Lowest-performing post: 1,200 views (a quantum physics topic that was probably too abstract)
For context, most new TikTok accounts average 200 to 500 views per post in their first month. We were hitting 10,000+ consistently by day four.
The most important metric for us, though, is not views. It is app installs. We include a link to NerdSip in our bio and mention it in captions where it is relevant (not every post). In two weeks, TikTok became our second-largest source of organic app installs, behind only App Store search.
What Worked
Science content is TikTok gold. People are genuinely curious. Posts about neuroscience, space, and human biology consistently outperformed everything else. The platform's audience skews young, and young people are hungry to learn things that feel mind-expanding.
The panorama format is a retention machine. Our average swipe-through rate (the percentage of viewers who swipe through all six slides) is around 68%. For comparison, typical TikTok slideshows see 30 to 40%. The panoramic continuity keeps people swiping.
Consistency matters more than perfection. We post twice a day, every day. The algorithm rewards accounts that post regularly. Our automated pipeline makes this trivially easy because the content is always ready.
Hooks are everything. The first line of the caption and the first slide determine whether someone stops scrolling. OpenClaw generates hooks that are specific and curiosity-driven. "Your brain deletes memories every night. Here's why." beats "How memory works" every single time.
Batch generation eliminates decision fatigue. We generate two weeks of content in one sitting. No daily scramble to figure out what to post. No staring at a blank screen. The pipeline outputs a folder of ready-to-post content, and we schedule it.
What Did Not Work
Not everything was a win. Here is what failed or underperformed.
Abstract topics bomb. Quantum physics, string theory, abstract mathematics. These topics sound fascinating as blog post titles but die on TikTok. The audience wants tangible, relatable science. "Why you can't tickle yourself" outperforms "The measurement problem in quantum mechanics" by a factor of 50. We learned to filter topics through a relatability test before generating images.
Text-heavy slides get skipped. Early on, we put too much text on each slide. Three sentences per frame. Nobody reads that on TikTok. We cut it to one sentence per slide, sometimes just a phrase. The image does the heavy lifting; the text is the punchline.
Some AI images look obviously AI. Nano Banana 2 is excellent, but it is not flawless. About one in ten images has telltale artifacts: impossible geometry, blurred details in faces or hands, text-like shapes that are not actual text. We built a manual review step into the pipeline. Every image gets a 5-second human check before it goes into the upload queue. This is the one part of the process that is not fully automated, and we think that is fine.
Hashtag flooding hurts. Our first batch used 15+ hashtags per post. Performance was mediocre. When we cut it to 5 to 7 highly targeted hashtags, views jumped. TikTok's algorithm seems to interpret excessive hashtags as a spam signal.
Posting time matters less than we thought. We spent hours researching optimal posting times. 7 PM EST. Lunchtime. Sunday mornings. In practice, our best-performing posts were published at random times because the scheduler does not optimize for time zones. The algorithm surfaces good content regardless of when it is posted. Time optimization is a distraction for accounts under 100K followers.
The Full Pipeline, Step by Step
Here is the complete workflow from start to finish.
Step 1: Content generation (OpenClaw). We input our target categories and audience profile. OpenClaw outputs 20 to 30 topic-caption-hashtag bundles in a JSON file. Time: about 15 minutes of prompting plus review.
Step 2: Image generation (Nano Banana 2). A script reads the JSON file, constructs a panoramic image prompt for each topic, and calls the AI Studio API. Each call returns a single 4K ultrawide image. Time: about 30 seconds per image, runs in parallel.
Step 3: Image processing (sharp + Node.js). Another script takes each panoramic image and slices it into six portrait-orientation segments. It adds text overlays to each segment using the caption content. Time: near instant.
Step 4: Human review. We scroll through the generated slides in a preview folder. Flag any images with obvious AI artifacts. Regenerate the flagged ones. Time: 5 to 10 minutes for a batch of 20 posts.
Step 5: Upload and schedule. We upload the image sets to TikTok and schedule them across the next two weeks. This is currently manual (TikTok's scheduling tools are limited), but it takes about 30 minutes for a full batch.
Total time investment for two weeks of content: roughly 90 minutes.
Compare that to what it would take to produce 24 pieces of content manually. Even if each post only took 30 minutes to research, design, and write, that is 12 hours. Our pipeline compresses 12 hours into 90 minutes and produces more consistent, higher-quality output.
Why This Matters for Indie Developers
Here is the thing nobody tells you when you launch an app. Building the product is the easy part. Finding users is the hard part.
The app stores are graveyards of excellent software that nobody ever found. The discovery problem is brutal, especially for indie developers without marketing budgets, without audiences, without the kind of personality that translates well to social media.
We are not natural marketers. We are not comfortable on camera. We do not have the instinct for viral content that some creators seem to be born with. But we can write code. And it turns out that writing code to solve marketing is a legitimate strategy.
The tools exist now. A year ago, this pipeline would not have been possible. AI image generation was not good enough. AI content strategy tools were not specific enough. The combination of OpenClaw's marketing intelligence and Nano Banana 2's image quality creates a workflow that genuinely works at scale.
This is not about replacing human creativity. Our best-performing posts still required human judgment: choosing the right angle, writing the right hook, catching the image that looks slightly off. The AI handles the volume and the execution. The humans handle the taste and the quality control.
If you are an indie developer sitting on a good product with no audience, this approach is worth trying. The barrier to entry is a weekend of scripting. The cost is essentially zero (Nano Banana 2 through AI Studio is free for developers). The worst case is that you learn something about content marketing. The best case is that you find an audience for the thing you built.
What Comes Next
We are not done iterating. Here is what we are working on now.
Video generation. Slideshow carousels work, but TikTok is still fundamentally a video platform. We are experimenting with using AI video models to turn our panoramic images into slow-pan videos with narration. Think Ken Burns effect, but the entire scene is AI-generated. Early tests look promising.
Automated A/B testing. We want to generate two versions of each post (different hooks, different image styles) and let the data decide which approach works better. Right now our review process is gut-based. We want to make it data-driven.
Cross-platform expansion. The same panoramic images work on Instagram Reels and Pinterest. We are building pipeline branches that reformat the content for each platform's specifications. One generation run, three platforms.
Tighter app integration. Our TikTok posts are about science topics. NerdSip teaches science topics. The connection is obvious, but right now it is just a bio link. We want to create deep links that take a TikTok viewer directly to a NerdSip course on the exact topic they just saw a slideshow about. Swipe through six slides about neuroplasticity, tap the link, land on a full micro-course about neuroplasticity. That is the funnel we are building.
The Honest Conclusion
We would be lying if we said we expected this to work as well as it did. Our hypothesis was modest: automate TikTok posting, get some views, maybe drive a few app installs. The reality exceeded our projections by an order of magnitude.
350,000 views is not viral by TikTok standards. People with ring lights and charisma get that on a single post. But for two developers who spent zero hours in front of a camera and 90 minutes total on content creation, it is a meaningful result. It is proof that the leverage exists.
The uncomfortable truth about marketing as an indie developer is that you have to do it. You cannot just build a good product and hope people find it. The app stores do not work that way. Search does not work that way. You need distribution, and distribution requires content, and content requires either time or automation.
We chose automation. It is working. We will keep iterating on the pipeline, keep publishing the numbers honestly, and keep sharing what we learn.
If you are a developer who has been avoiding marketing because it feels foreign and uncomfortable, consider this your permission to solve it like a developer. Write the script. Automate the pipeline. Let the machines handle the volume while you handle the vision.
That is what we did. 350,000 people saw the result.
Full disclosure. We are not affiliated with OpenClaw or Google. We pay for these tools ourselves. We are sharing this because we wish someone had shared it with us six months ago.
Frequently Asked Questions
How do you generate TikTok slideshow content with AI?
We use OpenClaw to generate content strategy, captions, and topics, then Nano Banana 2 (Gemini 3.1 Flash Image Preview) to create 4K ultrawide panoramic images. We split each panorama into 6 portrait frames, which become a slideshow-style TikTok post. The entire pipeline runs on autopilot.
What is the panorama-to-slideshow technique for TikTok?
You generate a single 4K ultrawide (21:9) panoramic image using an AI image model. Then you slice it into 6 equal vertical segments, each in portrait orientation. When uploaded to TikTok as a photo carousel, the viewer swipes through what feels like a seamless visual journey. The continuity between frames keeps people swiping.
Is it ethical to use AI-generated content on TikTok?
We believe so, as long as you are transparent about it. We do not pretend a human photographer shot these images. The value is in the educational content and curation, not in passing off AI art as handmade. TikTok's own guidelines allow AI-generated content when disclosed.
How much does this automated TikTok pipeline cost to run?
Nearly nothing. Nano Banana 2 is free through Google AI Studio for developers. OpenClaw has variable pricing for the content strategy portion. The biggest cost is the initial time investment to build the pipeline scripts, which took us about two days.
📚 Keep Learning
See What We Built
NerdSip is the AI learning app we market with this pipeline. Try it free and see what all those TikTok slideshows are about.
