My Secret AI Workflow for Viral YouTube Thumbnails

I have made thousands of YouTube thumbnails. I have hired a designer in the Philippines. I have paid for multiple courses to try to figure this out — thousands of dollars in courses alone. Thumbnails are a documented blind spot of mine, and I am always honest about that. But something changed over the last few weeks, and people started complimenting my thumbnails. That has basically never happened before. So I recorded exactly what I am doing, step by step, because the answer is AI — specifically a workflow that takes about 15 minutes per video and produces results that match or beat what I was paying a freelancer to do.

This post walks through every step of that workflow: how I pull the transcript, how I use ChatGPT to generate better titles first (that part matters more than most people realize), how I feed those titles into an AI image tool called Nano Banana to generate six or seven thumbnail options, how I post-process the best ones in Photoshop to add drama, and how I finish with YouTube’s own built-in A/B testing feature. If you have a blind spot for thumbnails too, or if you want to offer thumbnail refresh services to other YouTubers, this is the exact process I use.

What You’ll Walk Out With

A clear understanding of why click-through rate determines whether YouTube pushes your video to a wider audience
The exact order of operations for an AI thumbnail workflow that takes roughly 15 minutes per video
How to pull a clean transcript from Adobe Premiere Pro so you can feed it to ChatGPT
How to use ChatGPT to optimize your video title first, then use that title to prompt Nano Banana for thumbnails
A simple Photoshop post-processing step (vibrance, saturation, contrast) that makes AI-generated images pop on a busy YouTube feed
How to run YouTube’s built-in “test and compare” to let data pick your best thumbnail instead of guessing
Why turning your videos into blog posts compounds your results over time through backlinks
How to identify which platform is the right fit for your specific skills and situation at finder.platformproof.com

The Thumbnail Problem Nobody Talks About Honestly

Click-through rate, or CTR, is the percentage of people who see your thumbnail and actually click. YouTube measures it out of 100 impressions. A typical YouTube video lands somewhere between 3% and 8%. If you can get above 10%, YouTube reads that as a signal that the video is resonating and will push it out to a broader audience — meaning YouTube is essentially doing free promotion because your thumbnail did its job.

The thumbnail is not a cosmetic thing. It is a conversion tool. Alston uses an analogy that makes this concrete: imagine you are driving down a major highway, and you see the giant golden arches of McDonald’s. That sign is competing with a dozen other signs for your attention at 65 miles per hour. Your YouTube thumbnail is doing the same thing in a feed full of competing videos. It needs to be the sign that makes somebody pull off the highway. A great thumbnail gets people into the restaurant. But if the food is bad — if the video does not deliver — the thumbnail can only do so much.

The problem is that thumbnail skill does not come naturally to most people, including experienced YouTubers. Alston hired a designer from the Philippines who did technically good work, but the CTR did not move. He paid for multiple courses. Nothing consistently improved his numbers until he built this AI-assisted workflow.

Step 1: Pull the Transcript from Adobe Premiere Pro

The workflow starts with the transcript, not the thumbnail. This is the part that surprises people. Before touching any image tool, Alston uploads the video into Adobe Premiere Pro, goes to Export, and exports a TXT file of the transcript. That file goes into his Downloads folder and becomes the input for the next step.

Why start with the transcript? Because the title drives the thumbnail. You want the thumbnail to visually represent the strongest possible version of your title, not the title you originally published. Most older videos have titles that were written quickly and never revisited. The transcript is the raw material for generating a better title, and a better title is what you will feed into the image generator to get a relevant, compelling thumbnail prompt.

This step also works when you are going back to update old videos, which is exactly what Alston was doing in this video. Updating an old thumbnail and title takes 15 minutes and can add views to a video you published years ago. You do not need to film anything new. You just need the original transcript.

Step 2: Use ChatGPT to Optimize the Title, Description, and Tags First

The second step is to take that transcript into ChatGPT with a long, detailed prompt designed specifically to generate improved titles, descriptions, and tags. Alston keeps this prompt saved in his notes and pastes it in fresh each time. The prompt itself is long enough that he stores it externally rather than retyping it. He mentioned making it available in the comments for people who ask.

The reason for doing titles before thumbnails is sequencing. ChatGPT gives you multiple title options based on the actual content of the video. You pick the strongest one. Then that title becomes an input inside the thumbnail prompt you will use in the next step. The image generator needs to know what the video is actually about so it can produce something visually relevant. If you skip the title step and just generate a generic image, you are likely to end up with something that looks fine in isolation but does not communicate anything specific to a viewer scrolling past at speed.

For the example in the video, ChatGPT generated a title like “Screenflow Tutorial for Beginners: The Easiest Video Editor for Mac Users.” Alston copied that, pasted it into the designated placeholder inside his thumbnail prompt, and used it to drive the next step.

Step 3: Generate Six or Seven Thumbnail Variations in Nano Banana

Nano Banana is the AI image tool Alston uses to generate the actual thumbnail visuals. He initially called it Sora in the video and then corrected himself — it is Nano Banana. The workflow here is to paste the thumbnail prompt into Nano Banana, attach a photo of himself (he keeps a specific image that he uses for almost every thumbnail), and let it generate a few options. He does this multiple times in the same session to build up a pool of choices.

The prompt he uses in Nano Banana is long and detailed, combining what he describes as “the best suggestions and ideas” for thumbnail design into a single structured prompt. He generates roughly six to seven variations per video, sometimes more. Each generation takes a few seconds. When he runs out of tokens in a session, he exits, starts a new session, and continues — fresh token allocation every time.

Having six to seven options matters because thumbnail quality is subjective and not every generation lands. By producing a pool of options rather than stopping at the first acceptable result, Alston can pick the two or three that genuinely look strong before moving to post-processing. The process is fast enough that generating extra options does not meaningfully add to the total time.

Step 4: Post-Process the Best Options in Photoshop

AI-generated images often look slightly flat compared to high-performing YouTube thumbnails. The thumbnails that stop someone mid-scroll tend to have high contrast, punchy colors, and a slightly dramatic quality. Alston’s post-processing step in Photoshop addresses that.

His process once an image is in Photoshop: first, rasterize the layer. Then go to Image Adjustments and pull up Vibrance. He turns vibrance up significantly — enough to make the colors feel energetic without looking fake. Then he increases Saturation as well, though he notes you can push it too far. Next, he goes to Brightness and Contrast and raises the Contrast setting. He keeps Brightness only slightly higher than default. The goal is an image that reads as a little more dramatic than reality. That quality — drama — is what performs on social media.

You do not need Photoshop specifically. Any photo editor that lets you adjust vibrance, saturation, and contrast will work. Alston uses Photoshop because it is his existing tool, but the editing moves themselves are basic enough to execute in GIMP, Canva’s image editor, or any comparable software. He exports the finished file with a date-based filename — in the example, something like “10301” for October 30th — so he can keep track of versions easily.

Step 5: A/B Test Using YouTube’s Built-In “Test and Compare”

The final thumbnail step is YouTube’s own testing feature, which Alston calls “test and compare.” This tool lets you upload multiple thumbnail options for the same video and lets YouTube serve different versions to different viewers to see which one generates a higher click-through rate. It is built into YouTube Studio and does not require any third-party software.

Alston typically loads his top two or three finished thumbnails into the test. YouTube rotates them and tracks which version gets more clicks over time. The data decides. This removes guesswork from the final selection and means you are not relying on your own visual judgment — which, as Alston openly acknowledges about himself, can be unreliable when thumbnails are a known weak spot.

When Alston does this for older videos, the goal is incremental gains. He is not expecting one thumbnail swap to transform a dormant video overnight. But picking up 20 extra views per day per video, across many videos, adds up. And occasionally a strong thumbnail on an older video catches YouTube’s algorithm at the right moment and generates a meaningful spike.

Not sure if YouTube is even the right platform for you?

Answer a few questions and find out which online income path actually fits your skills and schedule at finder.platformproof.com.

The Bonus Step: Convert Videos Into Blog Posts for Backlinks

At the end of the workflow, Alston showed one additional step he does for older videos: turning the video into a blog post. He uses another detailed ChatGPT prompt to generate blog post content from the transcript, then pastes it into WordPress and publishes it. He mentioned having a specific “blog post prompt” for this purpose, separate from the title and thumbnail prompts.

The reasoning is backlinks. Every time your YouTube video appears somewhere else on the internet — a blog post, a forum discussion, an embedded share — it creates a connection that search engines count as a signal of authority. Alston described this as the “lazy way of explaining backlinks,” but the core idea is sound: a YouTube video embedded in a blog post on your own site means the video exists in two indexed locations instead of one. That is better for long-term discovery.

He is also clear about his preferences here: he would rather viewers watch the video than read the blog post. The blog post is infrastructure, not primary content. He publishes it quickly, without heavy editing, because the goal is presence across the internet rather than a polished written piece. If people read it, great. If it helps YouTube rank the video, even better.

This Workflow as a Service You Can Sell

Alston flagged this twice in the video: this entire workflow is something you could offer as a paid service to other YouTubers. Many creators with large back catalogs of videos know their old thumbnails are underperforming but do not have the time or skills to fix them. A thumbnail refresh service — going through their existing videos, running this AI workflow, generating new options, testing them — is a concrete, deliverable service with a clear benefit.

The key to selling it, as Alston put it: “sell the benefits, not the service.” People do not want to hear about Nano Banana or Photoshop or vibrance sliders. They want more views. They want higher click-through rates. Frame the offer around the outcome — help them get more traction on videos they already made — and the technical details become irrelevant to the pitch. The 15-minute execution time per video also means you can price this service at a meaningful rate while keeping your own time investment reasonable.

Honest Drawbacks

The thumbnail workflow Alston describes has real results — his last video before recording this had a 6% CTR, above the normal range. But there are a few honest limitations worth understanding before you commit to this approach.

AI image likenesses are inconsistent. Alston mentioned in the video that the generated images “sometimes don’t look 100% like me.” When you use a photo of yourself as an input, the AI will approximate your appearance, not reproduce it exactly. Some outputs will look accurate. Others will drift. You have to generate enough variations to find one that works, and you need to check each output before publishing.

A good thumbnail does not save a weak video. Alston is direct about this. CTR gets people in the door. Once they are watching, the video itself has to deliver watch time, engagement, and retention. YouTube’s algorithm looks at what happens after the click as much as it looks at CTR. A thumbnail that dramatically overpromises what the video delivers will hurt your average view duration, which cancels out the CTR gains.

The prompts require investment to build. Alston’s thumbnail prompts and title prompts are long and detailed for a reason. He did not generate them in five minutes. Getting to a prompt that reliably produces strong options takes iteration. If you are starting from scratch, plan to spend time refining your prompts before the workflow becomes fast. The process he shows took time to develop — he’s been refining it for three to four weeks, based on what he said in the video.

Nano Banana token limits reset per session. You get a limited number of image generations per chat session in Nano Banana. When you hit the limit, you exit and start a new session. This is a minor inconvenience rather than a blocker, but it means the workflow is not fully automated and requires you to actively manage the tool.

The Complete Workflow at a Glance

Step 1: Upload the video to Adobe Premiere Pro, export a TXT transcript, save it locally
Step 2: Paste the transcript into ChatGPT with a title/description/tags optimization prompt; copy the best title option
Step 3: Open Nano Banana, paste the thumbnail prompt with the new title inserted, attach a photo of yourself, generate six to seven variations
Step 4: Download the best two or three, bring them into Photoshop, rasterize the layer, increase vibrance and saturation, raise contrast, export with a date-based filename
Step 5: Go to YouTube Studio, open the video, use “test and compare” to upload multiple thumbnail versions and let YouTube pick the winner via data
Optional: Use a blog post prompt in ChatGPT to convert the transcript into a WordPress post for backlink distribution

Find Your X

The thumbnail workflow in this video is one piece of a larger question: is YouTube actually the right platform for your situation right now? For some people, the answer is yes — they have the patience for a channel, they enjoy video, and they are willing to work on skills like thumbnails over time. For others, there are faster paths to a first income online using skills they already have.

If you are trying to figure out where to focus, visit finder.platformproof.com. It asks a few questions and matches you to the income path that fits your specific setup — no guessing, no generic advice.

Frequently Asked Questions

What is a good click-through rate for YouTube?

Most YouTube videos land between 3% and 8% CTR. Getting above 10% is a strong signal — YouTube typically responds by pushing the video to a wider audience because it reads high CTR as evidence that viewers find the content compelling. Alston mentioned his last video before recording this tutorial hit above 6%, which is above average for his channel.

What is Nano Banana and where do I access it?

Nano Banana is an AI image generation tool that Alston uses to create YouTube thumbnail visuals. You can access it through its web interface and provide text prompts along with a reference image. It generates multiple image options per session, with a token limit per conversation that resets when you start a new session. Alston uses it specifically because it can incorporate a photo reference of himself to approximate his likeness in the generated images.

Do I need Adobe Premiere Pro to get my transcript?

No. Alston uses Premiere Pro because it is his existing video editor, but you can get a transcript from any editor that supports caption export, from YouTube’s own auto-generated captions (available under the video’s subtitle settings), or from free tools like yt-dlp. The transcript just needs to be a clean text file you can paste into ChatGPT.

Why generate the optimized title before making the thumbnail?

The title drives the thumbnail. When you feed an updated, stronger title into the Nano Banana prompt, the AI has specific context about what the video covers and can produce visuals that match. If you skip the title step and just ask for a generic thumbnail, you are likely to get images that look fine but do not communicate anything specific to a viewer. Title first, thumbnail second is the correct sequence.

Do I need Photoshop for the post-processing step?

No. Alston uses Photoshop because he already owns it, but the edits he makes — vibrance, saturation, and contrast adjustments — are available in free tools like GIMP, the image editor built into Canva, or other basic photo software. The goal is just to make the image pop more against a busy feed, which any tool that supports those three adjustments can accomplish.

What is YouTube’s “test and compare” feature?

Test and compare is a built-in YouTube Studio feature that lets you upload two or three different thumbnail versions for the same video. YouTube serves each version to different segments of your audience and tracks which one earns more clicks over time. After the test collects enough data, you can see which thumbnail performed better and keep that one as the permanent option. It removes the guesswork from thumbnail selection and lets viewer behavior make the final call.

Can I offer this as a service to other YouTubers?

Yes, and Alston specifically mentioned this. Many YouTubers with large back catalogs of videos have old thumbnails that underperform. A thumbnail refresh service — using this AI workflow to generate and test new thumbnails for their existing videos — is a legitimate offering. When pitching it, focus on the outcome (more views on videos they already made, without filming anything new) rather than the technical process. The 15-minute execution time per video means you can take on multiple clients without the work becoming unmanageable.

Will updating old thumbnails actually increase views on old videos?

It can, though not dramatically on every video. Alston estimates picking up 20 additional views per day per video on average. Across a large catalog of videos, that adds up over time. Occasionally, a stronger thumbnail on an older video catches YouTube’s recommendation engine and generates a more significant spike. The approach is most valuable when you have a large back catalog of videos on topics that are still searched regularly.

Sources

Alston Godbolt, “My Secret AI Workflow for Viral YouTube Thumbnails,” YouTube, https://youtu.be/YSUmG_4ktNc
YouTube Help, “Test and compare thumbnails,” support.google.com/youtube
Nano Banana, AI image generation tool, nanobanana.com
Adobe Premiere Pro, transcript export feature, adobe.com