AI images · checked 2026-10-19Can I vibecode ThumblifyAI?
YES· one-shottable
ThumblifyAI looks like a front end for hosted image models: a prompt, a few photos of your face, a reference thumbnail, 16:9 output. The site doesn't say which models it runs, but Gemini and OpenAI's image models both accept reference images and edit instructions, so an agent can wire a local studio over one sitting, and open-source builds on Gemini already exist. What stays behind is the library of proven thumbnails with view stats, the agent that researches your niche, and whatever tuning the vendor does. Likeness depends on the model, not the wrapper. Its own MCP server and CLI are the shorter road if you only want thumbnails from inside Claude Code.
$9.99/moprice
$120you'd save a year
one sittingbuild time
—people replaced it
What you need on hand.
DifficultyBeginner · No setup beyond your agent.
TimeOne to two hours for the first version
- Claude Code, Codex or CursorAny of them can run the prompt.
- Node.js 22sharp and the canvas package install prebuilt binaries on most systems.
- A Gemini or OpenAI API keyImage calls are billed per image; check each provider's current price list.
- Photos of your faceThree to five, different angles and lighting, no sunglasses.
Before you buildThumblifyAI has no free plan, but a $5 credit pack buys up to 20 thumbnails. If you make a handful of thumbnails a month, buying credits may cost less than running your own studio.
- 1. Install one agent: Claude Code, Codex, Cursor.
- 2. Make an empty folder and open it in the agent.
- 3. Paste the prompt below, press enter, and answer its questions as it builds.
The prompt
Build me a local YouTube thumbnail studio to replace ThumblifyAI. Requirements: - Node.js 22 + TypeScript: a Fastify server and a React UI built with Vite. `npm start` serves everything on http://localhost:4600. - Models: `@google/genai` with GEMINI_API_KEY in .env as the default image model, plus an OpenAI-compatible images endpoint (`openai` package, OPENAI_API_KEY) as a second option; the model name is a setting, and both calls go through one ImageProvider interface. - Storage: better-sqlite3 in ./data/studio.db for projects, generations (prompt, model, reference ids, estimated cost, status) and settings; images are PNG files in ./data/images/. A price table in settings feeds a running monthly cost estimate. - Personas and styles: a persona is a named set of 3 to 5 photos of a face, a style is a named set of reference thumbnails; choosing either attaches them as reference images to a generation. No training step. - Generate: pick a persona and an optional style, type the video title and a visual idea, ask for 1 to 4 variations. Output is 16:9 at 1280x720, resized with `sharp` when the model returns another size. - Edit: paint a mask over a region on a canvas, add an instruction, send an edit call, and keep every version in a history strip so I can step back. - Text: add the headline as a layer afterwards with `@napi-rs/canvas` (font, size, outline, position) so the spelling is always right; export PNG or JPEG under the size limit in settings. - Single user. No accounts, no telemetry, local-first: only the model calls leave the machine, and the README lists exactly what is sent (prompt and reference images). - Out of scope: YouTube research, view-count scoring, face-model training, a shared template library. - README: getting the API keys, the price table, and a note that likeness and spelling vary by model, so generate several variations and pick.
Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.
Then keep going
Next prompts, same folder.
The first prompt builds the core. Run these one at a time in the same folder to grow it into your own ThumblifyAI.
Add channel style matching to the thumbnail studio in this folder. Requirements: - A "Channel" page that takes a YouTube channel handle, fetches its recent video thumbnails through the YouTube Data API (YOUTUBE_API_KEY in .env) and saves them as a style set. - Ask the image model to describe the shared look (colours, layout, text treatment) from six of them, store that description, and add it to every prompt that uses the style. - Show which reference images were used and let me untick any of them. - Update the README with the API setup and the quota it uses.
Add a thumbnail test log to the studio in this folder. Requirements: - For each project, mark two generations as variant A and B, record the date range each ran on the video, and let me type impressions and clicks for each by hand. - Compute click-through rate for each variant and show a plain verdict, with the sample size and a note when fewer than 1,000 impressions make the difference unreliable. - Keep the winner's prompt and settings as a reusable preset. - Update the README.
Add a CLI to the thumbnail studio in this folder. Requirements: - `npm run gen -- --persona me --style channel --titles titles.txt --variations 2` reads one video title per line and writes images to ./data/exports/<date>/. - Print the estimated cost before starting and require --yes to continue when it is over $2. - Retry failed calls twice with a pause, then continue with the next title and list the failures at the end. - Update the README with an example titles file.
What you lose
- The library of proven thumbnails with view counts and click-rate stats
- The agent that researches your niche on YouTube and proposes ideas
- Face and style models that persist and are tuned by the vendor
- Region-aware editing tuned for thumbnails
- Credits spread across plans instead of paying an API per image
Why people still pay
Creators buy back their time: the site sets itself against a freelancer at $40 to $150 per thumbnail and against hours in Canva. A tool built for the YouTube crop, with your face and a format that has worked for others in your niche, saves the trial and error.
ThumblifyAI or your own copy?
Over
Your usage
3h ×$/h
ThumblifyAI, 1 yr$120
Your own copy, 1 yr$65running + build session
You keep$55pays for itself in 2 months
Estimate · Nothing runs in the background, so the cost is the image API: from a few cents to roughly twenty cents an image depending on model and quality (check current prices), so about $10 for 40 high-quality images. The build is token usage on your agent plan.
Take your data, then cancel.
1 · Take your data out of ThumblifyAI
- 1Thumbnails: download each finished image from the app; the FAQ says they come in high resolution at YouTube's recommended 1280x720 and can be used on any platform.
- 2Your own photos and reference thumbnails are the inputs: keep the originals, since we found no way to export a trained face or style model.
- 3The service offers an MCP server and CLI for agents; we found no bulk export of past generations in what we read.
2 · Cancel ThumblifyAI
- 1Manage or cancel the subscription from your account settings, any time.
- 2After cancelling you keep access until the end of the billing period.
- 3One-off credit packs have nothing to cancel.
- Use up your credits first: the FAQ says there are no refunds, and it offers no free trial.
- Download every thumbnail you still need before the period ends.
- Face and style models you trained live in the service, so keep the photos you used and expect to start again elsewhere.
The escape hatch
Don't feel like building it? These are already free.
YouTube Thumbnail AI GeneratorOpen sourceMIT React and Vite app: upload a reference image, describe the thumbnail and get three Gemini variations; your own API key, kept in the browser.★ 20 · active Jan 2026 · self-host, real ops
WebVisit ↗Canva FreeFreeHosted design editor with YouTube thumbnail templates, available on its free plan.they host it
WebVisit ↗PhotopeaFreeFree browser editor that opens PSD files and works with layers, masks and text; ad-supported.they host it
WebVisit ↗
WebVisit ↗Canva FreeFreeHosted design editor with YouTube thumbnail templates, available on its free plan.they host it
WebVisit ↗PhotopeaFreeFree browser editor that opens PSD files and works with layers, masks and text; ad-supported.they host it
WebVisit ↗
The numbers
ThumblifyAI pricing
| Plan | Monthly | Annual, per month | What you get |
|---|---|---|---|
| Credits (pay once) | custom | — | $0.25 per credit, minimum 20 credits ($5); the page opens on 40 credits for $10. Credits never expire and unlock every feature. |
| Creator | $9.99 | $7.49 | 40 credits a month (360 a year on the yearly plan), all generation, editing, face and style features, MCP and CLI access, email support. |
| Influencer | $19.99 | $14.99 | 80 credits a month (720 a year), everything in Creator, priority email support. |
| Pro Studio | $39.99 | $29.99 | 160 credits a month (1,440 a year), early access to new features, priority support and live chat. |
- Billing
- Monthly or yearly subscriptions, or one-off credit packs. Cancel in account settings and keep access until the period ends. The FAQ says there are no refunds.
- Hidden costs
- Everything is credit-based. By our arithmetic from the page, every plan and the pay-once packs work out to $0.25 per credit. Training a face model or a style model costs 5 credits each, and the YouTube research agent costs nothing. The checkout link carries a promo code, so the price you see at payment may be lower.
- Source
- thumblifyai.com
Also on the list
Related apps
None: the FAQ says there is no free trial; the smallest purchase is a $5 credit pack.
Yes. An AI coding agent such as Claude Code, Codex or Cursor can build a usable personal version in one session with the prompt on this page, running on your own machine.
About $9.99 a month on the Creator plan, $119.88 a year (checked 2026-10-19).
The library of proven thumbnails with view counts and click-rate stats; The agent that researches your niche on YouTube and proposes ideas; Face and style models that persist and are tuned by the vendor; Region-aware editing tuned for thumbnails; Credits spread across plans instead of paying an API per image. If any of those matter to you, keep paying.
Yes: YouTube Thumbnail AI Generator (open source), Canva Free (free), Photopea (free).