Short answers about the project
What the report can tell you, what it cannot tell you, and how the research handles your thumbnail.
What is Eyelicium?
Eyelicium is a school research project and thumbnail review app. It studies what catches the eye in a YouTube feed and turns that evidence into practical design feedback.
Why is it called Eyelicium?
The name combines the eye with an angler fish lure. A thumbnail can act like a small visual lure and stop a scroll before someone reads the full title.
What makes a good thumbnail here?
A good thumbnail makes one honest visual promise quickly. Its main subject is clear at feed size, important text is readable, competing details are controlled, and the picture adds something useful to the title.
Does the report predict clicks, CTR, or views?
No. Public YouTube data does not reveal thumbnail click-through rate. Topic, audience, distribution, timing, title, and other factors also affect performance. Eyelicium is a design review, not a performance guarantee.
What does Eyelicium measure?
The report keeps four readings separate: human first-choice attention patterns, seven design and perception checks, visual category conventions, and optional similarity to a channel’s recent public thumbnails.
What is saliency?
Saliency is a pixel-based estimate of which image areas stand out because of contrast, color, or structure. It is not eye tracking and does not prove where every viewer will look.
Why do I choose a category?
A technology thumbnail and a music thumbnail often use different visual conventions. The category changes the reference examples, not the fixed design checks.
What data did the study use?
The project used public YouTube thumbnails with public video category and view information. The first-choice study produced 330 usable choices. It did not use private YouTube Studio analytics or public CTR data.
How well did the models work?
The category model identified the exact category 53.5% of the time and included it in the top two 74.8% of the time. The public-view regression explained almost none of the held-out variation, with R² of 0.003. The app therefore treats these readings as context, not view prediction.
Does Eyelicium save my uploaded thumbnail?
The main review processes an image for the request and does not add it to a training library. Uploaded drafts are not written to persistent storage by the app.
How can I help improve the research?
The most useful next dataset would pair thumbnails with owner-approved impressions, CTR, audience context, and A/B test results. To volunteer anonymized YouTube Studio data, email jpaul.fernandez18@gmail.com.