- The screenshot feature was initially added as an experimental feature, then removed from Hayami entirely one update later due to a variety of reasons. This feature’s vision was overly ambitious, aiming to allow you to screenshot your current episode, and in-future updates, integrate this into the Hayami comment editor, seamlessly allowing image hyperlinking. To add to this, expanding the capture feature to recording a scene up to e.g. 20 seconds, transcoded to a GIF (likely using gifenc instead of ffmpeg.wasm for lightweight-nature) for action-packed scenes for use on e.g. Tenor. The feature had worked so you keybind screenshot your screen, or, element-picker choose an element on-page (e.g. a video) so it exports a raw photo of your screen, no matter if you were in fullscreen mode or not (uses Dynamic Host permission requesting as already-exists if player was on a different domain e.g. static.crunchyroll.com). This did, in part, achieve it’s goal on a variety of sites, but due to DRM, had difficulties on some streaming sites e.g. Netflix, requiring encouraging the disabling of hardware accelaration, which may be consequential to older hardware. The feature may not have been used a whole lot to justify the increase in bundle size (& since gifenc would be needed for video encoding), to top off other solutions like ShareX, and the built-in Windows snipping tool already having video recording & screenshot support out-the-box, making this feature overly bloat the extension’s intended purpose.
-
An “AI assistant” mode was added to help with mapping, but was scrapped entirely after improving the manual mapper. Initially, web-llm was used to experiment for a more privacy-respecting experience, however, the token limit kept being reached for processing larger-size HTML files. Other ways could’ve been used (i.e. using page context to detect what an anime is, then using the selector to grab this information), which was the technique used with Gemini Nano. This feature as a whole though was not a surefire way to accurately get elements on-page, and can be highly inaccurate at times; the feature in itself was more of a nuisance than actually aiding user experience in selecting elements on-page. For historical purposes, the page housing the documentation for this feature has been preserved here.
- About: The “AI assistant” mode brought support for Google AI Studio, Mistral, OpenRouter, Gemini Nano and any OpenAI-compatible API that aimed to process the HTML of your current page to get the CSS selectors & XPath for you of the anime name and episode number.
- Purpose: All third-party considered models for this feature had free plans, and the feature was disabled by-default, enabled manually. Reason for this feature initially is for further reach for sites that can sometimes be more difficult to find certain elements to enable mapping i.e. anime names due to hidden elements on-page visibly.
- Why it was scrapped: For AI-assistant functionality, feeding the entire page HTML uses too much resources (in order to have an accurate result), takes too long, and can be highly inaccurate (LLMs given long HTML files - too much noisy material). To reduce the noisy material, selections of i.e. where the episodes are would have to be done, going against the point of the entire feature (selections still have to occur user-side - what’s the point in the assistant?). After more digging, most sites have a corresponding theme to i.e. grab episode numbers (e.g. ‘active’ class when there’s a list format), allowing selections to be done accurately across sites, so e.g. if someone selects episode 8 (current episode they’re on) while hitting a matching pattern, it knows from this that’s the CSS selector, not the element itself. Adding an import/export feature further made the AI assistant unnecessary, as now if one-person did the effort of mapping, it could be shared & imported. To it top off, KomentoScript, allowing syncing from a third-party URL: this feature was deemed useless.
-
Crunchyroll archived comments can’t be added as a Hayami discussion platform due to Crunchyroll discontinuing the backend API for comments entirely.
- Debugging notes: It throws a 502 Bad Gateway error on /talkbox/guestbooks (assuming CR has withdrawn support on the backend entirely for comment discussions). The classic Crunchyroll site which used /comments with parameters of talkbox IDs throws a 404 Not Found error, assuming this was withdrawn some time ago upon migration to the new site.
- There are some partial discussions archived through the Internet Archive, but only the top initial comments. Anything past that (on-scroll to get more comments) isn’t archived.
Discarded features
This page is intended for note-taking of features that couldn’t be implemented, or were discarded over time. This page is sorted per newest features that were scrapped top-down.