This Monster Wants to Eat Me

 ## Staff | Role | Member | |-|-| | Translator | Crunchyroll | | Translation Checker | blagoblag | | Timer | Olivo28 | | Editor | Chika | | Encoder | sgt, Aliciä | | Typesetter | Nyarthur (Anarchy Subs) | | Song Translator | Yon | | Song Editor | arctan | | Karaoke Timing | arctan | | Karaoke Effects | arctan (OP), Yon (ED) | | Quality Checker | Ringtomb | | Special Thanks | Anarchy (for loaning us their Nyaa), ||The0x539 (for doing most of the karaoke work years back), ||You (for watching our release <3) | ## Specs | Video | Audio | Subtitles | |-|-|-| | CR WEBRip AV1 | Japanese E-AC-3 2.0 | Full Subtitles [Heartside] | [Mediainfo](https://rentry.org/wdutquvr) - [Comparison](https://slow.pics/c/WaKhW14b) This is a smaller and lower quality version of [ [Heartside] This Monster Wants to Eat Me (2025) - S01E01v2 (WEBRip 1080p HEVC E-AC-3)](https://nyaa.si/view/2028895). Apologies for the delay on this. We had a little too much fun trying to make this release as amazing as possible. Hopefully the next episode will be out sooner. Filtering for this mini encode is largely the same as the main encode. In regular scenes, we use a stronger although still protective denoise, while in flashbacks and most part of OP and ED, we use a much weaker denoise. Weak lineart retention is the single most difficult part in this mini encode. Given the target filesize, we tried our best to retain the lineart as much as we can. We use Progression Boost Preset-Character-Boost-Butteraugli-Mean as a base for bitrate allocation optimisation. For regular scenes, first, we limit the effect of metric-based Progression Boost module to save space, with `metric_min_crf` set to as extreme as `25.00`. Based on this, we hyperboost the characters, with `character_crf_boost_max` set to `10.00` and a modified `character_crf_boost_alt_curve` enabled. For regular scenes, we use 5fish/SVT-AV1-PSY v2.3.0-C `--tune 0` for the encode. We perform the encode in the slowest speed we can accept, using a mix of `--preset -1` and `--preset 0`. To retain character lineart, we choose to leave `--enable-cdef 1` and `--filtering-noise-detection 1` set, and use `--psy-rd 2.4` for retention. It has been a recent discovery that boosting chroma helps preserve character lineart. In this encode, we use `--key-frame-chroma-qindex-offset -25 --chroma-qindex-offsets [-19,-14,-13,-11,-10,-10]` to boost chroma. In addition, we add in the typical methods for better lineart retention, including `--max-32-tx-size 1` and `--low-q-taper 1` in scenes that are boosted to better `--crf`, as well as a universal `--noise-norm-strength 4`, `--qp-scale-compress-strength 3.4`, and `--qm-min 9 --chroma-qm-min 10`. To detect flashback scenes, we use `DFTTest().denoise({0.0:0.0, 0.04:0.0, 0.08:40.0, 0.28:32.0, 0.32:0.0, 1.0:0.0}, tbsize=1)` to separate the noise, and then compare noise diff against plain luma diff. With Alternative SVT-AV1, we use a different encoder than regular scenes, svt-av1-psyex v3.0.2-B `--tune 3`, in flashback scenes. Parameterwise, we use `--film-grain 28 --adaptive-film-grain 0` for grain synthesis, with the addition of a chroma boost of `--key-frame-chroma-qindex-offset -18 --chroma-qindex-offsets [-16,-14,-13,-13,-11,-11]`. [Join our Discord](https://discord.gg/EpCUu7ZA4u) to hang out and talk about ~~doomed~~ complicated yuri! Until next time! ~The Heartside Crew