TRIGGER制作アニメの制作素材データ/中間素材データ/中間生産物データ/中間制作物データをアカデミア研究者向けに提供開始
https://www.nii.ac.jp/news/release/2022/0315.html
Little Witch Academia (2013) 1 episodes 26 mins, animenewsnetwork
not the (2015 Kickstarter campaign) or (2017 television animation Little Witch Academia 25 episodes)
the dataset is for develop new and improved methods and technique
eg: VFI, VSR, 3D model generation etc.. more efficient production, management tools for industry
filling out an application here: https://www.nii.ac.jp/dsc/idr/trigger/
policy: https://www.nii.ac.jp/dsc/idr/trigger/documents/trigger-policy.html
@misc{tr2022a,
author = {TRIGGER Inc.},
date = {2022},
title = {Trigger dataset},
publisher = {Informatics Research Data Repository, National Institute of Informatics},
note = {dataset},
url = {https://doi.org/10.32130/idr.15.1},
doi = {10.32130/idr.15.1}
}
Little Witch Academia (2013) 1 episodes 26 mins, animenewsnetwork
not the (2015 Kickstarter campaign) or (2017 television animation Little Witch Academia 25 episodes)
the dataset is for develop new and improved methods and technique VFI, VSR, 3D model generation etc.. more efficient production, management tools for industry
all materials was scanned image data (tiff,tga,jpeg) (No animation data provided)
scenario/screenplay(JA:シナリオ)
only multi episode animation called Series composition(JA:シリーズ構成), episode contents director/(アニメ演出), moive, single episode just called scenario and director
storyboard/e-conte/(JA:絵コンテ) (197 files)
draft with picture and simple timing (variety action and dialogue), or Visual Storytelling, add the animation made it more details and smooth, coherence
drawing reference/(JA:作画設定&美術設定) as line art/monochrome picture also as designs/concept art (89 files)
include characters, scene, assets designs, pre-production notes
background (art) reference/(JA:美術ボード) as paint art
variety weather/time/season change scene reference but fixed perspective view for background drawing
background (art) drawing/(JA:背景画) (392 files)
as paint art but called drawing, also can be layered for compositing animation (背景美術 BOOK)
the background line drawing draft (JA:背景原図), has lower priority because drawing source from background reference and layout/LO. but higher priority when introduce AI for stylization background drawing
coloring reference/(JA:色彩設定) (67 files)
include model sheet, assets sheet, animation reference picture, in-production notes
cut envelope/(JA:カット袋) (for 383 cuts) (one cut mean count as one shot)
animated drawing content directory for each cuts, include in-production drawing notes, its covers, timesheet/X-Sheet(control of an envelope's material by timeline), (2d,3d) layout/LO, lots rough drawing, key-animation/key-drawing/"genga"/(penciltest), correction/addition drawing
this dataset estimated approximately 10,000 pages key-drawings, correction/addition drawings 7,000 pages, not sure it has douga?
(*) douga procedure (JA:動仕), nowdays mostly do with digitally, few analog.
-
when digital, drawings are scan into (RETAS Stylos) and directly do tracing (JA:色トレス) with tweening procedure (using bezier curve).
-
when analog, drawings was blend and clean-up with pen tweening became douga, then scan and standardized using (RETAS TraceMan).
(*) also not recommend "original picture"/"key frame" called for "genga", those word very confusing, it might due to machine translation error
tracing coloring/"cel"/(JA:仕上げ) (397 cuts), digital only
tracing easy to coloring (compare to illustration making), tracing following coloring reference to standardize lines colour, like solid line, coloring mark
cel is basic compositing unit. that can be sequence animation or static asset, include anything, parts, shadow etc. nowdays often blend cels with 3dcg.
modern frame animation, its hybrid of cels animation (compositing made animation, spatial featured between cels, like game sprites), and cel sequences animation (temporal featured like flipbook)
- its still not easy to do like related to (pre-production, post-production speech/dubbing, cuting editing and in-production timing)
generated pre-production contents like character designs (3D model generation etc..)
not included composite project file (layered animation) or in-production multiple stage animation (コンテ撮, 原撮, 線撮, 動撮/本撮), but animation compositing quite not easy from materials
And task like VFI, VSR, tracking, retrieval, generation etc... are video task need animation data (video), but no provide. so this dataset mostly usage is as images, image sequence
- too big filesize especially scanned background drawing and paper drawing
some time tiff,tga storage the cel very big(even 2Mb per image), that cel need convert TGA to 32bit PNG(only ~30kb), idk this material used picture resolution, usually reference/note:
7000x4000 png ~ 14Mb, 9000x6000 png ~ 30Mb, 4000x3000 tga ~ 35Mb
opentoonz dwanko run (one tif 1700x1100 might around 4Mb
retas/CLIP STUDIO sample ( forgot it, its also big scanned paper drawing and backgound drawing
netflix & producion I.G. , Sol Levante materials, 4 cuts cels
LO + backgound drawing psd, per 100Mb, 4000px tga per 2Mb
AfterEffects for animation EXPERT, used material
7Mb per LO, 8Mb per backgound drawing PSD
a 2010 television animation, one shot
72Mb 4000x2800 backgound drawing with (black outborder) PSD
- I already seen some materials
also not Traditional Paperless Animation/(デジタル作画), and drawing references, drawing/storyboard pictures are too rough or “trigger” style that not suitable use
base on idea, develop techniques that allow the researcher to easy 'reproduction' the finished animation
- paper drawing clean-up or tracing, also include apply AI upscaling for analog/digital drawing
this visionary comments from HN SolLevante4k comments, about 4K drawing animation
- generated secondary material( drawing effects animation, difference animation like animating mouth/lip/eyes (目パチ、口パク)
this animation has so many drawing effects animation, those can be generated training dataset
(*) but misuse not good, refers I watched a great anime Tentai Senshi Sunred!
- based on coloring reference or cels, coloring the paper drawing, paper douga or tracing/douga.
AI for coloring
-
cels management and checking, sync-with, replace-with various materials (動画検査, セル置き換え)
-
tracing/douga, paper drawing tweening/in-between (中割り,タップ割り,デッサン割り)
AI In-Between Animation
- animating object extraction from paper drawing (合成伝票, 原撮)
the in-production animation named "原撮", compositing quite not easy from paper, need extract animating object one-by-one from paper, which completely can apply AI.
-
automation blend cels, cels depth mask/map prediction, cels animation compositing animation automation (アニメ撮影, 被写界深度)
-
tracing anti-aliasing/stylization, celfx,2dfx like image harmonization/re-lighting/stylization (セル処理,アニメ撮影処理, アニメ動画特効)
refers to (CGWORLD アニメCGの現場 2017), (MdN 2017.11), fate liner 2015 special, jojo making
-
improve documentation and digitalization like import/export in-production material (storyboard, timesheet and various handwriten texts/numbers/instructions notes) (演出指示, ツメ指示, etc..)
-
paper drawing scan improve, pegs stabilize etc..
-
associate with terminology, semantic retrieval in-production material picture using CLIP pretrained
variety in-production notes enough collecting image-text pair dataset. you can refer to CLIPxGamephysics
- frames classification (whether key-animation or in-between), paper drawing correction/addition refine(作画監督, 作画添削) or temporal propagation
-
finished animation clip layered-neural modeling, learned layered prior/timing prior from layout, timesheet, effects note and cels animation parts (セル分け) (refers to soma making)
-
finished animation transcription, analyzer interface for teaching director
episode contents director dynamic supervise animation in-production, lost latest composition & effects project file from just literal text or storyboards almost impossible analysis director making details
refers to your ultimate guide to anime ending credits
refers to barikios animation workflow
-
from single episode script (text) to make "call sheet" and storyboard.
-
Meeting : preparing management materials, timesheets, requesting background art, drawing the picture of LO (is draft, with correction iteration) [actually lots animation do outsourced?]
-
begin drawing (no 3dcg involved)
in each shot, must know, what position and who involved (noted with colour paper, marked as "[ ]"). and idk how to translate terms
involved action and how those picture look like: making drafts, making addition/correction, or clean-up
-
production materials mainly from this stage, even for single episode, the data scale is giant. also too complicated and managed differently for each company and each animation
-
production process is fully black box, industry use no unify terms, also the staff credit too, it is not possible to list all position in staff credit for each shot, this resulting lots ”sakuga weebs“
LO レイアウト, レイアウト修正[yellow paper],演出修正[red paper] ->原画,原画修正,作画監督[yellow paper],総作画監督[green paper] ->動仕:動画(動画検査)& デジタル動画(デジタル動画検査) ->仕上げ
or (without 第二原画)
LO レイアウト, レイアウト修正,演出修正 ->原画,原画修正,作画監督 ->第二原画,作画監督,総作画監督 ->動仕:動画(動画検査)& デジタル動画(デジタル動画検査) ->仕上げ
-
Meeting : each shot pre-compositing [原撮, 線撮]: (re-take shot iteration), LO (camera/cels/"BOOK" motion correction), mark the timing: pre-dubbing(audio-visual), and sync like animating mouth/lip/eyes
-
output all shot animation sequences [本撮], and request backgrounds done
-
final-compositing, アニメ動画特効, アニメ撮影 (celfx, 2dfx, lighting, "BOOK", camera/cels motion, etc..)
-
final-dubbing, audio-visual remix
- around 20s ~ animation studio leading/ OVA Original Video Animation/ trace machine is sort of copy printer
but end of the century animation studio leading turn into form anime production committee, also widespread drop analog coloring procedure
- around 2004 ~ digital system like (RETAS/Animo, Adobe), digital douga procedure, small part of CGI
lots manga/lightnovel/game/visalnovel adaption
- around 2016 ~ tried full CGI/crowdfunding/small part of AI
digital production system, modern celfx, 2dfx and global streaming.
refers to (CGWOLRD デジタル作画, sunrise, toei, SANJIGEN digital production system) and (MdN 2017.11)
widespread use outsourcing and digital in-production procedure (動仕, 2原, even LO!)
- nowdays ~ ONA Original Net Animation, indie anime
at present, japan animation industry really need some type of "motivation", not only improve every animator job pay in japan, or just increase in-production efficiency, there so many factors.
start with a news 'All animation production may be replaced by AI' AT-X, President Iwata. https://www.itmedia.co.jp/news/articles/1702/13/news073.html
DeNA use zunko crowdfunding animation Production material for AI In-Between Animation
https://www.youtube.com/watch?v=X9j1fwexK2c
https://www.ieice.org/publications/conference-FIT-DVDs/FIT2018/data/pdf/CH-008.pdf
https://repository.dl.itc.u-tokyo.ac.jp/record/54835/files/48176425.pdf
Dwango use 4 cuts of Idol Incidents Production material for AI In-Between Animation
shinkai makoto think AnimeGAN interesting
https://twitter.com/shinkaimakoto/status/1293130397312643072
Geek Pictures/GEEK TOYS animation use cinnamonAI developed AI for coloring tracing
https://cinnamon.is/news/20210119_geekpictures/
toei animation experimental research use PreferredNetworks developed AI for stylization background drawing
https://cgworld.jp/feature/202104-actftaaf01-2.html
toei experimental research PEROs & OLM R&D developed AI for coloring tracing