01
Fire
Ogne Огне
Tiger-dragon
Hot-tempered and caustic — he hides his fear behind rage.
Live · dracats.ru
IP & Content
Solo + AI · ~2 days
Dracats
Дракоты. Эра Стали и Искр
An illustrated audiobook app and a whole fantasy IP around it — full-cast voices, sound design, a cinematic trailer, songs, four languages. One author, one book, built solo with AI in about two days.
4
Languages
~17h
Produced audio
2,827
Time-synced sentences
489
Sound effects to the word
01 · The challenge
One author. One book.
Make it a world people can hear, read and share.
Aleksey Stepikin wrote a YA dark-fantasy novel. The usual path — a PDF, a publisher, a year of waiting — leaves a story silent. The brief was the opposite: turn the manuscript into a finished product and a brand, fast, without a studio, a cast or a budget.
Claude acted as the entire engineering and production team: app, narration, sound, translation, editing, video, landing pages. The author stayed in the director's chair — approving voices by ear, correcting stresses, rewriting the ending. Built in ~2 days (29–30 Sep 2026), then iterated live on his feedback.
Manuscript+AI production team=App · Audio · Film · Brand
02 · The world
Aetheria,
age of steel
and sparks.
A human empire — the Inquisition — pumps the planet's magic, Aether, out of the ground to feed its machines. Against it stand six Dracats: giant winged big-cat dragons, guardians of the elements, awakened 500 years ago by the Great Mage's soul shards.
Aetherpunk, YA dark fantasy, for readers 12–30. Every hero got an element, a colour, a voice and a sound — the whole product is keyed to these six palettes.
01
02
Time · Magomatics
Charo Чаро
Emerald leopard-dragon
The eldest. Reads the world as formulas — magic as mathematics.
03
Earth
Skala Скала
Stone lion-dragon
The pacifist of the six — a stone giant who would rather not fight.
04
Lightning & wind
Molna Молна
Cheetah-dragon
The youngest of the six, riding lightning and wind.
05
Ice
Ledo Ледослава
Snow-leopard dragon
A lone snow leopard, cold and proud. Fur and ice-crystal scales.
06
Antimatter · Void
Echomir Эхомир
Black panther-dragon
The sixth. Captured by the Inquisition and locked inside the Juggernaut.
From photo reference
to character sheet
Before a single scene was rendered, every character got a photoreal reference sheet — anatomy, fur, scales, armour. Ledo alone went through four variants until “D”: fur plus ice-crystal scales. The sheets keep all stills and video on-model.
The Inquisition
Steel against Aether. The antagonists got the same treatment — and their own sound: servos, metal, static.
General Weyland
The Inquisition's general. A mechanical left arm.
Dr. Isabella Crow
Cyborg bio-engineer.
The Juggernaut
White-gold war golem with a black-hole reactor.
The Silencers
Anti-magic mercenaries, and the Inquisition's army behind them.
03 · The product
Listen. Read.
Together.
A free PWA that started as a Claude Design prototype and became a real audiobook player: an audio engine, sentence-level sync, offline chapters. “Together” mode highlights every sentence as it's spoken — karaoke for a novel.
51 scene illustrationsHero cardsWorld glossaryCollection & badgesBookmarks & quotesOffline per chapterSleep timerSpeed controlLock-screen controlsResume by sentence“Listen next” flow4-language UI, auto-detect
Same scene.
Four languages.
Russian by default, English, Chinese and Hindi picked up automatically from the browser. Same illustration, same sentence, same sound effect — in every language.
РусскийRU · 4h 26m
EnglishEN · 4h 04m
中文ZH · 4h 00m
हिन्दीHI · 4h 50m
04 · The hard part
One wrong stress
breaks the spell.
Free TTS engines mispronounce Russian — and the author demanded perfect stress: «очистнОго, not очИстного». A narrator who stumbles on a word pulls the listener out of the world.
edge-tts→Silero→Yandex SpeechKit+RUAccent
So we built a stress pipeline: RUAccent, a dictionary and human-verified homographs — plus a proofreading page: tap a word, fix the stress, re-voice only that phrase. Every author correction was applied book-wide.
~1,800
stress & ё decisions
169
context homographs reviewed by hand
1 tap
to fix a word and re-voice its phrase
✕ Machine
очистного
✓ Author
очистного
✕ Machine
дракоты
✓ Author
дракоты
Even the title had to be taught: the product is дракотЫ, and every one of 2,827 sentences was voiced against the same dictionary.
05 · Sound design & voices
Cast by ear.
Not one narrator, but a full cast: narrator, six heroes, villains, soldiers and machines — 15 voice presets in Russian, chosen with the author in listening rounds, each with its own sound design. English, Chinese and Hindi got separate casts with the same treatment.
VoiceSound effect on the exact word
- NarratorMystical male
- OgneGrowling baritone + growl layer
- SkalaDeep bass
- CharoMystical
- EchomirVoid echo
- MachinesMetal
- EN · ZH · HISeparate neural casts, same sound design
489
sound effects placed to the exact word via Whisper word timings
~90%
land on the exact word
≥14 dB
voice above the ambience, auto-ducked and checked per chapter
88
CC0 sounds in a stereo ambience layer
06 · From cartoon to cinema
Stills that
feel like film.
From the reference sheets came photoreal scene stills (Qwen-Image-Edit) and then motion (Alibaba Wan 2.5, image-to-video with lip-sync). The heroes speak on screen.
A 35-second trailer with a real score.
Photoreal AI shots, heroes speaking with lip-sync, and stereo trailer music built the classic way — braams, taiko, a riser, a sub drop. Cut from the same assets: a vertical cut for stories and a full music video.
0:35Cinematic trailer, 16:9, stereo
9:16Vertical stories cut
3:25Music video «Inside Power» — Ogne's arc
07 · One book → four languages
Aligned
sentence by sentence.
English, Chinese and Hindi — about 29,000 words each — translated 1:1 with the Russian book, under one shared glossary. Sentence N is the same moment in every language, so illustrations, sound effects and bookmarks stay in sync. Done in a day.
RussianYandex SpeechKit
4h 26m
EnglishNeural voices
4h 04m
ChineseNeural voices
4h 00m
HindiNeural voices
4h 50m
Sentence 1Sentence 2,827
~29k
words per translation
1:1
sentence alignment with the Russian book
1
shared glossary for names & canon
~17h
of produced audio across four languages
08 · The brand layer
Built to be
shared and found.
An audiobook nobody finds is still silent. So the product ships with its own distribution: songs people replay, cards people post, and pages that AI assistants can read and recommend.
Songs
10 songs about the Dracats
An anthem in Russian and English, hero songs and hits — with lyrics, right inside the app.
- AnthemRU · EN
- Hero songs6 heroes
- Inside PowerMusic video
Share cards
Real cards, deep links
Quote card 1080×1350 and hero story card 1080×1920. Every card links to the exact sentence, hero or song.
«Tell your master I'm going to melt you down into souvenir keychains!»Ogne · dracats.ru
SEO · GEO
Readable by AI assistants
Static landings in 4 languages, hreflang, sitemap, JSON-LD (Audiobook, VideoObject, FAQ), robots.txt open to AI crawlers, llms.txt and llms-full.txt.
# dracats.ru/llms.txt
> Дракоты. Эра Стали и Искр
@type: Audiobook, VideoObject, FAQ
hreflang: ru · en · zh · hi
robots.txt: AI crawlers welcome
llms-full.txt: the whole canon
09 · By the numbers
Two days,
one universe.
~17 h
of produced audio in four languages
2,827
sentences, every one time-synced to text
489
sound effects placed to the exact word
19
chapters — new prologue «The Sixth», 17 chapters, finale
100
scenes
51
scene illustrations
15
voice presets in the Russian cast
~1,800
stress & ё decisions
126
continuity fixes across the book
10
songs with lyrics
4
languages, UI and narration
The editing pass mattered as much as the tech: an audit of the whole book against the canon — the “500 years” timeline, species, who stands where in each battle — a new prologue, and an ending rewritten on the author's call: Charo dies and is reborn.
10 · Process
29–30 Sep.
Then live.
- ~2 days · solo + AI
- Prototype → productA real PWA from a design prototype
Claude Design prototype turned into an app with an audio engine, sentence sync and offline chapters.
- VoiceStress pipeline & casting by ear
edge-tts → Silero → Yandex SpeechKit; RUAccent + dictionary; voices picked with the author in rounds.
- Sound489 effects to the word
Whisper word timings, CC0 library, stereo ambience, auto-ducking checked per chapter.
- CanonWorld consistency audit
126 fixes, a new prologue, a new ending.
- LanguagesOne book → four languages in a day
EN, ZH, HI, sentence-aligned, one glossary.
- CinemaFrom cartoon to cinema
Reference sheets → photoreal stills → video with lip-sync → trailer, stories, music video.
- BrandSongs, share cards, SEO/GEO
Then iterated live with the author's feedback.
Stack
★
ClaudeEngineering & production team
R
React PWAService worker, offline, Range audio
Y
Yandex SpeechKit v3Russian full-cast narration
E
edge-ttsEN · ZH · HI neural casts
Á
RUAccentStress & ё pipeline
W
whisper.cppWord timings for SFX & sync
W
Alibaba Wan 2.5Image-to-video, lip-sync
Q
Qwen-Image-EditPhotoreal stills from references
ff
ffmpeg + numpyMixing & mastering
fs
Freesound CC088 sound effects & ambiences
YC
Yandex CloudObject Storage, Functions, DNS, TLS
$0
Free tiersHosting, reviews, Metrika
Free · 4 languages · offline
Hear the
Dracats roar.
The book is live and free. Start with the prologue, «The Sixth».
Have a book, a world, a product?
Stepikin Studio turns ideas into shipped products — design, code, content and distribution, solo + AI, in weeks rather than quarters.