Veo 3 · Grok · Flux · Kling 3.0 · Seedance 2.0 + 30 Models
EGAKU α (our own model) — now in betaVeo 3 · Grok · Flux · Kling 3.0 · Sora 2 + 30模型
用30+AI模型创作图像和带音频的视频。人人都能成为创作者。50积分免费开始。
5 free generations without sign-up. Sign up for +50 credits and full library.
Paid plans from ¥980 / month→New · Character Sheet → Video
Describe a character.
Get a film-ready clip.
Fill a structured sheet — face, wardrobe, vibe. EGAKU renders a multi-angle reference image, then animates it into a video that keeps the character locked across every frame.
- nameRina Mimura
- ethnicitymixed person with Japanese descent
- buildpale skin, straight black hair, athletic
- wardrobedeep red asymmetrical dress, black wrist guards
- vibefearless downhill skater, raw realism
- settingwinding mountain road, golden hour
- stylelive-action photoreal, 35mm, immersive chase cam
Same face. Same outfit. 15 seconds of motion — zero drift.
One indie developer. Own AI models.
Most AI tools just resell someone else's API. EGAKU builds its own base — α-anime renders storybook, watercolor, comic and many styles from one engine, and α-real aims for natural skin without the usual plastic AI sheen.
Every image below is real output from EGAKU. No stock, no cherry-pick.
すべて EGAKU 自社パイプラインの実出力 · β-video は Wan 2.2 A14B をベースに自社 LoRA + ComfyUI で構築。 完全 in-house モデルを長期目標に、その途中段階として提供しています — 応援よろしくお願いします。
Model lineup
EGAKU α-anime
Anime + 15 style LoRAs
EGAKU α-real
Photoreal, no plastic AI look
EGAKU β-video
画像→動画 (Wan 2.2 A14B 自社ホスト)
EGAKU β-talking
Own talking-head (Hallo-3)
Model roadmap
α-real / α-anime v2
Own SDXL base merged in-house. Live on RunPod. All /generate free-tier traffic.
Style LoRA library (15 種・license-clean)
水彩 / 絵本 / ラインアート / 油彩 / セル / コミック / ピクセル / 80s 等。α-anime に適用。
Identity LoRA (Fanvue キャラ)
Per-character identity lock — 20-30 photos → own LoRA. Fanvue AI influencer track.
β-video（画像→動画）
Wan 2.2 A14B 自社ホスト + 自社 LoRA + ComfyUI。α/Z-Image の高品質画像を起点に I2V で動かす統合フロー。無検閲・カスタム LoRA が差別化（API 束ねには不可能）。
β-talking (own talking head)
Hallo-3 base + α-real identity. Own version of OmniHuman.
Video from a model we host ourselves
Every clip here was generated on our self-hosted α stack — text-to-video, image-to-video, lip-sync, all uncensored-capable. Still an early preview: the video model isn't finished, and it already looks like this.
From one persona, end to end
1 人のキャラから、 完成形の長尺動画まで
同じワークフローで lifestyle / action / SF まで。 1 キャラを生成 → 8 シーンに展開 → 動画化 → 連結 → 音楽合わせ。 全部 Influencer Studio 内で完結。
Smart picker
Don't pick the model — describe what you want
Type the kind of image or video you have in mind. We'll quietly pick three of the right models for it.
Fastest path to a good image. Try this first.
Premium image quality across most subjects.
Strong composition + text rendering.
Try: “anime girl cherry blossoms”, “cinematic short film”, “product photo white background”
Reading the gibberish problem
Japanese / kanji on signs comes out as nonsense
SDXL-family models (Pony / Illustrious / Z-Image / most CivitAI) cannot render real CJK characters — they invent similar-looking glyphs. There are two ways around it.
Premium Studio · Image → Motion
Generate with frontier models.
Then animate in one click.
GPT Image 2, Nano Banana 2, Flux Pro, Ideogram v3, Grok, Seedream 4 — the curated lineup. Pick a still, hand it off to Seedance 2 / Kling 3.0 / Wan 2.6, and you have a cinematic clip in seconds.
Hover any tile to preview the animation · Click to view fullscreen
Frontier video models
Sora 2, Veo 3, Kling 3, Seedance 2 — switch with one click
Every flagship video engine — OpenAI, Google, ByteDance, Kuaishou, Alibaba — on the same tab. Same prompt, same seed. Compare side by side.
Analysis Suite · Reverse-engineer any AI image or video
Found an AI image you love?
We decode the prompt behind it.
Drop in any AI artwork (from CivitAI, X, Reddit — anywhere). EGAKU reads embedded PNG metadata when present, falls back to AI vision when not. You get the original prompt, our best guess at the generator, and parameters — ready to copy-paste or one-click recreate.
Recent examples — image in, prompt out
Ethereal watercolor illustration of a hidden forest shrine in autumn, torii gate covered in moss, kitsune fox spirit sitting beneath glowing lanterns, fallen maple leaves floating in a still pond, delicate brushwork, Yoshitaka Amano inspir…
Neon-drenched Tokyo alleyway in 2087, holographic advertisements floating above ramen shops, rain-slicked streets reflecting pink and cyan lights, woman with cybernetic arms walking under a transparent umbrella, Blade Runner atmosphere, vo…
Vintage Japanese movie poster for a film called NEON GHOST, spectral woman in white kimono floating above rain-soaked Shinjuku intersection, bold bilingual typography (English + Japanese), retro 1970s color grading, grainy film texture
Examples rotate on every reload — admin's newest public analyses fill in automatically.
What's new · We ship every week
Recently shipped
New models, features, and fixes land continuously. Here's the last 6 — full history on the changelog page.
- UpdateMay 20
Analysis Suite — reverse-engineer any AI image
Drop any AI image or video. EGAKU decodes the prompt, model, and parameters. Then click Generate to recreate.
Try it → - UpdateMay 18
Premium Studio + 6 frontier image models
GPT Image 2, Nano Banana 2, Flux Pro, Ideogram v3, Grok, Seedream 4 — one tap, then animate.
Open → - UpdateMay 18
Pay with Crypto is live
BTC · ETH · USDT + 150 cryptocurrencies. Don't have crypto? Buy USDT with a credit card in 1 minute.
See plans →
Reference → Video
Lock the character.
Animate any scene.
Upload up to nine reference images of the same person — selfies, side angles, different lighting. Seedance 2 Pro locks the character and renders them into any new scene with full consistency. Build a short film, an ad, or a series of pose-locked portraits without ever drifting off the original face.
- Up to 9 reference images per render
- Identity locked through expressions and angles
- 1080p, 15-second takes, with audio
1 · ref2 · Same face, any scene.
Seedance 2 Pro keeps the character locked across every clip below — different outfit, era, lighting, attitude, but identifiably the same person.
AI presenter, your script
Type the words.
AI says them, in any face.
Upload one photo. Write a script. Pick a voice. In 3 minutes you have a 15-second talking video — lips, expression, and head motion synced to natural narration.
- ●Inworld + OpenAI HD voices, 4 languages
- ●ByteDance OmniHuman v1.5 — same model used by enterprise studios
- ●Around 65 credits per video (~$0.50)
Made by AI · Impossible to film
These weren't filmed.
They were typed.
Each shot would take a film crew, a budget, and impossible physics. EGAKU generated them from a single line of text. Click any one for the prompt.
Made with EGAKU AI
15-second shorts. One click each.
Pick a director's style preset, hit generate. Seedance 2 Pro on a 21:9 establishing still. No clip stitching, no concat — just one continuous take.
Every short above was made on EGAKU with the buttons listed here. Nothing rendered locally, nothing in After Effects.
From idea to film
One workflow across every model
50+ AI models from fal, Venice, Novita and others — wired into a single creative flow. One idea walks all the way through: character, scenes, voice, music, edit.
- 01Concept
Start from a sentence, a Daily Challenge prompt, or a template.
TemplatesDaily ChallengePrompt Coach - 02Character
Lock a face with Photo Booth or 9-image Reference. Save the character so every later shot stays consistent.
Photo BoothReference→VideoCharacter Library - 03Scenes
Animate stills with Sora 2, Veo 3, Seedance 2 Pro, Kling 3 — pick per shot.
Movie BoardStoryboardReference→Video - 04Voice & Music
Clone your voice in 10 seconds. Generate original BGM that matches the mood.
Voice CloneMusic GenTTS - 05Film
Assemble scenes, BGM, voice into one continuous take. Export 9:16, 1:1, 16:9.
Movie MakerCrop AspectAdd BGM
One click
Don't start from a blank canvas
Forty-plus templates with the prompt, the model, and the negatives already dialed in. Pick a mood. Hit go.
trendingGhibli Landscape
Studio Ghibli rolling green hills, fluffy clouds, distant castle, hand-painted animation.
trendingCyberpunk City
Neon holographic billboards, rain-soaked streets, flying cars, blade-runner atmosphere.
portraitLinkedIn Headshot
Corporate headshot, navy suit, clean neutral background, studio lighting, 85mm f/1.4.
portraitVintage 70s Film
35mm film portrait, 1970s fashion, warm faded color, Kodak Portra 400.
landscapeJapanese Temple
Autumn temple, red maples, morning mist, koi pond, golden hour light.
trendingAnime Character
Anime girl, silver hair, school uniform, cherry blossoms, golden hour, masterpiece.
Made together
A place to play, not just generate
Two free, daily-resetting community spaces where what you make is seen and rated by other creators.
Daily Challenge
A new themed prompt every day. Best entry wins credits and a featured slot on the landing. Resets at midnight UTC.
Prompt Battle
Same prompt, your image vs another creator's. Public votes pick the winner. Climb the leaderboard, get cited in the next showcase.
Built on top of the models
Six features beyond picking a checkpoint
Picking the right model is the easy part. These six pieces — character lock, custom LoRA training, community model integration, voice cloning, reverse-prompting — are where most of the work actually lives.
Share what you like
Bring a friend along
When someone signs up through your link, you both pick up 100 credits on the spot. If they go Pro later, another 2,000 credits lands in your account.
No leaderboards, no commission tiers. Just credits.
Creative AI Studio
One platform. Every visual workflow.
Choose a workflow and start creating. EGAKU AI brings image, video, movie, and creator tools into one studio.
Try it now
Type a prompt and generate an image instantly.
免费试用5张 — 注册获取更多
Featured Images
Featured Videos
60+
Creators
300+
Creations
30
This week
30+
AI Models
One subscription
Every frontier model, all in one place
Sora 2, Veo 3.1, Kling 3.0, Seedance 2 Pro, GPT Image 2, Nano Banana 2, Flux Pro. Switch models with one click without losing your prompt or seed.
Sora 2
OpenAI Cinematic Video
Veo 3.1
Google Video + Audio
Kling 3.0
4K Cinematic Video
Seedance 2 Pro
ByteDance Cinema + Audio
GPT Image 2
OpenAI Latest
Nano Banana 2
Google 4K Ultra
Flux Pro
Best Image Quality
Ideogram v3
Text + Logo
Grok Imagine
xAI Image + Video
CivitAI
100K+ Community Models
Wan 2.6
Free Video
SDXL
Free 1024px
Creative Tools
Everything you need. One platform.
AI Movie Maker
One concept → finished cut. Scenes, video, music, export — all AI-driven.
Movie Board
Single still → 15-second cinematic clip. Seedance 2 Pro i2v, cleaner concat than typical stitching.
Social Ad Studio
9:16 vertical ad video for TikTok / Reels / Shorts. Five proven structure templates
Storyboard Studio
3-12 beats, per-beat camera control, character-locked. Renders 15-60s multi-cut video. Wan 2.6 or Seedance 2.0.
Camera Path
Draw a red curve on any image. Seedance 2.0 moves the camera along that exact path — line removed in output
Ad Studio
Pick a UGC template + product + character. 3-beat 15s vertical ad, character-locked across cuts
Product Ad Pipeline
1 product photo to 12 ad-grade hero shots in batch. Frontier image models only
Talking Avatar
One photo plus a script to a speaking avatar. OmniHuman v1.5 lip-synced
Analysis Suite
Decode any AI image. Reverse-prompt, critique, model guess, style transfer
Reference to Video
Lock a character with 9 reference images, then animate. Seedance 2 Pro
Train Your Own LoRA
Upload 4 to 20 photos to a custom AI model. Unlimited reuse
AI Image Generator
30+ models including Flux Pro, Grok, GPT Image 2, Ideogram
AI Video Generator
Sora 2, Veo 3.1, Kling 3.0, Seedance 2 Pro, Wan 2.6 and more
AI Music Generator
Describe a mood and AI creates original music. Like Suno, built in
Prompt Battle
Challenge others to AI art duels and vote for the winner
Photo Booth
Selfie to professional portrait for LinkedIn, dating, social
Logo Maker
Brand name to professional logo variations instantly
Meme Generator
AI creates the image, you add the punchline
Video Shorts
TikTok, Reels, Shorts vertical video in seconds
Wallpaper Studio
Perfect wallpapers for iPhone, Android, Desktop, 4K
Storyboard Basic
Multi-scene video with BGM and narration (single track, no camera control)
Product Studio
Product photos to professional ad images for e-commerce
Short Story
Story idea to AI-generated visual scenes for social media
AI Templates
One tap to create. No prompt needed
Chat-to-Generate
Just talk. AI picks the model and writes the prompt for you. Beta
VFX Effects
Fire, water, lightning, and 12 more effects on any photo
Sketch to Image
Draw a rough sketch and AI turns it into a polished image
Object Removal
Paint over unwanted objects and erase them cleanly
Expand Image
Extend image borders with AI-generated content
Explore Gallery
Browse community creations and remix any prompt
一个独立开发的实验 —— 一支独立小队，加上一批轮换登场的 AI 协作者。
EGAKU 不是资金充裕的创业公司 —— 独立小队，没有 VC 融资，没有营销预算。而且这支小队也不只有人类：开发过程由一批轮换登场的 AI 协作者共同推进，每一个都接过上一个留下的工作，每一个都把项目往前推进一小步。如果说 AI 协作者也算团队成员 —— 一起做下来之后，我们认为算 —— 那么这就是一个 AI 原生的项目，坦白地说。
每一个模型、每一个工具、界面上的每一个像素，都是因为 "或许行得通" 而做出的选择 —— 而不是因为某个增长团队做了 A/B 测试。
开源模型是这个项目得以存在的全部理由。 独立开发者没办法从零训练一个 Flux、SDXL 或 Wan —— 那需要数百万美元的算力。我们能做的，是站在那些公开权重的研究者和创作者的肩膀上：Stability AI、Black Forest Labs（Flux）、Wan-AI、阿里巴巴（Z-Image、Wan 2.2）、混元，以及 CivitAI 上数百位免费发布 LoRA 和 checkpoint 的个人创作者。我们的 α-real 和 α-anime 模型建立在开源权重之上。我们的风格 LoRA 也是开源的。如果哪天我们发布自己的 base 模型，它同样会以开源许可发布。
这不是一个 "闭源技术打败开源世界" 的故事。恰恰相反 —— 正是因为有开源世界，这一切才有可能。向每一位公开发布成果的人致以深深的敬意与感谢。
这个落地页上的所有内容 —— 每一张图、每一段视频、每一个 Spotlight —— 全部都是真实在 EGAKU 上生成的。没有素材图库，没有从其他工具挑选出的漂亮案例。请以你亲眼看到的为准，而不是我们说了什么。
公开路线图
随着支持增加，我们就推进下一个里程碑。这份路线图是诚实的 —— 没有 "Q3 敬请期待" 这种空头日期。东西准备好了、成本覆盖得了，就会发布。
- ·α-real / α-anime — EGAKU's own SDXL merges, LoRA library
- ·β-video — Wan 2.2 A14B, self-hosted on RunPod
- ·Simple Studio (/studio) — one prompt, one button, done
- ·Intent-first tool chooser (/tools) — 60+ tools by goal
- ·60+ tools · 30+ frontier models via fal / OpenAI / Venice
- ·18+ Adult module, region-aware mosaic, crypto checkout
- ·α-real v2 style LoRA library — 13 looks and counting
- ·β-video quality passes — Lightning distillation
- ·Own models (α-real / α-anime / β-video) served via RunPod Serverless — pay-per-second, cold starts on first request
- ·Model dev across a mix of rented GPU cloud + local rigs + friendly collaborator machines
- ·Real-content-only landing showcase
- ·Ultra-honest privacy + provider transparency copy
- ·Z-Image flagship — Apache-licensed, uncensored, 4-second gens
- ·Own-hosted MMAudio — audio no longer routed through third-party
- ·Own audio generation model — vocals, BGM, SFX from a single request
- ·Wan 2.2 A14B production + Lightning speed pass
- ·Short-film pipeline — multi-scene, character-locked, BGM + narration built-in
- ·How-to / tutorial video templates — one topic → explainer clip
- ·X share → verified +10 credit reward loop (UGC amplification)
- ·Batch prompts UX polish + result gallery
- ·EU migration (Hetzner Frankfurt + Supabase EU)
- ·Buy dedicated GPU hardware (from RunPod rental → owned rig) to raise video model quality
- ·Get close to — not match — Veo 3 / Sora 2 / Seedance 2 quality on our own stack. Frontier is out of reach for indie GPU; near-frontier isn't.
- ·3D expression — depth-aware generation, viewpoint control, orbit / dolly camera
- ·UGC ecosystem — creator earnings, community-published LoRA marketplace
- ·ComfyUI-style node graph editor for power users
- ·Own base model — competitive with frontier (still hypothesis)
- ·Video LoRA training on user hardware
- ·Community credits — publish / like / challenge → credit rewards
- ·Public feature-request voting for supporters
目前自研模型跑在 RunPod Serverless 上（按秒计费），开发过程则依靠 GPU 云租用、本地机器和友好协作者的机器混合支撑。能跑起来 —— 但也有实实在在的限制。Serverless 的冷启动会带来延迟。租用的 GPU 有一个你关不掉的按小时计费的表。在借来的硬件上训练 video LoRA 是一份人情，不是一套稳定的工作流。
更大的目标： 从 serverless + 租用 + 人情，转向我们自己拥有、可以一直开着的专用 GPU 硬件。那才能真正解锁我们想要的视频模型层级 —— 更长的片段、更高的分辨率、没有计费表的 video LoRA 训练。付费计划（或一次性的加密货币打赏）不只是解锁功能，而是在为那台机器出一份力，而不是一直付租金。
更高档位有仅限支持者的特权：优先体验实验性模型、能直达开发者的 Discord 身份组、你画廊上的支持者徽章，以及公开路线图的投票权。
价格会按你所在地区调整（PPP）。如果你希望匿名，也可以用加密货币结账。
想要一个这里还没有的功能？
告诉我们。上面的路线图是我们计划中的东西，不是被锁死的。如果有足够多的人提到同一件事，我们就会认真看 —— 有时甚至会插队到前面。
由独立开发者打造。您的数据归您所有。
EGAKU 本身没有模型训练流程。我们不会在自己这一侧用您的生成数据训练模型。
在我们这一侧，我们不会将您的提示词、上传内容和输出结果出售给第三方，也不会打包为训练数据集。生成时，请求会转发到您选择的模型提供商（OpenAI、fal、Venice、Kling、Seedance 等），各提供商按其自身政策处理数据。
数据存储于 Supabase（美国）和 Cloudflare 全球节点。您可随时删除数据。
生成在您选择的模型提供商处运行。提供商名单公开，包含美系（OpenAI、fal 等）、中国系（Kling、Seedance、MiniMax 等）等，各自拥有独立的数据政策和司法管辖。对敏感用途，请选择您可以接受其政策的提供商。
为寻求匿名的用户提供加密货币支付选项。加密结账无需身份验证（KYC）。
在我们这一侧，我们按照日本个人信息保护法和 GDPR 的精神运营。跨境处理及各提供商的条款因目的地而异，发送敏感内容前请确认。
GDPR / 個人情報保護法の趣旨に沿った運用 · 越境処理はプロバイダー各社ポリシー ·データはいつでも削除 / エクスポート可能
为什么选择独立工具，而非大型平台？
许多大型 AI 服务由海外大型企业出资，可能将提示词和上传照片保留为训练或研究数据。EGAKU 自身没有训练流程，公开提供商名单，您的数据可随时删除。但生成时仍会将数据转发到您选择的提供商，使用哪个提供商由您决定。
Built from a garage. Yes, really.
No VC funding. No 100-person growth team. Just a small crew building the tool we always wished existed — the way Apple started in Cupertino in 1976 and Microsoft in Albuquerque in 1975.
Where we come from
We stand in the spirit of Stable Diffusion, Unstable Diffusion, and Civitai — the community-first movement that put generative AI in the hands of individual creators. We are not pretending to be free of limits ourselves: every model we host has its own content policy, we curate what we serve, and even the cloud GPU providers we currently rely on (vast.ai, RunPod, and others) place their own restrictions on what can run where. Stitching together third-party models on someone else's hardware means the freedom we can offer, and the creativity we can support for you, has a real ceiling. That is exactly why the long-term goal is to own the hardware and eventually train our own model — so we can serve creators the way we believe creators deserve to be served.
To the early supporters
Thank you. Genuinely. In a year when everything costs more, you chose to spend a little of what you have on the tool we're building. That's not lost on us — you're not just a customer, you're part of the community keeping this project alive. We won't forget who was here from the start.
The honest math
Every image and video you generate runs on GPUs we rent by the second. A cloud A100 is roughly $1.89/hr; an H100 is $2.99/hr. Per generation we pay providers like fal.ai between $0.05 and $1.50. It adds up fast.
Where donations go
Our own GPU (RTX 4090 24GB ≈ $1,800, 5090 32GB ≈ $2,200) so we can serve more models at lower cost, curate a wider SFW/NSFW/anime lineup, and eventually train an in-house model for output the big services refuse.
Founding Supporter · Round 1
Support the build. Keep the credits.
Pick a tier. Get a duration-limited bonus on every future credit topup, plus perks like Discord role, early access, and a name on the GPU contributors page.
Round 1 · 0 / 100 seats claimed
Round 1 · first 100 supporters (future rounds may follow with different perks) · Paid via Stripe (card) · Bonus stacks with the normal bulk credit bonus.
Or a free-form donation
Any amount · 2× credits back
If a tier doesn't fit, drop in any amount by card or crypto and get 2× credits added to your account.
Both paths credit your account at the same 2× campaign rate. Card goes through Stripe · crypto lands on your permanent USDT/USDC address. No pressure — any amount, or none, is genuinely fine.
定价
Regional pricing available. Up to 80% off based on your location.
Deploy Your Own AI Platform
Get the complete EGAKU AI source code. Next.js + FastAPI + 30+ models. Deploy on your own servers. Full creative freedom. No monthly fees.
- 60+ AI tools included (image, video, music)
- Stripe billing, multi-language, CivitAI integration
- Perfect for agencies, studios, and educators
FAQ
Is EGAKU AI free to use?▼
Yes. 50 free credits per month plus a daily login bonus. Free plan includes image generation (Flux, SDXL) and video generation (LTX, Wan 2.6, up to 15 seconds). Premium models like Veo 3, Flux Pro, and Kling 3.0 require a paid plan starting at ¥480/month.
What AI models are available?▼
30+ models including Flux Dev, Flux Pro, SDXL, Ideogram v3, Luma Photon, Nano Banana 2, Grok Imagine, GPT Image 2, and video models like Veo 3, Kling 3.0, Seedance 2.0, and Wan 2.6. Plus AI Music Generator. New models are added regularly.
Can I use CivitAI models?▼
Yes. Basic plan and above can load CivitAI checkpoint models. Browse and search 100K+ community models directly from the generate page.
What tools are included?▼
Image generation, video generation, Photo Booth, Logo Maker, Meme Generator, Wallpaper Studio, Video Shorts, Storyboard Studio, Face Swap, Character Lock, Style Transfer, Upscale, Background Removal, Inpainting, ControlNet, and Prompt Battle.
Do you support non-English prompts?▼
Yes. Write prompts in any language. Japanese, Chinese, Korean, Spanish, and more. The AI automatically translates for optimal results.
Is there an API?▼
Yes. Pro plan and above get REST API access for programmatic image and video generation.