生成AIでIPチャレンジ

生成AIの気になるニュース、ツイートをクリップしています。
生成AIによる広告制作、マイクロドラマ、ゲーム、生成画像/生成動画/空間生成サービスを開発・事業化しています。
（ここでの発言、発信は、個人活動です）
Sony MVP 2004, Sony MVP2009

[生成AIでIPチャレンジ @GenAI_Creative]

minimax H3 MAX で、
インタラクティブ生成AI恋愛ゲームを作りました。

・リアルタイム（もう少し）、インタラクティブ生成

・設計されたゲーム空間でなく、生成空間を自由に無限にプレイ

・ユーザの実時間と連動したエピソード展開（向こうの彼女も同じ時間）

・季節や、イベント、時間や時事ネタなどのセリフ

・サプライズプレゼント（プレゼントしたジュエリー、洋服、花束などの着用動画生成にも対応）

・親密度が上がるにつれて積極的なデートも

・このゲーム内のアパレル、ジュエリー、アクセサリーのプロダクトプレイスメントの広告スポンサーを募集しています。

[NVIDIA AI @NVIDIAAI]

Available now: NVIDIA Nemotron 3.5 Bolt E4B 4GB ⚡️

Our newest small open model is built for local assistants, interactive characters and robots that can carry on a conversation and use tools to get things done.

Watch Reachy Mini chat and move, with Bolt handling conversation and NVIDIA Cosmos handling vision. Both models run locally on Jetson Orin Nano 8GB.

Learn more: https://t.co/B8ODaWMn0O

[fal @fal]

Kling 4.0 is now publicly available on fal, with API access open to all customers.

The latest model from @Kling_ai brings:

• Up to 30 seconds of native video generation with audio
• Up to 15 image, video, and element references
• Precise control with up to 10 keyframes
• Improved motion, character consistency, and cinematic storytelling

Start building with Kling 4.0 on fal today.

[Hao AI Lab @haoailab]

As promised, FastVideo open-sources @MiniMax_AI FastH3 Omni, training code included! 🚀

- Real-time gen @ 12x faster on NVIDIA Blackwell
- NVFP4/FP8 for Spark + RTX!
- ComfyUI day-0

Collab with @NVIDIAAI , Nuva Lab @nuvalab and Reactor @reactorworld

Open source must win! ♥️

[生成AIでIPチャレンジ @GenAI_Creative]

"AIバブル崩壊寸前、流動性の「枯渇」を警告する7つのサイン" - Forbes JAPAN 公式サイト（フォーブス ジャパン） #SmartNews https://t.co/YaHw7EmR97 #スマートニュース

[David Im @davidim]

Plans on the pricing of Sume API:

1. We're keeping 0% markup.
We've checked our prices against the foundation model providers more than 20 times. If you find any model where our price is different from what the foundation model company charges, tell me and we'll fix it.

2. "Is this business model actually sustainable?"
Yes. Even at 0% markup we're not losing money, so keeping it is not a problem. Because of the competition, video model prices will converge to 0% markup anyway. We're just doing it first.
Our plan is to bring in users by providing models, and later offer a separate SaaS product subscription that covers marketing workflows. So markup on video models stays at 0%.

3. "Kie AI is cheaper on some models?"
If you look at Kie AI, they act like they're giving big discounts on cheap models, but they put a pretty big markup on frontier models like Seedance. My guess is they heavily discount some low-usage models to look cheap, bring users in with that branding, then make their net profit from the markup on Seedance and others. We hate that kind of trick and care about transparency, so it's 0% markup on everything. No tricks.

As you know, other providers play games. The "most accessible" pricing only gets you that price if you subscribe annually to credits that expire every month. And Higgsfield looks cheap with their discount events, but when you buy top-up credits, they take a huge markup.

We learned trust is the most important thing in business. As an early startup, we're promising 0% markup to earn that trust. In return, just recommend us to your friends!

[生成AIでIPチャレンジ @GenAI_Creative]

"元miHoYoのAI研究責任者が設立した企業が1,000万ドル超を調達―3Dバーチャルアイドルと交流・育成するゲームを開発中" - Game*Spark #SmartNews https://t.co/APi3nY88e0 #スマートニュース

[生成AIでIPチャレンジ @GenAI_Creative]

サイバーエージェント: Cloud Run や Agent Development Kit を活用し、クリエイティブ生成 AI プラットフォーム「AI SCREAM」のエージェント化を推進 @googlecloud https://t.co/skUZrYx7lA

[MiniMax (official) @MiniMax_AI]

MiniMax M3.1 Flash weights are now public. 428B parameters total, 23B activated per token.

Agentic performance: A generational leap in agentic tasks, coding, and cybersecurity — 76.2 on BankerToolBench, 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2, 29.6 on Agents' Last Exam, and 82.2 on CyberGym.
Efficient by design: MiniMax Sparse Attention and DSpark speculative decoding improve inference efficiency.

Hugging Face:
https://t.co/JDYA1zZh54
ModelScope:
https://t.co/DQcmOjTwhO

API: https://t.co/fHRdSV7BwZ
M Plan: https://t.co/LYnYZcsesu
MCode for desktop: https://t.co/SGmLTI3fFU

[生成AIでIPチャレンジ @GenAI_Creative]

minimax h3などは止まっちゃうリスクもあるのね🤔⌨️🖱️🖥️

"ディープシークとミニマックスのAIモデル、中国が海外展開承認" https://t.co/ECOn2tL9C3

[生成AIでIPチャレンジ @GenAI_Creative]

"OpenAI、判断処理に特化した「Decisions API」を一般提供 GPT-6 Lunaを基盤に高速化" - マイナビニュース #SmartNews https://t.co/rrbgXUmdM5 #スマートニュース

[Sadao Tokuyama @tokufxug]

防犯カメラ×オープンソースAIで視線ヒートマップを作成

🔹YOLO11-poseで人物・骨格、Gaze-LLEで視線、Depth Anything V2で奥行き推定
🔹棚ごとの注目度を可視化し、Blenderで店舗を3D再現して棚を見る時間を推定

開発：Claudio Escudero氏
※リアルタイム処理ではなく数値は推定値
情報元はリプ欄。

[生成AIでIPチャレンジ @GenAI_Creative]

映像や画像を組み合わせて理解するAIモデルであるVLM（Vision Language Model）※1を活用し、視聴者があらかじめ設定した出演者（推し）のアップ映像と、ステージ全体の演出や見どころを捉えた映像を、AIがバランスよく自動で構成します。これにより、視聴者は視聴中にカメラ映像を切り替える操作をすることなく、嗜好に合わせて構成されたライブ映像を楽しむことが可能

AIで一人一人の嗜好に合わせたライブ映像を自動構成する「パーソナライズ映像自動構成技術」を開発 https://t.co/MukJwd1viK @PRTIMES_JPより

[Claude @claudeai]

Starting next week, sessions across Claude web, desktop, and mobile apps will run in auto mode by default on Pro, Max, and Team plans.

Rather than asking you to approve each step, an automated check blocks risky actions. Switch back to manual anytime.

[Emily @IamEmily2050]

ByteDance Seedream 5.5 coming soon with big improvement.

100 Layer Scene Decomposition (Up from 10+ in 5.0 Pro): Splits dense posters, presentation slides, and motion templates into up to 100 individual raster layers with automatic occlusion inpainting behind extracted foreground objects.

80+ Rounds of Drift Resistant Local Editing: Locks unedited pixels (product geometry, facial identity, brand labels) across 80+ sequential single-variable editing passes (background swap -> relighting -> material change -> text addition) to minimize VAE re encoding degradation.

Native 4K Output at 16:9 (Up from 2K Cap in 5.0 Pro): Generates native ⁠3840x2160⁠ widescreen frames, eliminating the 2K (⁠~2.36MP⁠) resolution bottleneck on 5.0 Pro for widescreen banners and multi panel storyboard grids.

Native 4 Channel RGBA Generation (With Partial Alpha): Outputs true transparent PNG assets directly from the initial prompt, supporting semi transparent alpha values for glass refraction, soft contact shadows, sheer fabric, and fine hair strands without post generation background removal.
Optical Font & Style Recognition for Text Re-Rendering: Reads existing font families, weights, colors, and layout hierarchy inside an image so you can translate or swap campaign copy (across English, Chinese, Japanese, and Thai) and re render matching typography in place.

Sub Headline & Micro Copy Multilingual Fidelity: Sharpens stroke accuracy on dense, small-scale text specifically pricing, units, dates, bottle volume markings, and promotional fine print where 5.0 Pro was limited to larger poster headlines.

Relighting Consistent Material & Portrait Shaders: Upgrades surface physics for ceramic micro texture, brushed metal anisotropy, glass refraction, and skin pores so that shifting lighting conditions (such as daylight to nighttime) updates reflections and contact shadows without smoothing skin texture or warping product labels.

Role Assigned Multi Reference & Negative-Space Planning: Formalizes reference routing by assigning distinct jobs to each input image (Reference A = subject identity, Reference B = color palette, Reference C = pose, Reference D = layout) while automatically reserving clean negative space for headlines and CTAs.

[OpenAI Developers @OpenAIDevs]

Add near real-time decisions to your workflow with the Decisions API, now generally available.

Build faster, more cost-efficient apps and agents that can classify text and images, evaluate inputs, and determine what happens next without generating a full model response.

[生成AIでIPチャレンジ @GenAI_Creative]

ご利用は計画的に。

[生成AIでIPチャレンジ @GenAI_Creative]

画像生成は、https://t.co/RyyH1GhZDkの方が安そう。

[生成AIでIPチャレンジ @GenAI_Creative]

画像生成は、https://t.co/RyyH1GhZDkの方が安そう。

[生成AIでIPチャレンジ @GenAI_Creative]

10月23日時点　1ドル160円換算で比較してみた。

[生成AIでIPチャレンジ @GenAI_Creative]

10月23日時点　1ドル160円換算で比較してみた。

[生成AIでIPチャレンジ @GenAI_Creative]

kieよりさらに安いのか。ホントか

ーーー
たとえばSeedance 2.5 1080pの場合
• falより50%安い
• higgsfield apiより50%安い（最大割引でも29%）
• kie aiより28.02%安い

ルーティングインフラ層は決して堀ではありません。私たちは、APIは商品化されるべきだと信じています。そして、堀はアプリケーション層で構築されるでしょう。

[David Im @davidim]

yes.

we payed the full creation rights, and not charging it to our users

unlike hig.....

[生成AIでIPチャレンジ @GenAI_Creative]

中国AIチップ市場、ファーウェイが米エヌビディアに肉薄　輸出規制下で国産勢が拡大(36Kr Japan)
#Yahooニュース
https://t.co/ViNGuk1o4f