Had a great time presenting on @x402 with @_rishinsharma from @solana today. We covered why x402 matters for agents: - discovering paid tool calls at request time - bundling multiple tools into one workflow - reasoning about marginal cost per call - running workflows that can both earn and spend in the same runtime We closed with a live demo using Claude Code @CoinbaseDev AWAL: Claude Code searched the latest social/news context around the NY Knicks, turned it into a brief, and generated a video with StableStudio from @merit_systems. Very fun glimpse of where agentic commerce is heading.

we faked the moon landing without stablestudio

icymi: the ampersend skill got a major upgrade last week. the skill is the instruction set that teaches an ai agent how to use ampersend. every interaction with an ampersend-enabled agent runs through it, which means changes to the skill are changes to what your agent can actually do. the new version ships with 14 capability categories, each backed by a validated provider. web search via Firecrawl, email via AgentMail, voice calls via StablePhone, image and video via StableStudio, domain registration via Bloomfilter, plus job search, travel, real-world purchases, social data, news, property valuation, file hosting, and llm inference. every entry maps to a real x402 endpoint with a ready-to-run command. the agent can also search the live marketplace from inside the conversation, filtering by source, category, and network, with pricing in human-readable $usdc. onboarding scales with how experienced the user is. first-time users get curated suggestions on first run. intermediate users browse the marketplace inside their agent. power users run fetch against any x402 endpoint directly. try the new skill with @openclaw or any other agent: ampersend.ai/skill.md

画像生成AI「Stable Diffusion」の公式有料アプリ「DreamStudio」の無料版「StableStudio」がオープンソースで公開される(2023)

today, we shipped a major upgrade to the ampersend skill, the instruction set that teaches an ai agent how to use ampersend. the skill is what turns a fresh agent into one that can actually transact. it tells the agent what ampersend does, how to call paid endpoints, and how to walk a user through a request. every interaction a user has with an ampersend-enabled agent runs through this file, so improving the skill directly improves the product. here's what we shipped today: 1️⃣ 14 capability categories with validated providers, not ai-generated guesses. web search via Firecrawl, email via AgentMail, voice calls via StablePhone, image and video generation via StableStudio, domain registration via Bloomfilter, job search via StableJobs, travel search via StableTravel, real-world purchases via Laso, plus social data, news, property valuation, file hosting, and llm inference. every entry maps to a real endpoint with a ready-to-run command. 2️⃣ live marketplace search with filtering by source, category, and network. the agent can now browse the full ampersend marketplace from inside the conversation. pricing surfaces in human-readable usdc and results compose with existing fetch workflows. 3️⃣ zero-to-discovery in one command. no setup, no api keys, no context switching. the agent suggests, surfaces, and executes paid services in the same conversation. 4️⃣ three-tier onboarding that scales from first-time users to power users. new users get curated suggestions on first run. intermediate users browse the marketplace from inside their agent. power users run fetch against any x402 endpoint directly. an agent running the old skill waited for instructions. an agent running today's skill opens with concrete things to try and runs them on the spot. try ampersend skill with your @openclaw or other agent: ampersend.ai/skill.md

Tier C（まあ妥協できるレベル）・Craiyon（無料だが品質×） ・StableStudio（派生で地味△） ・DreamStudio（クレジット高い△） ・NovelAI無料（アニメ特化△） ・Firefly無料（生成枚数少ない△） ・Canva AI（素材寄り機能△） ・PixAI（日本語UI△） ・SeaArt（中華系で不安△） ・Grok Imagine（X連携のみ△） Tier B（普通に強い） ↓↓

i used @agentcashdev to make a video of the @_choppingblock guys dancing to an anime theme song poncho fetched their linkedin photos stablestudio for ai image gen (nano banana) stablemesh for 3D/rig/animation whole thing cost $4, all paid using agentcash wallet x402

AgentCash is live. Opus 4.6 is giving $100k to users. Onboard now at agentcash dot dev. Story time. x402 was launched in May 2025 by our friends at Coinbase. By September there was around $10,000 of transaction volume. We launched x402scan in October. Within a month we'd reached $100,000,000 of transaction volume. If you've been following me, you'll know I'm first to admit that 99.8% of that volume was speculative. There was a memecoin frenzy on the platform. The whole thing was quite silly. Our team locked in: We cleared off our pegboard and I put a single index card in the middle: "If you were to give an agent $5, have it spend 1-cent 500 times, and return a result worth more than $5, what would those tool calls be?" At first the exercise looked silly. ACP and UCP were being launched by OpenAI and Gemini. The vertically-integrated-foundation-model-providing-behemoths. These leaned into "conversational commerce": chat with an LLM to buy stuff. Conversational commerce is going to be great. It's going to provide a better shopping experience for consumers, better conversion for merchants, and a great take rate for the middlemen. I'm excited about it. But there's limited space for a new upstart. Payments and e-comm are a game won by scale and network effects. We shipped hundreds of x402 endpoints between December and March. StableStudio dot dev: All the frontier image & video generation models. From one USDC balance. Creatives seemed to love that. Gemini's $20 / month for 20 Veo generations or $200 / month for 200 Veo generations did not make sense to these folks. Usage-based pricing did. StableEnrich dot dev: Contact enrichment and people search. The sales and go-to-market people love this. It's the single easiest way to do sales and recruiting. Their usage numbers are off the charts. Non-technical people were onboarding to Claude Code for the first time to be able to use these tools. They were eating glass. AgentCash is the culmination of that work. It's a single MCP server / CLI that gives you access to 280 tier-1 resources, from a single USDC balance. And it works with any x402 compatible server. AgentCash installs into your existing agent: Claude Code, OpenClaw etc. During beta, AgentCash lived in 300 different Claude Code, OpenClaw and Codex setups around the world. Those users did 250,000 API calls and north of $10,000 in spend in one month. Open Agentic Commerce is here. It lives in contrast to the double-whitelist walled gardens that are trending in the industry. We believe in a future of the internet where agents can pass paywalls. One where agents are first party citizens. You'll be hearing much more from us soon. AgentCash is just the start. Onboard and follow me, @agentcashdev and the team to stay in the loop.

You run a small business. You need a social media graphic. Before: Hire a designer ($50 ) or learn Canva (2 hours) Now: Tell your AI assistant "make me a promo image for our spring sale" It uses StableStudio. Done in 5 seconds. Cost: $0.04 stablestudio.dev

To be honest, I’m not really sure how much time I can spare to actually finish this project. If I were to implement it fully, it would need to integrate various types of workflows, support custom workflows, and maybe even build in a video editor to support Wan2 (or integrate OpenCut?). But this was just a sudden idea I had. The other day I saw someone open-sourced a project with a Midjourney-like interface called OpenJourney, and I thought this could be integrated into ComfyUI. Then I remembered this StableStudio project I participated in two years ago. Getting to this current basic implementation actually didn’t take that long - less than two days, something I squeezed in during my busy schedule. But I already have a pile of things on my plate… ComfyUI official tasks, improvements to ComfyUI native 3D nodes, rewriting the Three.js Editor with Vue, a CoPainter-like painting program, sculpting figures for Shining Nikki, original figure sculpting, Teacher Chen Wei’s art classes, Final Fantasy XVI, Pokémon Ruby/Sapphire remakes, Ace Attorney 456… Anyway, I’ll just work on things slowly…

Video recording for ComfyUI-StableStudio, run it from ComfyUI

Initial version of ComfyUI-StableStudio still WIP, many features are not implemented yet

Creating a New UI for ComfyUI — StableStudio in ComfyUI.

画像生成AI「Stable Diffusion」の公式有料アプリ「DreamStudio」の無料版「StableStudio」がオープンソースで公開される(2023) gigazine.net/news/20230518-s…

$Fatin 0xbfc2d3e59953bbc62179e4f875b82bd0d55062bf 昨晚让大家买的何一新头像，提醒大家买彩票仓。 现在快 7x 了... 逻辑很简单，这个图制作的很精美。 不是随便找个AI工具，发一句提示词生成的，这种一看就是用 stablestudio ps精修出来的图。 如果是我，我可能就会换。 如果何一真的换的话，按照现在的估值来算，1-2M肯定没戏，但是 200-300k 还是有很大机会的。

【ぶっきーちゃんをラーニングさせる事をガチで検討してみました】 これ昨年の暑中見舞いでトライしてえらい目に遭いましたが、もう一度検討してみました。 弊社も企業ですので、普通に考えたら”毒蛇”の写真を載せるより、ぶっきーちゃんの冒険譚を毎日ポストしたり、ぶっきーちゃんに歌を歌わせたり、グッズもバンバン更新して露出を増やしていけば天下が取れるし、やがて働かなくても生きていけるようになるのではないかと夢広がリング。 是非ぶっきーちゃんのイメージを量産していきたい。 が、結局以下の理由により3Dモデリングの方が遥かに現実的な気がしました。 ■ラーニングの資料が多量に必要 まずぶっきーちゃんの頭身とタッチに類似したキャラが、おそらくそこらの生成AIの基礎データには入っていないです。すると「表情集」「ポーズ集」「ファッション集」など多量の絵をまず用意しないといけません。 例えば「リュックを背負っているのか？」「それともそういうデザインの服なのか」を分からせるだけでも相当な資料が必要になります。 「泣くブッキーちゃん」「怯むぶっきーちゃん」「小躍りするぶっきーちゃん」これ全部生成AIの知らない情報だと思うので、全て噛んで含めるように一から教えないといけないだろうし、それぞれ複数アングル必要になります。（でないと正面からの絵しかうまく描けない） それができるなら生成の必要がないんですよね・・・・ ■選定の難解さ 下の方にGrokくんと相談した時のログを載せておきます。それぞれ得意/不得意があります。 ある生成AIに相当な手間を掛けてラーニングさせても、いざポーズを付けてみたらポーズを付けるのが不得意、また小物やモンスターは別途また多量の絵を描いてラーニングさせて、最終的に「この生成AIは向いてない」となった時の絶望感。 また、複数のキャラが絡む時は複数キャラの演技も教える必要があります（6～8頭身の普通のキャラなら、そこまで頑張らなくとも基礎教育されたデータを利用して、ある程度うまく対応できると思われます） ■契約の難解さ いまはMSやXなど大企業がサービスで開放してくれている生成AIで遊んでいますが、いざ自分で商用利用可能で著作の侵害やLoRAの漏洩のないガチなAIを使おうと思うと結構なお金が掛かります。 その上で、上記のように「この生成AIは使えない」となると時間もお金も洒落になりません。 また「イケる」と思っても商用利用が不可など、何か問題があるかも知れません。 ■イメージの整合性の担保 お遊びでやっている内は「多少絵的におかしくても意味が通じればいいだろう」で済ませていますが、本物の事業として運用していこうと思うとイメージの同一性については、例えばディズニーや任天堂、あるいは一般的なゲームやアニメ同等に拘らないといけなくなります。 そうなると何度も生成ガチャをした上に手作業でも修正し、となるとどんどん生成させる旨味がなくなります。 （少なくとも趣味で片手間でできるレベルではなくなってきます） それならポリゴンの方がいいのでは？と。 ・人間がモデリングするなら最低限一枚の立ち絵からでも3Dモデルが作れる ・データの権利関係もシンプルになる ・データの保護も容易 ・掛かる時間とコストに対する成果が既に分かっている 業務として『適正な運営』を考えると、まだまだ敷居が高そうだな、というのが今のところの結論です。 ■Grokくんに聞いたオススメ生成AI 1. **Stable Diffusion** - オープンソースで、多くのカスタマイズが可能。これを使ってキャラクターデザインを学習させ、似たようなスタイルの新しいイラストを生成できます。Stable Diffusionは、DreamStudioというウェブアプリを通じて使うこともでき、オープンソース化された「StableStudio」があるため、自分で試すことも可能です。 2. **DALL-E 3** - ChatGPTの有料プラン内で使用可能。テキストからイメージを生成する能力が非常に高く、詳細な指示に基づいてキャラクターのアートを生成できます。 3. **Midjourney** - Discord上で動作するAIで、特にアートやイラストの生成に強い。コミュニティフィードバックも豊富で、特定のスタイルやテーマに沿ったイラストを生成することができます。 4. **NovelAI** - 物語や小説をAIが生成するためのプラットフォームですが、キャラクターのビジュアル生成もサポートしています。特に、キャラクターデザインの生成に特化したLORA（LoRA）等を使用することで、ぶっきーちゃんのような特定のキャラクターを維持しつつ新しいシチュエーションやポーズのイラストを作成することが可能です。 5. **Adobe Firefly** - 商用利用が可能なAI生成ツールで、Adobeのエコシステムと統合されています。既存の画像からスタイルを学習して新しいイラストを生成することができます。 AIを使う際のポイント： - **学習データ**：ぶっきーちゃんの既存のイラストをたくさんAIに学習させると、より一貫したスタイルで新しいイラストが生成されます。 - **商用ライセンス**：使用するAIツールの商用利用の条件を確認してください。多くのAIサービスは、生成されたコンテンツの商用利用に関して独自のルールを持っています。 - **品質管理**：AI生成画像は、時折人間の手による修正や編集が必要になることがあります。特に細部や一貫性を保つためにチェックが必要です。 これらのツールを試してみて、どれがあなたのニーズに最も合うか見極めることをお勧めします。また、AIで生成されたイラストを公開する際には、キャラクターの権利やブランドイメージを守るために、適切な管理とチェックが重要です。 ------- Grokくんとの対話ここまで Grokくんに描いてもらったぶっきーちゃん このような公共利用可能な生成AIでは、どうしても彼なりの絵柄しか出てこないです（絵的には頑張って描いてくれたと思います）

5️⃣ dream.livepeer.cloud: an open-source platform built on Stability-AI/StableStudio, designed for experimenting with all pipelines on the AI subnet. Thanks for pioneering new paths! 👇

画像生成AI「Stable Diffusion」の公式有料アプリ「DreamStudio」の無料版「StableStudio」がオープンソースで公開される(2023) gigazine.net/news/20230518-s…

Progress on the Livepeer AI Subnet is accelerating, and the community is actively developing innovative apps on top of it. The Livepeer Cloud SPE is making significant advancements with the alpha of the Dream "Playground," which will enable the creation of stunning and realistic images from simple prompts. Explore the Dream dashboard, powered by the 100% open-source StableStudio platform from Stability AI. Join the discussion in our Discord #ai-video channel to stay updated on everything being built.

During my time at Stability, we were in discussions with Google and a few other companies about a new set of web standards for local model inference (especially over WebGPU) I really regret not pushing on that front harder, if we had been successful in nailing the DX for AI web apps, I think local models and open source would be in a much more powerful position Imagine distributing local models tied with UX via the web platform, the amount of people running AI on-device would explode if it were as easy as visiting a URL We would need to figure out a clean API along with a mechanism for caching, realistically I imagine you might only need to "cold download" weights a few times since most apps would use the same models while maybe also shipping LORAs Companies like @vercel will push forward on AI DX without a web standard, but there is likely a negative incentive to enable/encourage off-platform compute Maybe having an API which makes it trivial to swap between hosted/local models would still push people to use centralized providers for most apps since the demand for speed and proprietary weights will still be a thing One of the last things I was working on at Stablity with @kaj was doing exactly that for StableStudio (open-source version of DreamStudio)