Kling 4.0 完整版放出实机片段，10 月发布前预热

AlexAImaginator @TraffAlex原推发布于 2026.09.30

在 X 上打开原推

🎨 AI ART & VISUALS ROUNDUP — September 30, 2026 1️⃣ RUNWAY JOINS OPENAI MARKETPLACE AS LAUNCH PARTNER Runway announced it is joining OpenAI's new marketplace as a launch partner. Starting today, creators and teams can access Runway's image, video, audio, and editing models through the marketplace storefront, and part of an existing OpenAI enterprise commitment can be applied toward purchases made with Runway. For studios already spending through OpenAI, that turns Runway's suite into a single-billing, single-procurement path — no separate contracts needed. 🔹 @runwayml 2️⃣ KLING 4.0 SHOWS FULL POWER IN ACTION Ahead of its October release, Kuaishou's Kling team shared footage of the full 4.0 model in action. The teaser pushes on the upgrades the team promised — higher visual realism, tighter creative control, and more complete narrative scenes — and lands right after the 4.0 Flash variant started rolling out to Ultra yearly subscribers last week. The gap between the Flash preview and the flagship is now visible in production-quality clips. 🔹 @Kling_ai 3️⃣ KREA AGENT SPOTLIGHT: WINDOW LAB MOTION-CONTROL APP Krea kept promoting its Agent-built app ecosystem with "Window Lab," a tool where you simply upload a video and then dial in dedicated motion controls until the movement plays out the way you want. It is a small but telling example of how the Krea Agent app system is turning narrow post-production tweaks into shareable one-click utilities — each app built by the Agent itself rather than by Krea engineers. 🔹 @krea_ai 4️⃣ PIXVERSE LAUNCHES EARN PROGRAM FOR AGENT VIDEO WORKFLOWS PixVerse opened an earn program tied to its Plugin: show what the PixVerse Plugin finished — an ad, a short film, or a brand video — and approved posts earn a base payout, with view-driven bonuses that can scale up to a full three-figure sum on short-video platforms and a mid-three-figure sum on X and YouTube. It is a practical way to push adoption of agentic video pipelines: creators get paid to document the workflow, and the feed gets filled with real Plugin output. 🔹 @PixVerse 5️⃣ PRUNA OPEN-SOURCES 6.3× FASTER QWEN-IMAGE-2.1 ADAPTERS Pruna released Pruna-Qwen-Image-2.1 as open source — a set of few-step LoRA adapters that make Qwen-Image-2.1 up to 6.3× faster for generation and editing. Instead of the usual 40 steps, images can be generated or edited in 5 or 8 steps, with the 8-step adapter as the recommended default: 1K-resolution text-to-image plus single or multi-image editing with up to three reference images. The adapters load directly on top of the base model, keep the pipeline unchanged, and each runs without CFG on its own optimized sigma schedule. 🔹 @PrunaAI 6️⃣ RUNWAY AGENT GAINS ASSET-LEVEL TAGGING Runway shipped Agent Tagging: you can now tag the Runway Agent directly next to your assets to request edits, alternatives, or fixes in place. It shifts the agent from a chat box you have to brief in prose into a collaborator you point at a specific element and critique — closer to how an editor leaves notes on a timeline. 🔹 @runwayml 7️⃣ A 10-HOUR-REFINED OPUS 5.5 VIDEO-GENERATION PLAYBOOK A creator published a dense, hands-on playbook for using Claude Opus 5.5 as an orchestrator for full video production: drive it from Claude Code rather than the app, route all image, video, and audio model calls through one OpenRouter key, use GPT Image 2.5 Sunburst for keyframes with Veo 3.1 or Seedance 2.5 for generation, and assemble with OpenTimelineIO. The stack adds skills for script planning, ASR-based word-level subtitles, and an automatic critic that screenshots and transcribes the output to validate quality — and the post closes with a four-minute business explainer built entirely in this pipeline. 🔹 @deedydas 💭 The thread running through all of this: visual AI is becoming infrastructure rather than a tool. Marketplace integrations, asset-level agent tagging, and creator earn programs are all the same bet — the competitive edge is moving from raw model quality into how deeply a product plugs into your billing, pipeline, and daily workflow. Meanwhile, open speed LoRAs and publicly shared playbooks keep lowering the floor, which means a solo creator today can chain a 40-step model down to 8 and assemble a four-minute explainer with nothing but a laptop and an API key. Which of these would actually change your workflow first — marketplace access, asset-level agent tagging, or faster open-source models? 👇 #AIArt #AIvideo #AIGraphics