ByteDance's Seedance 2.5 Skips the Stitching

PLUS: Google's Nano Banana 2 Lite arrives and Microsoft quietly swaps in its own AI

Author

Gen AI Art
July 14, 2026 • Estimated Reading Time: 3 minutes

Sponsored by

ByteDance just skipped four version numbers to make a point. Seedance 2.5 generates native 30-second clips at full 4K resolution in a single pass — finally killing the stitch-together workaround that’s plagued every AI video tool before it.

The bigger flex is control: up to 50 reference inputs versus the old model’s 12, plus audio generated in the same pass as the visuals instead of bolted on after. If the quality holds up outside cherry-picked demos, is this the model that finally makes AI video good enough for real production work instead of just social clips?

Today in AI:

  • ByteDance’s Seedance 2.5 generates uncut 4K video

  • Google’s Nano Banana 2 Lite speeds up image generation

  • Microsoft quietly benches OpenAI and Anthropic in Office

The best voice models, now across all channels

Most CX platforms do not own the voice. They orchestrate a workflow, then call a third party for speech and transcription. Every hop adds latency, cost, and another vendor to manage.

ElevenAgents is the opposite. They make the voice models the market builds on, and ElevenAgents puts full orchestration on top. Voice, transcription, text-based chat, and reasoning run in one vertically integrated pipeline, so responses come back in <400 milliseconds and sound human, not synthetic.

Plus, you keep full control. Plug in any LLM, integrate tools, webhooks, and MCP servers, and ground responses in your knowledge base. Get an agent live in minutes, then A/B test with Experiments, enforce Guardrails, and version every change.

The payoff: more human conversations, lower latency, and far less time stitching infrastructure together. You build on the models you already trust. Pricing is transparent and flat at $0.08 per minute.

What’s new? ByteDance unveiled Seedance 2.5 at its Volcano Engine FORCE conference — a video generation model that renders a full 30-second clip at native 4K in a single diffusion pass instead of stitching together shorter segments.

What matters?

  • The model accepts up to 50 multimodal reference inputs — images, audio clips, 3D models, and style references — up from just 12 in the previous version, giving creators far finer control over style and motion.

  • Audio and video are generated together in the same latent space, so on-screen actions and their sound effects land in sync without a separate audio pass.

  • An enterprise beta is live now through BytePlus, with a public launch targeted for mid-July.

Why it matters?

Stitching has always been the tell-tale seam in AI video — the moment continuity breaks and gives away the trick. A model that holds 30 seconds together natively, in 4K, pushes AI video closer to something a real production could actually use.

GUIDE

What’s new? Google rolled out Nano Banana 2 Lite, its fastest and cheapest image generation model yet, built for developers who need speed and volume over maximum fidelity.

What matters?

  • The model turns a text prompt into an image in about four seconds, priced at $0.034 per 1,000 images for high-volume pipelines rather than one-off art.

  • It’s rolling out across Google AI Studio, the Gemini API, and consumer surfaces like the Gemini app, NotebookLM, and Google Photos, so the same model powers developer tools and everyday apps alike.

  • Anyone can try it directly in Google AI Studio without writing a line of code.

Why it matters?

Fast, cheap image generation matters more than another quality leap right now — it’s what lets apps generate images on the fly instead of making users wait. Nano Banana 2 Lite is Google betting that ubiquity beats prestige.

SPONSORED BY LINDY

“Who is this person again?”

You’ve had that moment. Walking into a call, scrolling through old emails, trying to remember what you promised. Lindy texts you a brief 15 minutes before: attendee context, past discussions, open items, talking points. All pulled automatically. Try Lindy free.

What’s new? Microsoft has started routing Excel and Outlook AI prompts to its in-house MAI models instead of OpenAI or Anthropic — marking the first disclosed production shift away from its longtime AI partners.

What matters?

  • The swap targets high-volume, low-complexity tasks — drafting email replies, summarizing threads, generating simple spreadsheet formulas — while frontier-grade reasoning requests still route to OpenAI or Anthropic.

  • Microsoft hasn’t disclosed which model answers which request, so Office users may already be getting MAI-generated results without knowing it.

  • The move builds on Microsoft’s broader MAI model family, seven in-house models spanning reasoning, coding, image generation, and voice that launched in June.

Why it matters?

This is Microsoft testing how much of its AI bill it can bring in-house without anyone noticing the difference. For a company spending billions on outside API calls, even shaving off the commodity-task volume adds up fast.

Everything else in AI

Google launched Video Remix in Google Photos, letting Gemini Omni restyle ordinary clips into watercolor, oil-painting, or relit cinematic versions with one tap.

ElevenLabs rolled out a new Eleven Music toolset headlined by Voice to Song, which turns a hummed or sung recording into a studio-quality track.

1X unveiled a tendon-driven robotic hand for its Neo humanoid with 25 degrees of freedom and tactile sensors precise enough to pour tea and sort grapes.

AI International Film Festival screened 10 AI-made films and music videos selected from 250 submissions across 35 countries at its Hollywood celebration.

Essential AI Guides - Reading List:

Let us know!

Work with us

Reach 100k+ engaged Tech Professionals, Engineers, Managers and decision makers. Join brands like MorningBrew, HubSpot, Prezi, Nike, Ahref, Roku, 1440, Superhuman, and others in showcasing your product to our audience. Get in touch now →