In partnership with

AI for Real Life Weekly: Major Video Model Updates (Week of Sept 24th, 2025)

What a Week!

It’s been another crazy, busy week in the AI image and video world. In just the past couple of days, it feels like there’s a race to release models before the next one drops. Kling rolled out their 2.5 Turbo update, and I thought that was an impressive upgrade. The day before, they launched their new Avatar feature, which was also strong. But I think part of that timing was because WAN 2.5 was coming out, and that looks like it’s going to be huge. It’s already breaking the internet, with servers crashing under the demand. From my first impression, WAN 2.5 isn’t revolutionary in that no one’s tried it before, but it’s doing what Veo 3 did, and maybe better in some areas. The real question is how much better it is with movements, actions, realism, and overall performance. That’s going to take some testing and time. For now, let’s dive into each of these updates so you can see what they’re all about.

🌍 WAN 2.5: The First Real Contender to Veo 3

WAN launched version 2.5 this week, and demand was so high that the servers crashed on day one. Here’s what stands out:

• Integrated voices. Unlike other generators that only add sound effects or music, WAN 2.5, besides Veo 3, is the first I’ve seen to output full voices directly in the video. They’ve integrated the multimodal voice element. Veo 3 had the crown here, but WAN 2.5 looks ready to challenge it.

• Smooth motion and realism. The model focuses on stable motion and natural dynamics.

• True competition for Veo 3. This is the first time we’ve had a serious contender. Side-by-side testing will show whether WAN 2.5 can match or surpass Veo 3, especially with voices and motion.

WAN 2.5 Prompting Cheat Sheet

Here are practical tips and examples for prompting WAN 2.5, based on newly released guides and expert breakdowns:

Prompt Structure and Tips

• Be detailed and structured. Specify scene, audio, and visuals in a clear order.

• Direct dialogue. Write exact words for dialogue, assign lines to characters, and note speaking order. For scenes without speech, add “no dialogue” or use a negative prompt.

• Audio cues. Add background or ambient sound for realism (e.g., “soft rain tapping” or “fast-paced action music”).

• Scene details. Define camera work, lighting, mood, and physical elements for cinematic output. Example: “Wide shot of a mountain road at sunset, golden light, cyclist racing, energetic music.”

Cheat Sheet Table

Special Techniques

• Prompt weighting. Add weighting (e.g., “(prompt:2)”) for stronger results.

• Negative prompts. Rule out unwanted elements like extra speech, specific objects, or visual details.

Explicit layering of these elements produces the most natural, professional-quality video and audio with WAN 2.5.

👉 Sources: Fal Blog, Reddit Tips, X prompt examples, Official WAN 2.5

🎥 Kling 2.5 Turbo: Faster, Cleaner, Sharper

Kling just released 2.5 Turbo, and it’s a strong step forward:

• Better prompt adherence and temporal control. Handles multi-step actions, character interactions, and scene transitions more accurately.

• Fluid and stable motion. Trained for real-world physics and smoother camera moves with fewer distortions.

• Consistency with references. Colors, lighting, textures, and style remain stable across frames.

• Lower cost. Five seconds of 1080p now costs 25 credits, down from 35.

👉 See Kling’s official release notes (with prompt examples and outputs):
https://app.klingai.com/global/release-notes/2025-09-19?type=dialog

Flow Update (Sept 24): Prompt Expansion + Veo 3 Push

The latest Flow update introduces:

• Gemini inside Flow. Expand and enhance prompts directly in the platform, saving time and avoiding tool switching.

• Nano Banana integration. Edit images step by step inside Flow, eliminating the need for external tools.

• Veo 2 phase-out. Features once exclusive to Veo 2 (like nFrame) are moving to Veo 3. This raises the question: is Veo 4 on the horizon? Time will tell.

Kling AI Avatar: Talking Characters

Kling also rolled out AI Avatars. Upload a photo or pick a stock avatar, drop in audio, and generate a lip-synced performance.

Why it matters: This makes dialogue videos and branded content more accessible without hiring actors. It’s not perfect, but it’s a solid step forward.

Luma Ray3: The HDR and Reasoning Challenger

Luma just launched Ray3, a “reasoning” video model that delivers native HDR (10/12/16-bit EXR) output and evaluates its own results. It’s available now in Adobe Firefly and Luma’s Dream Machine.

Key features to explore:

• Draft mode for quick prototyping

• Upscale and “Hi-Fi” for refined shots

• Keyframes, looping, and animation controls

• Greater consistency and realistic motion

Why it matters: Ray3 is aimed at professional pipelines, not just short clips. The reasoning aspect makes this especially interesting. The model uses reasoning to plan and understand what comes next, making it a different kind of tool for building out video. If it performs well under complex use cases like dialogue, physics, and composites, it could shift how creators move from concept to final video.

What to Watch Next Week

• YouTube Shorts + Veo 3: YouTube is rolling out Veo 3 Fast into Shorts. Expect motion transfer, style blending, and lightweight generations tailored for short-form creators.

• Meta’s Vibes: Meta introduced Vibes, an AI-powered video feed where users can generate or remix content with visuals and music, then share across Instagram and Facebook. This could bring AI video creation to a massive mainstream audience.

The Bottom Line

This week brought:

• WAN 2.5 with voices and realism

• Kling’s Turbo model and Avatars

• Luma’s HDR reasoning model, Ray3

• Flow’s Gemini integration and Nano Banana tools

And the platforms are shifting too. YouTube with Veo 3 Shorts, Meta with Vibes. If you’re building AI videos, now is the time to explore. These aren’t small changes. They open workflows that didn’t exist even a few months ago.

Final Note

Things are moving fast, so fast that weeks like this are starting to feel almost normal. Don’t feel pressure to dive into every new release, because that’s a quick way to overwhelm yourself. Instead, just take note of what’s out there. If you have time, sample a little bit of each one. The more tools you’re familiar with, the more options you’ll have to pull from when you’re creating. Think of it like building a toolkit. You don’t need every subscription or model running at once. Just knowing what’s possible helps you decide what to try when the right situation comes up.

On a personal note, I’ve been enjoying taking what extra time I do have to start building out some bigger projects. At the same time, I’m developing workflows that I can share with you or anyone else who wants to follow along. Stay tuned, because I plan on releasing guides and possibly courses in the near future. These will help walk someone who’s just getting started with AI video, or even a seasoned pro who could use some extra tools and a framework to analyze and apply. I’m working on that in the background, but until next time, keep creating and keep learning.

“As I’m grinding away on these projects, the best way to support this journey right now is to check out some of our sponsors, partners, or partner newsletters by clicking their links below. Every bit of support helps keep this work going.”

Thank you all!

-Khalil

Smart dictation that understands you

Typeless turns your raw, unfiltered voice into beautifully polished writing - in real time.

It works like magic, feels like cheating, and allows your thoughts to flow more freely than ever before.

With Typeless, you become more creative. More inspired. And more in-tune with your own ideas.

Your voice is your strength. Typeless turns it into a superpower.

Download for Mac today