This week’s highlights point to a more settled phase in AI tools for creative work. Across voice, video, image workflows, and production systems, the shift is less about novelty and more about how these tools perform in actual use.

Some of these updates make creative processes easier to manage. Others are a reminder that speed, access, and convenience do not always mean long-term reliability. For creatives, the question is becoming less about what a tool can do once and more about whether it fits into a workflow that needs consistency over time.

This Week’s Highlights:

Freepik’s new 3D Scenes feels less like another image generator and more like a virtual set builder. You can start from a preset scene, your own image, a Street View panorama, a 3D model, or even a blank canvas, then place objects, move the camera, and capture multiple shots from different angles without rebuilding the environment each time. Freepik’s official docs focus on that workflow, while Freepik’s own social post describes the feature as “Powered by World Labs spatial AI.”

What it actually does

Read full details here.

Gemini 3.1 Flash Live is built for real-time voice interaction, with the focus on faster responses, more natural rhythm, and stronger performance in longer or messier conversations. It is available in preview through the Live API and is also being rolled into voice-facing products and enterprise use cases.

Read more here.

OpenAI’s reported decision to discontinue the Sora video platform points to a bigger issue in AI video right now: platform stability. What makes this notable is not only the product decision itself, but how quickly the direction appears to have changed. Just days earlier, OpenAI’s Sora release notes were still rolling out new editing features, while its developer docs now show a formal deprecation path for the Videos API and Sora 2 model aliases, with removal scheduled for September 24, 2026.

For anyone following AI video, this looks less like a small product cleanup and more like a reminder that even high-profile tools can shift course fast. Sora helped set the tone for the current wave of cinematic text-to-video, but the latest reporting suggests OpenAI is redirecting focus toward areas it sees as more commercially important, including coding and enterprise products.

Read the full article here.

ElevenLabs has introduced Flows inside ElevenCreative, a node-based creative canvas designed to connect image generation, video, text to speech, lip-sync, sound effects, and music in one workspace. The pitch is straightforward: instead of bouncing between separate tools, exporting files, and rebuilding the same sequence over and over, teams can map a pipeline once and reuse it with different inputs. ElevenLabs says Flows is available now in ElevenCreative, with API access coming soon.

For creatives, the more interesting part is not just that another platform added more models. It is that ElevenLabs is pushing a workflow-first idea. A Flow can be reused, rerun, and adjusted at the step level, which makes it easier to test variations without regenerating an entire project from scratch

For the creative community, the value here is not simply “more AI in one place.” Plenty of platforms are moving toward all-in-one positioning. What stands out is the shift toward repeatable creative infrastructure.

That matters because a lot of production time is lost in the handoff layer:

Flows is trying to reduce that friction by turning a creative process into something more modular and reusable. The upside is speed, testing, and more consistency across outputs. The trade-off is that a more automated pipeline can also make it easier to produce a large volume of average work. For creatives, the difference will still come from direction, taste, and knowing where a workflow should stay flexible instead of becoming too templated.

Read it full here.

Midjourney’s Version control lets users switch between model generations with --version or --v, and the current default is V7. The official docs now position versioning less as a technical detail and more as a practical creative choice, since each model handles prompts, style, speed, and feature compatibility differently.

The bigger shift is that Midjourney now has a clearer split between stable production use and early testing. V7 remains the default model, while V8 Alpha launched on March 17, 2026 on the alpha site only, not on the main site or in Discord. Midjourney says V8 Alpha is its fastest model so far, with standard jobs rendering about 4 to 5 times faster than earlier versions, and adds optional 2K HD generation through --hd.

Read the full details here.

Edmond Yang and the Nerve Lab team just released a new promotional film for Rednote Redgalas, inspired by the story of a sheep determined to become a sheepdog. It is a strong example of how AI can support a larger production pipeline when paired with clear creative direction, collaboration, and craft across story, art direction, sound, and post. The project was built entirely in Freepik, with Spaces used for collaboration, iteration, and finalization.

Read more here

Henry Duabrez shared JUNKYARD KING — Episode 0 is an early look at how AI video tools are moving from short-form generation into more structured storytelling. Built over roughly four weeks, the project explores how a small creative setup can develop an animation prototype with more of a studio mindset, using prepared assets, consistent visual references, and a more directed workflow. What stands out is not just the final piece, but the process behind it: using tools like Seedance 2.0, Google Flow, and Gemini to build a more intentional pipeline for series development.

Check full details here.

Taken together, this week’s updates suggest that creative work is becoming more shaped by workflow decisions, not just model quality. The more useful question now is not only what a tool can generate, but whether it supports the way a team actually works, adapts when platforms change, and helps maintain quality from draft to delivery.

For the creative community, that makes judgment, flexibility, and process design just as important as the tools themselves. The technology keeps moving, but the value still comes from knowing what to use, when to use it, and where to stay cautious.

That’s it for this week, folks.