Blender adds splats, DLSS and ripple edits

In partnership with

Welcome to VP Land! AI is moving inside the software artists already run, from Gaussian splats in Blender and multi-model generation in Premiere to coding agents in Unreal, while Disney picks Adobe to build models on its own characters.

In today's edition:

Blender adds native Gaussian splats, DLSS denoising and ripple editing
Premiere generates video, sound and music in the timeline, with Firefly and four partner model makers to choose from
Disney names Adobe its AI innovation partner
Gemini 4 Carbon, Unreal's in-editor agents, MiniMax H3-Image and free Megascans
Nithya Raman wins LA mayor on a film czar platform

Blender Gets Native Gaussian Splats, a DLSS Denoiser in Cycles and Ripple Video Editing

Blender 5.3 adds native 3D Gaussian splat support, so splats reconstructed from photos or video can be imported, processed with Geometry Nodes and rendered directly in Cycles or EEVEE. Captured scenes now sit in the same open-source package as the rest of a Blender project.

The Blender 5.3 release imports splats from PLY, SPZ and glTF files and stores them as a new point cloud type. It also brings an NVIDIA DLSS denoiser to Cycles, ripple editing in the Video Sequencer, panoramic cameras in EEVEE and an optional Projects system for grouping files and settings.

Splats are editable, with rough relighting. Each splat carries attributes for scale, rotation, radiance with opacity and spherical harmonics, and renders with an emissive shader by default. Swapping in a Diffuse BSDF gives what Blender calls "rudimentary" relighting, and renders can show inaccuracies because most splat tools work in sRGB while Blender works in linear color.
DLSS denoising in Cycles. The denoiser uses NVIDIA's DLSS 4.5 Ray Reconstruction, which Blender says brings near-final quality to the viewport and virtually eliminates flicker during animation playback. It requires NVIDIA driver 620 or newer.
Ripple editing and OpenTimelineIO. The Video Sequencer update adds a ripple overlap mode, Ripple Add, Trim and Delete shortcuts, a rewritten blade tool, hardware decoding and initial OpenTimelineIO import and export.
Fisheye in EEVEE, an effects stack in the Compositor. EEVEE now renders equirectangular, cubemap, fisheye and mirror ball cameras, and the Compositor's effects stack applies Bloom, Chromatic Aberration, Film Grain and Vignette like modifiers, with GPU playback up to 2x faster, according to Blender.

For previz and virtual production teams, a captured location can now live in the same scene as animation, lighting and the cut, though splat performance is still being optimized. Our breakdown of Blender 5.3 covers the splat attributes, shading changes and Projects system in more depth.

SPONSOR MESSAGE

The voice-to-text app that built its own speech model

Most dictation apps are a wrapper around someone else's speech model. Wispr Flow trained its own.

It's called Canto, and it was built for the way people actually talk to AI: thinking out loud, changing direction mid-sentence, half-whispering in an open office.

Wispr tested it on 10 hours of real dictations from more than 2,300 speakers. Canto had the lowest word error rate of every model tested, including models from Google, OpenAI, AssemblyAI and Deepgram. That means your prompts to ChatGPT, Claude and Cursor come through the way you said them.

Talk naturally in any app and Wispr Flow types clean, formatted text, 4x faster than typing.

Try Wispr Flow free

Adobe Premiere Adds Multi-Model Generation in the Timeline and Previews Local MCP Access for Third-Party Agents

Premiere's Generative Media Tool creates video, sound effects, soundscapes and music directly in the timeline from Adobe Firefly or partner models from Google, Kling, Runway and Luma. Adobe also previewed local MCP integrations that will let editors point their own third-party agents at native Premiere projects.

The Generative Media Tool debuted in Premiere before MAX, and No Film School's MAX roundup pairs it with the show's new Premiere news: the MCP preview and Color Mode reaching general availability. Editors select Generative Media, mark where the result should land and describe what they need, and Premiere returns editable clips without leaving the project.

Four generation tools, five model sources. Generate Video fills timeline gaps using reference frames sampled from the project, Generate Sound Effects runs on Adobe's commercially safe audio model, Generate Soundscape (beta) analyzes up to 15 seconds of video, and Generate Music (beta) offers variations with tempo and looping controls, per Adobe's launch post.
Agents work in native project files. The MCP preview covers Photoshop, Illustrator, Premiere and After Effects, and Adobe's press release says bins and sequences stay editable on the user's own computer. An agent connected to other MCP tools could pull a brief from Google Docs and share the result in Slack.
Claude cut a podcast in the keynote. Adobe showed Claude finding relevant moments to assemble a podcast edit in Premiere, MakeUseOf reports, adding that MCP arrives as a private beta with supported agents not yet confirmed; Adobe has only said early 2027.
Color Mode ships, cleanup tools enter beta. Color Mode is now generally available, Generative Remove arrives in beta for boom mics and logos, and Enhance Audio, Separate Crosstalk and Dynamic Auto Ducking are in public beta.

Adobe hasn't said which partner models serve which generation type or how partner-model usage is billed inside Premiere. We covered Premiere's generative tools and the MCP preview in more depth.

Disney Names Adobe Its AI Innovation Partner, With Custom Firefly Foundry Models Trained on Disney Characters

Disney named Adobe its official AI innovation partner in a multiyear alliance, and the two will build custom Firefly Foundry models trained on Disney's own characters, environments and visual styles.

As Variety reported, the deal gives Disney's creative teams early access to Firefly, Firefly Foundry, Premiere and Photoshop, plus dedicated Adobe engineering and research resources for its production pipeline. The companies say the models will produce "commercially safe and production-ready outputs."

Imagineering already uses Foundry. The R&D team at Walt Disney Imagineering runs a sketch-to-image model that turns hand-drawn concepts into rendered 2D concept art, plus a custom image model for "franchise-accurate" assets.
Adobe joins Disney's R&D programs. Adobe becomes a StudioLAB Innovation Partner at The Walt Disney Studios and will work with ESPN Edge on AI for sports content and ESPN's archive.
Marketing is part of the pact. Adobe gets activations across Disney parks, Disney Cruise Line, ESPN Edge and StudioLAB, according to Adobe's announcement.

Disney says the models will be "human-centered and artist-led," and the alliance puts a major studio's production work behind Adobe's creative-AI stack. We covered the Disney and Adobe alliance in more depth.

🎥 HiAPI. Unified, OpenAI-compatible API for image, video, music and text generation models, with hosted MCP access and pay-as-you-go billing.

🌄 masoc. Lightroom alternative for Apple devices, offering on-device photo and video editing, RAW support, film looks, and AI-assisted adjustments.

🔊 WaveXML. Syncs multicam footage and dual-system audio on Mac and Windows.

Browse the full tool directory

Gemini 4 Carbon, Unreal's In-Editor Agents and Archon Push AI Into 3D and Editing, as Megascans Goes Free

Google's Gemini 4 Carbon, Epic's Unreal Engine 5.9 MegaJam Preview and Selendis' Archon put agents and models to work on 3D scenes and edits, while MiniMax H3-Image and a free Megascans library add open weights and real-world scans to build with.

Agents in 3D tools

Gemini 4 Carbon. Google's frontier model for long-horizon agent work ships with a /swarm preview in Antigravity that splits one prompt across a coordinated team of agents, and in Google's demos it modeled 3D assets in Blender and built Three.js scenes that render in the browser. Introductory pricing is $2 per million input tokens and $10 per million output tokens.
Unreal Engine 5.9 MegaJam Preview. Epic's first release built around agentic workflows auto-launches Unreal MCP with expanded rendering, level, audio and UI toolsets, adds an in-editor panel that runs Claude Code, Codex, Cursor or OpenCode, and extends Python APIs to Niagara, PCG, materials and MetaSounds, per the Unreal Engine forum post. That gives VP teams more ways to script stage and scene setup.

Images and edits

MiniMax H3-Image. The open-weight model from the MiniMax H3 family generates 16MP images, up to 6272x2688 at 21:9, without an upscaler and edits from up to 16 reference images in six sampling steps, according to the model card. MiniMax claims 4K generations in 10 seconds, and it runs in ComfyUI or through MiniMax's API and Hailuo AI.
Archon. Selendis, co-founded by Steve Aoki, says its video foundation model turns raw footage into a finished edit in minutes and learns each creator's taste from their catalogue, cited references and decisions in its multiplayer editor. The company has shared no architecture, specs or pricing on its official site, and access starts with a waitlist.

Assets

Quixel Megascans. The full library, more than 23,000 assets by GameFromScratch's count, is going free on Fab, and the updated terms allow training AI models on Megascans and Megaplants. Buyers from 2025 or later get refunds to their Epic Account Balance, and Fab listings switch over through November 12.

An iPhone Running Lightcraft's Jetset Tracks a Low-Budget Bluescreen Shoot Into Unreal Engine

In a behind-the-scenes video, Lightcraft Technology breaks down a bluescreen test shoot where an iPhone running Jetset tracks the camera into Unreal Engine, placing actors inside medieval stone interiors on live monitors. Lightcraft says Jetset's $80-a-month subscription compares with $20,000 or more for dedicated tracking hardware, and Blender and Unreal Engine are free, which puts the setup within reach of indie budgets.

Stories, projects, and links that caught our attention from around the web:

🖐️ Maxon is bringing Touch support to Cinema 4D and ZBrush on Apple's MacBook Pro with PROLED XDR Display, and Cinema 4D adds Open in Spatial Preview so clients can review scenes life-size on Vision Pro, both arriving in late November.

🗳️ Nithya Raman won the LA mayor's race over Karen Bass on a platform that includes a film czar in the mayor's office, fairer indie permit fees and a push to lift California's $750 million film tax credit cap.

🎬 DGA president Christopher Nolan says it's on studios as copyright holders to fight AI training on their films, rejecting fair-use defenses, and notes The Odyssey's end credits carry an anti-AI-training notice.

📺 Fuji Television says every live-action drama it makes will use generative AI starting October 2027, turning AI video tools into a network-wide production mandate.

⚖️ The Presley estate group and Eminem's publisher Eight Mile Style filed separate suits against Suno, with the Presley suit also invoking Tennessee's ELVIS Act over prompts that clone his vocal timbre.

🤖 The Camerimage cinematography festival made AI its central theme, opening with an address from a humanoid robot, video messages on AI from Coppola, Nolan and Spielberg, and an honor for gen-AI advocate Darren Aronofsky.

GPT Image 2.6 Adds GIFs. Pika H3 Flex Goes Open Weight.

In Denoised, we test Pika's open-weight H3 Flex, compare how Google's Omni 1.2 Flash and ByteDance's Seedance treat real cinema-camera footage, and look at GPT Image 2.6's GIF output, saved references and faster Flare mode.

We also debate Epic opening Quixel Megascans to AI training and look at a GPT workflow that turns phone scans into CAD files. Our episode recap has the timestamps and takeaways.

Watch/Listen & Subscribe

Spotify | Apple Podcast | YouTube