AI Video Feed

Sharing exciting new things about GenAI video

[Black Forest Labs @bfl_ai]

Introducing FLUX 3.

One multi-modal model for Image, Video, Audio and Action-Prediction. Creations are truer to life in every kind of style.

FLUX 3 Video is now available in early access (link below).

Jointly trained in one unified architecture, our model can be extended to predict actions for robotics. See our work with mimic and Audi in the thread.

[Siqi He @starryhe214]

Definitely a great combination: Seedance 2.0 + Google Omni.
Use Seedance for stunning, cenimatic multi-shot video, and then Omni for faster, cheaper video editing—perfect for testing more scenes.

Share prompt in comment

[Min Choi @minchoi]

Holy smokes... how is this AI?

[AI Video Feed @aivideofeed]

Try it this week:

Open the free MetaView Space: https://t.co/INaiBC6zy2
Upload one clean, well-lit photo
Set yaw and pitch for the new angle you want
Similar tools: Luma, Runway
5 minutes and no cost to see if it fits your shots.

4/4

[AI Video Feed @aivideofeed]

Turn one photo into a scene you can orbit around.
Kwai MetaView builds new camera angles from a single image.
It holds geometry even on big viewpoint swings.

1/4

[AI Video Feed @aivideofeed]

My read: most useful for creators who need parallax or a camera move from a single still, product shots, thumbnails, establishing frames. Quality dips on extreme angles, so I'd keep it to b-roll and quick mockups for now.

3/4

[Amir D @starks_arq]

We made a short film that asks one question: "what is reality?"

For a year, every AI film we made answered it for us: you could still tell.

Then BytePlus handed us Seedance 2.0 in native 4K, and we couldn't.

Filmmakers have a word for footage clean enough to ship: final pixel.

AI video has never been there, because resolution was always the ceiling.

This is the first time the frame held everything we asked of it. It finally looks the way we imagined it.

"Reality is our workshop. Imagination is our tool."

Watch REALITY in 4K, then try it yourself:

@BytePlusGlobal @lumina_ai_aiart

[AI Video Feed @aivideofeed]

How to try it this week:

- Grab code and weights
- Confirm a 24GB GPU (a 4090, or rent an L4/A10 on Runpod)
- Run the demo, type an action, then edit the text and waypoints live
- Adjacent tools if you want hosted: Cascadeur, DeepMotion. Setup: an afternoon.

4/4

[AI Video Feed @aivideofeed]

NVIDIA open-sourced ARDY: type an action and the character does it live.
Change the text mid-walk, drop a waypoint, it re-plans instantly.
Runs locally on 20-24GB VRAM. 🎥

1/4

https://t.co/UIrKbpkEX2

[AI Video Feed @aivideofeed]

My read: the part that matters is editing motion while it plays. Most text-to-motion tools bake a clip, and you regenerate to change anything. Live re-planning changes how you block a scene. I think this hits games and previz before it hits film.

3/4

[World Cup Anime @WorldCupAnime]

The Reason Why Sorloth Didn’t Pass To Haaland.

[PJ Ace @PJaccetturo]

Everyone said AI ads would never win at Cannes.

Two of them just did. Both built on Kling.

Here's why it's a bigger deal than the headline: 🧵👇

[AI Video Feed @aivideofeed]

How to try it this week:

- Open Topview and start a project in 3D Shot Composer.
- Drop in your characters and props.
- Position the camera for the framing you want, set angle and distance.
- Generate the video from that setup.

4/4

[AI Video Feed @aivideofeed]

@TopviewAIhq added a 3D stage where you place characters, props, and the camera before you generate.

You block the shot in 3D, then render the video from that exact framing.

Fewer rerolls to land the composition you wanted.

1/4

[AI Video Feed @aivideofeed]

I think 3D blocking is where AI video is heading. Placing a virtual camera gives you the exact frame a text prompt can't guarantee, and you can reuse the setup across shots for consistency. I'd bet the interfaces that win end up feeling more like a viewport you direct.

3/4

[AI Video Feed @aivideofeed]

ComfyUI launched an MCP: Claude can now run its node workflows directly.

My take: the MCP doesn't replace the workflow, it replaces the busywork. Claude just executed the graph.
Prompting and model choice become the skill. Node-wiring stops being one.

[ComfyUI @ComfyUI]

We just launched Comfy MCP in public beta, the first MCP built for production pipelines.

Connect Claude, Codex, Cursor, or Hermes to the entire ComfyUI ecosystem.

→ Run any workflow in natural language

→ Search models, nodes, and templates

→ Share a workflow URL - your agent picks it up

→ Re-run saved workflows on new inputs without touching a single node

→ Hundreds of pre-built workflows. Always auto-updated

20 product-placement images for your brand ads. Character design for your game art. Script-to-shot ideation at scale and so on!

Your agent is now a creative technologist.

[AI Video Feed @aivideofeed]

Try it this week:
- Capture or generate a 3D Gaussian splat scene (Luma AI or Polycam both export compatible formats).
- Open it in SuperSplat (free, browser only).
- Export as 360 video and upload to YouTube as a VR180/360 video.

4/4

[AI Video Feed @aivideofeed]

3D Gaussian splats can now export as 360 video, no extra software.
The feature just shipped in SuperSplat (https://t.co/VQUf8tPmCS), PlayCanvas's free web editor.
Turn any splat scene into a 360 video walkthrough.
1/4

[AI Video Feed @aivideofeed]

My take: this closes a real gap for anyone using Gaussian splats for virtual production or archives. Before this, sharing a splat scene meant sending people a link to an interactive viewer. A 360 video works everywhere: YouTube or Meta Quest that reads equirectangular video.
3/4