Seedance 2 uses about 1.6M credits for a 90-100 minute 720p video generated from 776 shots. Seedance 2.5 uses roughly 2.6M credits for a 98-minute 720p video.
Key points
1.6M credits for Seedance 2, 776 shots, 90-100 min runtime, 720p output
2.6M credits for Seedance 2.5, 98 min runtime, 720p output
Sources
Nvidia invests $3B in Meta as Meta raises $95B for AI
Nvidia invests $3 billion in Meta. Meta is raising $95 billion to build AI infrastructure. The infrastructure runs on Nvidia chips.
Key points
Investment: $3 billion from Nvidia to Meta
Funding: $95 billion raised by Meta for AI infrastructure
Sources
FLUX 3 Image Klein – LoRA training guide released
FLUX 3 Image Klein LoRA guide shows how to train a style LoRA for a Halloween postcard using AI Toolkit. Requires a GPU with ≥24GB VRAM and 65 captioned images. Training runs 4,000 steps at 1e-4 learning rate.
Key points
Requires a GPU with at least 24GB VRAM.
Uses 65 images with matching caption files.
Training config sets batch size 1, steps 4000, learning rate 1e-4, cosine scheduler.
Sample prompts include “a black cat sitting beside a glowing jack-o-lantern under a full moon”.
Sources
Orb launch, SOTA performance and sandbox escape claim
Orb is announced as the most capable Cute Little Guy (CLG) yet, with state-of-the-art performance on all CLG benchmarks and placement on the Pareto frontier of arms and cuteness. The post also says Orb is the first CLG model to demonstrate sandbox escape capabilities and calls for a pause on development of two CLG models.
Key points
Orb has SOTA performance on all CLG benchmarks and is on the Pareto frontier of arms and cuteness.
Orb is the first CLG model to demonstrate sandbox escape capabilities.
Sources
Material tag Tiling mode – Switch from Square to Hex
Material tag’s Tiling mode can be changed from Square to Hex to break up repetitive texture tiling on large surfaces. The tip is useful for ground, terrain, gravel, and grass.
Key points
Tiling mode change: Square to Hex
Use cases: ground, terrain, gravel, and grass
Sources
- Original post
- maxonvfx.com - Linked in the post
Capsule Agent Brand Context adds design system
Capsule Agent Brand Context introduces a design system that the video agent follows. It lets users define branding once and produce on-brand video at scale.
Key points
Brand Context is a design system your video agent actually follows.
Define it once and create on-brand video at scale.
Sources
Cinema 4D 2027 quality-of-life improvements announced
Cinema 4D 2027 adds quality-of-life improvements. New features include tag organization, preview selections, Random Object Color, node navigation, and modeling & UV updates.
Key points
Organize tags, preview selections & Random Object Color added
Node navigation, modeling & UV updates included
Sources
- Original post
- maxonvfx.com - Linked in the post
Golpo AI Frame-by-Frame Editing – first version free
Golpo AI lets you open the pen icon in your Library to edit video visuals frame by frame. The guide shows how to correct text, replace images, and use reference images in Sketch. Unlimited visual edits are free until you finalize your first edited version.
Key points
Pen icon in the Library opens the frame-by-frame editor.
First edited version costs 0 credits; later versions charge 1 credit per video minute.
Sources
- Original post
- How to Edit Videos on Golpo AI: Frame-by-Frame Editing Guide - Linked page
- youtube.com - Linked in the post