🎬 AInVFX-News

---

📅 Episode Archive

🤿 Paint fluid simulations in ComfyUI: free LTX 2.5 IC-LoRA for smoke & fire | AInVFX September 13

Paint a first and a last frame, get a fluid simulation. ainvfx-fluid is a free IC-LoRA for LTX 2.5 that turns flat painted blobs into smoke, steam or fire, with a 5-second element generated in about a minute on a laptop GPU. Model, ComfyUI workflow and painted example frames are free on Hugging Face.

📁 Model, Workflow & Assets: AInVFX/ainvfx-fluid on Hugging Face

🔗 Resources:

The model:

• ainvfx-fluid model card • LoRA • Example workflow

LTX 2.5:

• LTX-2.5 models (gated) • ComfyUI-LTXVideo
• LTX-2 repository & community license • LTX documentation

Training:

• Musubi-tuner (ltx-2-dev branch — LTX-2) • Musubi-tuner (main)
• Pexels License • ComfyUI • GIMP

Related Tutorials:

• INNOCENCE — custom Z-Image & LTX-2.3 LoRAs • Demystifying ComfyUI

📰 Read the full article: Paint your fluid simulations: a free LTX 2.5 IC-LoRA for smoke and fire in ComfyUI

---

🎬 INNOCENCE — From Chinese ink drawings to an AI short film: custom Z-Image & LTX-2.3 LoRAs | AInVFX April 6

A film about an icicle, my son, and everything it takes to become something — just before it disappears. Made entirely with open source models as part of the Arca Gidan Prize, a one month open source AI film contest, in partnership with Banodoco, ComfyUI, and Lightricks.

📁 ComfyUI Workflows & Assets: episodes/20260406

🔗 Resources:

Arca Gidan Prize:

• Wathch all submissions • Banodoco • Lightricks • ComfyUI

Training:

• Musubi-tuner (main — Z-Image) • Musubi-tuner (ltx-2-dev branch — LTX-V)
• Z-Image model • LTX-2.3 checkpoint

ComfyUI Nodes:

• ComfyUI-SeedVR2_VideoUpscaler by NumZ/AInVFX • ComfyUI-VideoHelperSuite

Related Tutorials:

• SeedVR2 v2.5 Complete Guide

📰 Read the full article: From 20-year-old ink drawings to an AI short film: training custom LoRAs for Z-Image and LTX-2.3

---

🤿 Demystifying ComfyUI: Complete installation to production workflow guide | AInVFX December 19

ComfyUI doesn't have to be overwhelming. This comprehensive guide takes you from manual installation through creating your first complete AI workflow—image generation, multi-angle art direction, video creation, post-processing, and HD upscaling. Everything you need to go from beginner to confident.

🔗 Resources:

ComfyUI:

• ComfyUI Official Repository

Installation Prerequisites:

• Python 3.13 • UV Package Manager • Git
• NVIDIA Drivers • PyTorch

Tools & Nodes Featured:

• Z-Image • QwenImageEdit • Kandinsky-5
• KJ Nodes • SeedVR2 VideoUpscaler

Related Tutorials:

• SeedVR2 v2.5 Complete Guide

📰 Read the full article: Demystifying ComfyUI: Complete installation to production workflow guide

---

🤿 SeedVR2 v2.5: Official Guide from the ComfyUI Integration Team | AInVFX November 7

After 4 months of community-driven development, SeedVR2 v2.5 brings a complete architectural redesign. New 4-node system, GGUF quantization for 8GB GPUs, torch.compile optimization, native alpha support, and production-ready CLI make professional upscaling accessible to everyone. Breaking change but worth it.

📁 ComfyUI Workflows & Assets: Directly in the ComfyUI Template Manager or at https://github.com/numz/ComfyUI-SeedVR2_VideoUpscaler/tree/main/example_workflows

🔗 Resources:

SeedVR2 Implementation:

• Official ComfyUI Integration GitHub • ComfyUI Registry
• Installation Guide

Model Repositories:

• Main models (numz) • AInVFX models • GGUF models (cmeka)

SeedVR2 Research:

• SeedVR2 Project Page • SeedVR Paper • SeedVR2 Paper
• Official ByteDance GitHub

📰 Read the full article: SeedVR2 v2.5: The complete redesign that makes 7B models run on 8GB GPUs

---

🤿 One-Step Video Upscaling: Complete ComfyUI SeedVR2 Guide (Free workflow included) | AInVFX July 11

ByteDance's SeedVR2 transforms video upscaling with one-step restoration instead of 15-50. Complete tutorial covering ComfyUI setup, BlockSwap for consumer GPUs, alpha channel workflows, and multi-GPU processing.

📁 ComfyUI Workflows & Assets: episodes/20250711

🔗 Resources:

SeedVR2 Research:

• SeedVR2 Project Page • SeedVR Paper • SeedVR2 Paper
• Official SeedVR & SeedVR2 GitHub

ComfyUI Implementation:

• ComfyUI-SeedVR2_VideoUpscaler by NumZ
• ComfyUI-CoCoTools_IO by Conor-Collins
• ComfyUI-VideoHelperSuite by Kosinkadink

📰 Read the full article: One-step 4K video upscaling and beyond for free in ComfyUI with SeedVR2

---

🎬 Speed up WAN 2-3x with MagCache + NAG negative prompting + One-step upscale | AInVFX News June 21

Four practical techniques for faster, better video generation: MagCache accelerates diffusion 2-3x, NAG brings back negative prompting to distilled models, DLoRAL upscales videos in one step, and MIT shows how AI can be use for Art restoration.

📁 ComfyUI Workflows & Assets: episodes/20250621

🔗 Resources:

MagCache: Fast Video Generation with Magnitude-Aware Cache

• Project Page • Paper • GitHub
• WanVideoWrapper Integration

NAG: Normalized Attention Guidance

• Project Page • Paper • GitHub
• Self-Forcing LoRA

DLoRAL: One-Step Video Super-Resolution

• Paper • GitHub • Demo Video

Physical Art Restoration with AI

• MIT News • Nature Paper • Nature Video

📰 Read the full article: Speed Up Video Generation 2-3x: MagCache, NAG, DLoRAL & AI Art Restoration

---

🎬 Democratizing AI: train models with 256 NPUs, track through occlusions | AInVFX News June 18

Four groundbreaking papers democratizing AI development - from training competitive video models with 256 NPUs to tracking through occlusions, streaming video generation, and climate modeling.

🔗 Resources:

ContentV: Efficient Training of Video Generation Models

• Project Page • Paper • GitHub

CoTracker3: Tracking Any Point Through Occlusions

• Project Page • Paper • GitHub
• ComfyUI Node
• See our LEGO DeepDive for CoTracker + ATI workflow

Self-Forcing: Autoregressive Video Diffusion

• Project Page • Paper • GitHub

CBottle: Climate Foundation Model

• NVIDIA Earth-2 • Blog
• Paper • GitHub

📰 Read the full article: ContentV, CoTracker3, Self-Forcing & CBottle - Democratizing AI Development

---

🤿 Create your own LEGO animation: complete AI workflow from photo to final shot | AInVFX June 13

Transform a single LEGO photo into a complete animated shot! Join Adrien for an in-depth tutorial combining the latest open-source AI tools to bring our favorite toys to life.

📁 ComfyUI Workflows & Assets: episodes/20250614

🔗 Resources:

WAN 2.1 + ATI (Any Trajectory Instruction) + VACE + CausVid

• ComfyUI WANVideoWrapper
• WAN, ATI, VACE, CausVid Models
• WAN BF16 Models

CoTracker

• Research Paper
• ComfyUI CoTracker Node

SAM2 (Segment Anything 2)

• Official GitHub
• ComfyUI Implementation

Additional Resources:

• KJ Nodes
• Video Helper Suite

📰 Read the full article: LEGO Animation DeepDive: WAN + ATI + CoTracker + SAM2 + VACE Complete Workflow

---

🎬 Master art direction in AI video: normals, bokeh, camera control & trajectories | AInVFX June 6

Learn to art direct Wan 2.1 - Join Adrien for an in-depth ComfyUI tutorial covering four game-changing research papers that enable unprecedented art direction in video diffusion models.

📁 ComfyUI Workflows & Assets: episodes/20250606

🔗 Resources:

NormalCrafter: Learning Temporally Consistent Normals from Video Diffusion Priors

• Project Page • GitHub • Paper
• ComfyUI Wrapper
• ComfyUI Workflow • Input Video

Any-to-Bokeh: One-Step Video Bokeh via Multi-Plane Image Guided Diffusion

• Project Page • GitHub • Paper

Uni3C: Unifying Precisely 3D-Enhanced Camera and Human Motion Controls for Video Generation

• Project Page • GitHub • Paper
• ComfyUI Wrapper • Model • CausVid LoRA (optional)
• ComfyUI Workflow • Input Image • 3D Cube

ATI: Any Trajectory Instruction for Controllable Video Generation

• Project Page • GitHub • Paper
• Model
• ComfyUI Workflow (Start) • ComfyUI Workflow (Final) • Input Image

📰 Read the full article: Art direct Wan 2.1 ComfyUI - ATI, Uni3C, NormalCrafter & Any2Bokeh

---

🎬 ILM turns 50, Cannes surprises, and AI gets 10x faster | AInVFX News May 28

Join Adrien as we celebrate 50 years of Industrial Light & Magic, explore Jafar Panahi's inspiring Palme d'Or win at Cannes, and dive into the latest AI developments transforming the VFX industry.

🔗 Resources:

ILM 50th Anniversary

• ILM's Audacious Start • Creating the Impossible
• The Dykstraflex • John Dykstra Profile
• Rob Bredow TED Talk • Original 70s Footage

Cannes 2025

• Quentin Tarantino Interview
• Palme d'Or Announcement • Film Clip

Industry & Research

• Cinesite TechX: Portal • Company
• SpatialScore: Towards Unified Evaluation for Multimodal Spatial Understanding Project • Paper • GitHub
• Jenga: Training-Free Efficient Video Generation via Dynamic Token Carving Project • Paper • GitHub
• agenticSeek: Private, Local Manus Alternative GitHub

📰 Read the full article: ILM's 50th, Cannes, TechX, SpatialScore, Jenga & AgenticSeek

---

📢 Stay Connected

---