AInVFX

I submitted a small personal contribution to the Arca Gidan contest — a 2-minute short film made entirely with open source models, built around 73 of my own ink drawings.

The full training process is shared freely: the Z-Image style LoRA, the LTX-V 2.3 video LoRA, and every step in between. All open, all documented.

But honestly — go look at the other submissions first. There's a lot of great work in there, and voting is open until April 6th. Watch a few, leave a score, get inspired.

🎬 Teaser: https://youtube.com/shorts/hq1jg6s91y0
🎬 Full selection: https://arcagidan.com/submissions/entry/5ca70873-e0c6-481a-96ef-5e15809451be

73 of my own hand drawings, 2 LoRAs. One short film — made entirely with open source models. This teaser covers the full pipeline: training a Z-Image style LoRA for image generation, then an LTX-V 2.3 LoRA to bring it to life. The entire process — da...

To celebrate the holiday season, we're releasing a new deep dive: a complete guide to ComfyUI from manual installation to full workflows.

ComfyUI's node-based interface intimidates many people before they even start. The scope of possibilities makes it hard to know where to begin. This guide takes a different approach: instead of overwhelming you with every feature, we walk through one complete workflow from start to finish. You'll see how installation, image generation, art direction, video creation, post-processing, and HD upscaling all connect in context.

The goal isn't to teach you everything—it's to give you the foundations to explore confidently.

🤿 Watch the tutorial: https://youtu.be/VG0hix4DLM0
📰 Read the article: https://www.ainvfx.com/blog/demystifying-comfyui-complete-installation-to-production-workflow-guide/

Thank you all for your support this year. Wishing you a wonderful holiday season, and we'll see you in 2026!

🤿 Demystifying ComfyUI: Complete installation to production workflow guide | AInVFX Dec 19 🎨 Stop feeling overwhelmed by ComfyUI. This comprehensive guide takes you from manual installation all the way through creating your first complete AI workf...

🎬 Today we're exploring how to create a realistic crowd.

Step 1: Open SideFX Software Houdini
Step 2: Do your magic (it's Houdini after all)
Step 3: Render a flipbook
Step 4: ...
Step 5: Okay fine, you also need a bit of ComfyUI.

Creative upscaling is changing how studios approach previz. That rough blocking you rendered in 10 minutes? It doesn't have to stay that rough.

Our monthly Masterclass covers how to turn any previz into near-final quality — and more importantly, how to art direct it through notes and iterations. 4 hours of foundational ComfyUI knowledge, plus a private 45-minute mentoring session tailored to your goals.

We don't teach you every model — they change daily. We teach you how the ecosystem works, so you can keep learning on your own.

Also works on astronauts. 🧑‍🚀

🎟️ Enroll now: https://www.ainvfx.com/masterclass
https://www.youtube.com/shorts/Jp05MPRnr7M

From Previz to Final: Creative Upscaling in ComfyUI | GenAI Masterclass for Film & TV | AInVFX 🎬 Turn any previz into (almost) final videoThat rough Houdini flipbook? That temp astronaut blocking? They don't have to stay rough. Creative upscaling is t...

🎬 Last Generative AI Masterclass of 2025

Traditional and AI workflows are finally merging, and there's never been a better time to get up to speed. We're hosting our final live session of the year - a 4-hour deep dive into ComfyUI and generative AI for visual storytelling, plus a private 45-minute mentoring session tailored to your goals.
We don't teach you every model - they change daily. We teach you how the technology and ecosystem works, so you can keep learning on your own.
Whether you're a seasoned pro or just getting started, join us before the year ends.

🎟️ Enroll now: https://www.ainvfx.com/masterclass
https://youtube.com/shorts/iF1BNiocgVg

Last Generative AI Masterclass of 2025 | ComfyUI for Film and TV + personalized mentoring | AInVFX 🎬 Don't miss the final Generative AI Masterclass of the yearStudios and filmmakers often ask if they've missed the AI train. Truth is, the train left a long...

🤿 SeedVR2 v2.5 released
After 4 months of R&D with support from the open-source community, we're pleased to release updated inference code for SeedVR2. We're calling it v2.5 and it's a breaking change requiring workflow updates.
Main changes:

Available both as command line interface tool for batch processing/multi-GPU and ComfyUI integration
Modular node-based architecture to better accommodate different users & hardware requirements (now fully supports consumer hardware from as low as 8GB of VRAM)
Native alpha channel processing
GGUF, tiling, torch.compile, Apple Silicon support... and much more
Now available in ComfyUI Manager
Inference code & models released under Apache 2.0/MIT license - commercially permissive

Thanks to the brilliant open-source community for tirelessly testing/breaking/fixing over the last 4 months.
🤿 Watch the tutorial: https://youtu.be/MBtWYXq_r60
📰 Read the article: https://www.ainvfx.com/blog/seedvr2-v2-5-the-complete-redesign-that-makes-7b-models-run-on-8gb-gpus/

🤿 SeedVR2 v2.5 Video Upscaling: Official Guide from the ComfyUI Integration Team | AInVFX Nov 7 🎬 SeedVR2 has evolved! Version 2.5 brings a complete architectural redesign with modular 4-node system, VAE tiling, GGUF quantization support for 8GB GPUs, ...

🤿 New AInVFX Deep Dive: One-step video upscaling is finally here

ByteDance just changed the game with SeedVR2 - transforming low-res footage to 4K and beyond in a single step instead of the usual 15-50. And they released it under Apache 2.0 license.

In today's deep dive, we walk through everything you need to know:
- How Diffusion Adversarial Post-Training achieves this breakthrough
- Running 7B parameter models on consumer hardware using BlockSwap
- Production workflows with image sequences and alpha channel support
- Real-world performance and honest limitations

The game-changer? NumZ's ComfyUI integration makes this accessible to everyone, for free. We demonstrate how to run the 7B model on just 16GB VRAM.

Yes, there are challenges - high memory usage, occasional oversharpening, VAE bottlenecks. But we now have production-ready AI upscaling that's both fast and temporally stable.

🤿 Watch the full tutorial: https://www.youtube.com/watch?v=I0sl45GMqNg
📰 Read the article: https://www.ainvfx.com/blog/one-step-4k-video-upscaling-and-beyond-for-free-in-comfyui-with-seedvr2/

🤿 One-Step Video Upscaling: Complete ComfyUI SeedVR2 Guide (Free workflow included) | AInVFX July 11 🎬 Transform low-resolution videos to 4K and beyond in a single step with SeedVR2, ByteDance's groundbreaking video restoration model now available under Apa...

🎬 New AInVFX News: Speed up your video generation workflows by 2-3x

Today we dive into four practical techniques that solve real production challenges:
MagCache - Skip computation steps intelligently without complex calibration. Get 2-3x speedup on Wan2.1 and other models with minimal quality loss.
NAG - Finally use negative prompts with distilled models like CausVid and Self-Forcing. No more CFG=1 limitations.
DLoRAL - Upscale videos in a single step instead of 30-50. Dual LoRAs handle detail and temporal consistency separately.
AI Art Restoration - MIT shows how AI and polymer films can restore damaged paintings 66x faster than manual methods.

Complete ComfyUI workflows included, with step-by-step parameter guidance.

Watch the full episode: https://www.youtube.com/watch?v=YGTUQw9ff4E
Read the article: https://www.ainvfx.com/blog/speed-up-video-generation-2-3x-magcache-nag-dloral-and-ai-art-restoration/

hashtag hashtag hashtag hashtag hashtag hashtag hashtag hashtag hashtag hashtag hashtag

Speed up WAN 2-3x with MagCache + NAG Negative Prompting + One-Step Upscaling | AInVFX News June 21 🎬 Learn how to accelerate your video generation workflows with four practical techniques. MagCache speeds up Wan2.1 and other diffusion models by 2-3x witho...

🎬 New AInVFX News: Four papers democratizing AI development

ByteDance just proved you don't need thousands of GPUs to build competitive video models. ContentV achieves state-of-the-art results using only 256 NPUs in 4 weeks.

Today's episode explores breakthrough approaches making advanced AI more accessible:

- ContentV - Reuse image models, train progressively, work smarter not harder
- CoTracker3 - Track through occlusions, trained on real videos without manual annotations
- Self-Forcing - Enable real-time video streaming with autoregressive generation
- CBottle - Compress Earth's climate into a few gigabytes

The common thread? Innovation doesn't require infinite resources when you approach problems creatively.

Watch the full episode: https://www.youtube.com/watch?v=U6LoN10ZpxU
Read the article: https://www.ainvfx.com/blog/contentv-cotracker3-self-forcing-and-cbottle-democratizing-ai-development/

🎬 AInVFX News - June 18, 2025 | ContentV, CoTracker3, Self-Forcing & CBottle Join Adrien as we explore four groundbreaking papers democratizing AI development. From ByteDance's efficient video training to Meta's robust tracking, Adobe...

🤿 New AInVFX DeepDive: Transform a single LEGO photo into a complete animated shot!

In this comprehensive tutorial, we combine 5 powerful open-source tools:
- WAN 2.1 as the base model
- ATI for trajectory-based animation
- CoTracker for precise motion tracking
- SAM2 for automatic segmentation
- VACE for stylized 2D fire & smoke effects

The workflow covers everything from basic setup to advanced optimization, with complete ComfyUI workflows and all assets freely available.

Watch the full tutorial: https://youtu.be/7YmiJxPEMk0
Read the article: https://www.ainvfx.com/blog/lego-animation-deepdive-wan-ati-cotracker-sam2-vace-complete-workflow/

h

🤿 AInVFX DeepDive - Create your own LEGO animated shot from scratch: WAN+ATI+CoTracker+SAM2+VACE 🧱 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 f...

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.

Watch the full episode → https://youtu.be/0cw2N3W7nKo
Read the full article here → https://www.ainvfx.com/blog/art-direct-wan2-1-normalcrafter-any-to-bokeh-uni3c-and-ati-deep-dive/

🎬 AInVFX News - June 6th, 2025 | Art direct Wan 2.1 ComfyUI - ATI, Uni3C, NormalCrafter & Any2Bokeh Take control of every aspect of AI video generation! Join Adrien for an in-depth ComfyUI tutorial covering four game-changing research papers that enable unp...