November Highlights
Two landmark model families land simultaneously, NVIDIA ships optimised FP8 quantisations at launch, and the licensing landscape shifts in favour of commercial users.
Flux 2: Next-Generation Image Model
Black Forest Labs open-sources FLUX.2, delivering up to 4-megapixel photorealistic output with dramatically improved lighting, skin, fabric, and hand detail. ComfyUI provides day-zero support with dedicated Flux 2 Pro API nodes, optimised text encoder quantisation, and hardware-aware detection that routes FP8 operations to the most efficient execution path.
NVIDIA collaborates with BFL and ComfyUI to ship FP8 quantisations and RTX GPU optimisations at launch, decreasing VRAM requirements by 40% and improving inference performance by 40% on RTX 40-series hardware. Flux 2 reference-image memory estimation is refined to reduce OOM errors on 12 GB cards.
Market signal: The "FP8 at launch" pattern — model creators, hardware vendors, and UI platforms coordinating on day-zero optimisation — is now the expected standard for frontier models.
HunyuanVideo 1.5
Tencent releases HunyuanVideo 1.5, a substantial upgrade to their video-generation framework. ComfyUI adds native support with new HunyuanVideo workflow variants, improved display names, and dedicated nodes for the 1.5 architecture. Community plugins offer both simplified and complete node sets for quick usage or deep workflow customisation.
New features include LoRA training code for customised video effects (hair growth, object transformation, style transfer), image-to-video pipelines, and support for the LightX2V acceleration framework that brings inference speed closer to real-time on consumer hardware.
Significance: With LTXV (October) and HunyuanVideo 1.5 (November), ComfyUI now supports three distinct video-generation architectures natively — the platform is rapidly becoming the de facto orchestration layer for multi-modal AI.
FLUX.2 VAE: Apache 2.0 Licensing
BFL releases the FLUX.2 VAE under an Apache 2.0 license — fully open source, with no copyleft restrictions. This is a direct response to the enterprise adoption barrier highlighted in our July analysis: the 90% GPL/AGPL licensing problem that blocked commercial deployment.
For the first time, enterprises can integrate a frontier-quality VAE into self-hosted production workflows without licensing friction. BFL signals that future models (including Flux.2 Klein) will follow the same permissive licensing path, potentially reshaping the entire commercial layer of the ecosystem.
Impact: The licensing chasm narrows. Apache 2.0 VAE + permissive PaaS wrappers = viable enterprise pipelines without legal review bottlenecks.
Infrastructure & Developer Updates
CUDA 12.6 official release workflow and portable download support ship, simplifying GPU setup for new users. HunYuan 3D 2.0 compatibility is fixed for improved 3D model generation. New Topaz API nodes bring professional video upscaling directly into ComfyUI workflows. Nano Banana Pro integration and expanded Gemini model options round out the API node ecosystem.
- API security hardened with Content Security Policy headers
- Public API access issues fixed; Gemini model support expanded
get_frame_countandget_frame_ratemethods added for video workflow control- BlockInfo support for Chroma, Qwen-Image, and HunyuanVideo models
- Frontend updated to v1.30.6; Transformers version bumped
Key Takeaway
- Flux 2 + NVIDIA FP8 at launch sets a new coordination standard between model creators, hardware vendors, and UI platforms.
- The Apache 2.0 FLUX.2 VAE is the most commercially significant release of the quarter — it directly addresses the enterprise licensing barrier.
- HunyuanVideo 1.5 cements ComfyUI's position as a multi-modal orchestration platform, not merely an image generation UI.
- Three video architectures now native (LTXV, HunyuanVideo, Topaz upscale) — video workflows are no longer experimental.
Model Ecosystem Growth
| Model / Release | Type | License | ComfyUI Status |
|---|---|---|---|
| FLUX.2 (dev/pro) | Image Generation | Mixed (dev: non-commercial; pro: API) | Day-zero support, FP8 optimised |
| FLUX.2 VAE | Variational Autoencoder | Apache 2.0 | Fully integrated, enterprise-ready |
| HunyuanVideo 1.5 | Video Generation | Tencent Open Source | Native nodes + community plugin |
| HunYuan 3D 2.0 | 3D Model Generation | Open Source | Compatibility fixes shipped |
| Topaz Video AI | Video Upscaling (API) | Commercial API | New native API nodes |