NVIDIA released Nemotron 3.5 Transcribe on October 28, 2026: a free, open model with a browser demo that labels up to eight speakers. We ran a real podcast through it and priced an hour of transcript against Descript, Otter and the APIs.
NVIDIA released RTX Remix 1.6 on October 23, 2026: a free beta mod that path traces Minecraft Java Edition, DLSS 5 controls for modders, and ComfyUI texture and mesh generation inside the Toolkit.
Only one of 17 open-source AI 3D model families, TRELLIS.2, is MIT licensed and runs in all four creator apps we checked. We measured adoption, licenses and app support for every family, from Pixal3D and Hunyuan3D to TripoSplat and SHARP.
Bria Fibo 2.0 is the top-scoring open-weights image editor, trained on licensed images. It costs about what Nano Banana 2.1 does, scores lower, and the free download is non-commercial.
Black Forest Labs' FLUX 3 Image Klein is a 5.6B open-weights image model you can use for paid work if your business earns under $5M a year. It runs on 8 GB GPUs in ComfyUI and costs $0.009 per 1K image hosted. We tested its text rendering and editing.
Only one of 12 open-source AI music model families, ACE-Step 1.5, is free to sell from, runs in all five creator apps we checked and fits a 4 GB graphics card. We measured adoption, licenses and app support for every family, from YuE2 to MusicGen.
Comfy Org's free NVIDIA VFX Nodes bring RTX Video Super Resolution (up to 4x), Frame Generation (2x to 16x, RTX 40+) and SDR-to-HDR (400 to 2000 nits) into ComfyUI, alongside ComfyUI v0.40.0. How to run the chain and what it replaces.
OpenScreen is a free, MIT-licensed screen recorder for Windows, macOS and Linux that turns a raw capture into a polished demo: automatic zooms on your clicks, on-device captions and an MP4 or GIF export with no watermark. Here is the full workflow in nine steps.
Fish Audio released Drama 3 on 13 October, a voice model you direct like an actor. We compare it with Eleven v4 on price and control, and show how to direct a scene in the app.
Two anime models, Anima and Illustrious XL, drew 16,934 of the 22,959 LoRAs published on Civitai in the past month, 73.8% of the total. Every FLUX model combined drew 1.4%. We measured 29 open image model families on adoption, license and app support.
Mistral Large 4, a 1-trillion-parameter open-weight model, entered public preview on October 6, 2026 at $0.68 input and $2.09 output per million tokens, half its list price. It costs less than Mistral Medium 3.5 and GLM-5.3. Weights are due at the end of October.
Kandinsky 6.0 Video, released October 6, 2026 under MIT, generates 5-second clips with synced 44 kHz audio and runs on 16 GB GPUs. The catch is system RAM: Pro needs about 58 GiB.
ComfyUI v0.39.0 changed Save Video to CRF 18 for H.264 and 24 for AV1. On three 1080p clips, H.264 files grew 64-83% and AV1 files about doubled, for 0.3 to 1.5 dB more PSNR. EXR now defaults to 16-bit, and Luma Ray 2 nodes end 24 October.
llama.cpp 0.6.0 ships a Hugging Face model downloader and a memory-fit estimate. We tested both: downloads work through the API today, the model browser is still a pull request, and the estimate ignores context and vision files.
WanGP v17, released October 5, 2026, cuts 15 seconds of MiniMax H3 at 1080p from 25 GB to 11 GB of VRAM with its new MMGP v4 memory engine. Two savings are on by default; three settings you switch on yourself.
InvokeAI 7.0 alpha 1 adds video with generated sound, a rewritten canvas with PSD export, autosaving projects and semantic gallery search. What changed, which video model you can legally use, and how to install it without breaking v6.
Meta open-sourced Muse Gadgets on October 2: ESP32 firmware, a Linux SDK and 42 device skills. The Linux SDK lets Muse run any command as the account you choose. We tested the runner, found what the log and the sudo warning miss, and wrote a safe Pi setup.
Lightricks released Alpha Gen on October 2, 2026, an open-weights LTX-2.5 add-on that turns ordinary video into a frame-matched alpha matte with no green screen, masks or prompt. Here is how it compares, what it needs and how to try it.
rembg's default background removal model is now BRIA's RMBG-2.0, which is non-commercial and ran out of memory on our 12 GB test machine. We scored the models that do run on 500 photos: ISNet with ViTMatte edges was the best free setup.
Cloudflare's open Clef decision models would top the Decision Index at 61.2 on Cloudflare's own numbers, but cost 5.7 times Jev per token, Clef-flash fails hallucination checks, and 5,354 downloads went to GGUF files missing the decision head.
DeepFilterNet3 lifted speech quality on every noise we tested, from keyboard clatter to mains hum, on one CPU core. At its default setting it also removes a music bed almost completely, and the RNNoise inside OBS is the pre-2024 model.
DeepSeek Harness Desktop, a free MIT agent app for Mac and Windows, sends a structured session log with every request on DeepSeek's own model route by default. Our capture test shows what it carries, what it costs, and how to switch it off.
Hugging Face's new Open TTS Leaderboard puts the 82M-parameter Kokoro first in English. We read every model's license: in the multilingual and voice cloning views the #1 and #2 are non-commercial, and the first model you can ship is Qwen3-TTS at #3.
MiniMax H3 drew 150 of the 311 community models built on open video models in September, 48% of the total. Its license excludes the US, the EU, the UK and South Korea. We measured 13 families on adoption, license and ComfyUI support.