New by day
0 new today, 25 new this week (as of 2026-10-18). Grouped by the day each entry was added. RSS feed.
Use case: a music video made fully locally with MiniMax H3, YuE2 and HiDream O1
A creator made a three-minute music video using only open-weight models on their own machine: YuE2 for the song, HiDream O1 for character and world references, Qwen 3.8 27B to format prompts and MiniMax H3 for the video. The edit was done in DaVinci Resolve with only basic tools and no cloud AI.
Kling 4.0 launches globally
Kling 4.0 is Kling AI's new video model, which can generate up to 30 seconds of video in one go with keyframe control and sound. After early access in late September, it is now fully available, including through MCP and a command-line tool.
mflux 0.23 runs FLUX 3 Image Klein locally on Apple Silicon
The open-source mflux tool now runs Black Forest Labs' FLUX 3 Image Klein natively on Macs with MLX. Version 0.23 covers text-to-image, editing with up to ten reference images, LoRAs and 4, 6 or 8-bit quantization.
Open ComfyUI workflow turns text into a lip-synced talking video with Drama 3 and MiniMax H3
A free ComfyUI workflow from Skysay chains Fish Audio Drama 3 for expressive speech with MiniMax H3 for a lip-synced talking video, all on your own GPU. It ships as MIT-licensed graphs plus a script that runs both stages.
OpenMOSS releases MOSS-Image, an open text-to-image model with layout control
MOSS-Image is a new open-weight text-to-image family from OpenMOSS, with Base and SFT checkpoints under Apache 2.0. Besides plain prompts, it accepts tag lists and JSON layouts that place text and objects with bounding boxes.
Pika's ChatGPT plugin puts its video, image and sound apps inside the chat
Pika launched a plugin for ChatGPT that runs its 40+ creation apps and 55+ image, video, music and speech models inside the chat. You can ask ChatGPT to bring up an editing timeline, and the plugin works with your existing Pika library.
Plenio 1.0 turns ComfyUI into a free local song studio
Plenio is a free, open-source ComfyUI node pack that goes from a song idea to a mastered track on your own machine. A local language model writes the lyrics, YuE2 or MiniMax Music 3 performs them, and version 1.0 adds a browser player and lyric videos.
Use case: a 22-second drone flight over Taiwan from a Blender blockout and MiniMax H3
A creator made a 22-second FPV-style flight past four Taiwanese landmarks without a drone, using a gray Blender blockout and MiniMax H3 running locally on one RTX 5090. The write-up shares every step, the render times and what still breaks.
WanGP v18.17 adds FLUX 3 Klein, Fish Audio Drama 3 and MOSS Image
WanGP v18.17 brings FLUX 3 Image Klein, Fish Audio Drama 3 dialogue with voice cloning and MOSS Image 5B to the low-VRAM local app. It also adds a lighter H3 video decoder, INT8 math kernels and a Deepy Prime agent that plans a one-minute video from a brief.
Black Forest Labs opens FLUX 3 Image Klein
FLUX 3 Image Klein is a 5.6B-parameter open-weights model for image generation, editing and multi-reference work with up to 10 images. It supports native 4K and bounding boxes to control where things go.
ComfyUI v0.41.0 adds FLUX.3 Image and MiniMax H3 LightVAE support
ComfyUI v0.41.0 adds native support for Black Forest Labs' FLUX.3 Image models, including the open FLUX 3 Image Klein, and for MiniMax H3's LightVAE. It also adds ImageSDRToHDR and ReferenceMultiLatent nodes.
ElevenLabs Instant Soundtracks scores a video and places sound effects in one pass
Instant Soundtracks in ElevenCreative takes an uploaded video, writes an original score for it and places sound effects at the moments that call for them. The effects land on their own layers, and the API can return music and effects as separate files.
Krea 3 image editing model
Krea 3 is an image editing model that returns most edits in under 8 seconds at native 2K while keeping products and characters consistent. It costs $0.06 per image and is also offered via partners like fal and Replicate.
Mago Local runs the Mago AI film studio on your own GPU
Mago released a Windows desktop version of its AI video studio that renders on your own NVIDIA GPU, so local renders don't use credits. It can connect to an existing ComfyUI install or set one up, and the cloud models are still available in the app.
Sarvam launches Bulbul V4 Flash for Indian languages
Bulbul V4 Flash is a fast text-to-speech model with 83 voices across 11 Indian languages. It reaches first audio in under 250 ms and costs ₹20 per 10,000 characters.
SolarRenderer turns grey-box, skeleton or sketch video into finished shots with MiniMax H3
SolarRenderer is a pair of open LoRAs for MiniMax H3 that renders a structure video, such as a game-engine grey box, a skeleton or depth pass, or a pencil storyboard, into finished video in four sampling steps. The first frame and the prompt set the look, so the same blockout can come out photoreal, anime, pixel art or ink.
Video Rebirth launches BACH 1.0 Pro
BACH 1.0 Pro is a video model aimed at film, advertising and animation teams that need consistent characters and products across shots. Each generation can use up to 12 image, video or audio references, and 1080p API output is listed at $0.12 per second.
FastH3-Omni preview speeds up MiniMax H3 reference video
FastH3-Omni is an accelerated version of MiniMax H3 that conditions clips on images, audio, text and first and last frames. A hosted preview is live, with open weights and training code announced for October 20, 2026.
Gnani Timbre v3.0 adds audio tags
Timbre v3.0 is an Indian-language text-to-speech model that uses inline tags to make a voice laugh, sigh, whisper or pause. It launches with 13 voices in 10 languages.
Mirelo SFX 2.0 rebuilds the sound of a video and keeps the dialogue
Mirelo SFX 2.0 is a video-to-sound model that removes unwanted music from a clip, keeps the dialogue, and generates effects, foley and ambience synced to the action. It handles edits of up to 10 minutes in one pass and is now also available on Runway.
ComfyUI v0.40.0 adds camera-angle nodes and LoRA stack loaders
ComfyUI v0.40.0 adds CameraAngle and CreateCameraInfo nodes for defining camera parameters inside a workflow, plus native LoRA stack loaders. It also changes MiniMax H3 image decoding and fixes EXR and half-float file handling.
Concat is a free, open-source CapCut alternative that runs its AI locally
Concat is an open-source video editor for macOS, Windows, Linux and Android with auto-captions, text-to-speech with voice cloning and background removal that run on your own machine. It's in beta, has no watermark or account, and scripts or AI agents can drive it through a JSON-RPC API.
Fish Audio releases Drama 3 emotive TTS
Drama 3 is an expressive text-to-speech model in 83 languages that you direct with plain-language cues. It is available in the Fish Audio app and API, and as open weights under a community license.
HeyGen Video model arrives on Civitai
HeyGen's first general-purpose video model makes 5 to 15 second clips with dialogue, ambience and sound effects generated in the same pass. It is now available in the Civitai generator for text, image or reference to video.
NVIDIA VFX nodes bring local upscaling, frame generation and HDR to ComfyUI
NVIDIA's RTX video effects now run inside ComfyUI as custom nodes on your own RTX GPU. They cover video upscaling up to 4×, frame generation from 2× to 16×, and SDR-to-HDR conversion.
WanGP v18 adds Kandinsky 6.0 and much faster video VAEs
WanGP v18.00 rewrites the LTX 2, Wan and HunyuanVideo VAEs to be 2 to 3.5 times faster and lighter on VRAM. It also adds the Kandinsky 6.0 audio-video model, Wan 2.2 Prism Video and Wan 2.1 camera control.
Shengshu launches Vidu Q4 Preview with 4K output
Vidu Q4 Preview makes video clips up to 16 seconds long at up to 4K with 10-bit color. It accepts up to 15 reference images and 3 reference audio clips to keep characters and voices consistent.
Kandinsky 6.0 Video released as open audio-video model
Kandinsky 6.0 Video generates five-second clips with matching sound, including lip-synced speech, from text or an image. The Lite and Pro models, code and weights are released under the MIT license.
WanGP v17 runs 15 s of MiniMax H3 at 1080p in 11 GB of VRAM
WanGP, a free local app for open video, image and audio models, rebuilt its memory engine (MMGP v4) in v17. The developer says 15 seconds of MiniMax H3 at 1080p now needs 11 GB of VRAM instead of 25 GB, and 480p fits in 5-6 GB.
Sync Lipsync v3 available on Artlist
Sync Lipsync v3 re-renders the mouth in existing footage or a still image to match a new audio track, up to 4K. It holds sync on profile angles and when hands or mics block the mouth.
Whistle: on-device speech to text in 16.9 MB
Whistle is an open speech recognition model from Cactus that fits in a single 16.9 MB file and runs on a CPU without dependencies. It transcribes seven languages with word timestamps, which is handy for captions, and audio never leaves the device.
LTX-2.5 Layout to Render turns a 3D blockout into a finished shot
Lightricks released an open IC-LoRA for its LTX-2.5 22B video model that turns a grey clay viewport or blocky playblast from Blender or Unreal into a finished shot. The layout sets the camera, motion and object placement, and an art-directed first frame sets the look.
Microsoft ships MAI-Voice-2.1 and Flash
MAI-Voice-2.1 is a text-to-speech model covering 23 languages where a single voice keeps its identity across languages. The Flash variant targets high-volume use with about 150 ms end-to-end latency.
ElevenLabs launches Eleven v4
Eleven v4 is ElevenLabs' most expressive text-to-speech model, built for audiobooks, voiceovers, dubbing and character dialogue. A Turbo variant offers about 100 ms latency.
Use case: a family sci-fi short made with MiniMax H3 on a home PC
Chris Beach made a short space film starring his family, with every shot generated locally by MiniMax H3 in ComfyUI on one RTX 5090. He published the full workflow, render times and electricity cost.
Google releases Gemini 3.8 Flash TTS with voice design
Gemini 3.8 Flash TTS and Flash-Lite TTS let creators design new voices from a text prompt and direct dialogue line by line in over 100 languages. They are available in Google AI Studio, the Gemini API and Google Vids, and every clip is watermarked with SynthID.
Google Vids offers free HD AI video with Gemini Omni
Anyone with a Google account can now generate 1080p video clips in Google Vids using the Gemini Omni 1.1 Flash model at no cost. Every generated clip carries a digital watermark.
ComfyUI v0.36.0 adds Video Concatenate and native FastH3
ComfyUI v0.36.0 adds a Video Concatenate node that joins up to 100 clips in a graph, and brings the 8-step FastH3 version of MiniMax H3 into the core build. It also adds Marigold V2 depth, normal and albedo workflows and YuE2 music generation.
Creatify launches Boreal ad video model
Boreal is a text-to-video and image-to-video model built for advertising that costs one cent per second. It renders roughly in real time and is available via Creatify's API, its apps and fal.
Nvidia expands AI for Media with lip sync and 8K upscaling
Nvidia's AI for Media toolkit for broadcasters adds frame generation for slow motion, 8K upscaling and SDR-to-HDR conversion. It also adds real-time lip sync to translated audio and speaker detection.
Nari Labs serves fast, low-cost Qwen3-TTS
Nari Labs offers optimized versions of the open Qwen3-TTS and Qwen3-ASR speech models through a simple API. The company says they lead Coval's voice AI benchmark on the balance of quality, latency and cost.
Use case: 'Tusk Hunters' AI short film made with Flick
Filmmaker Alan Xiao released Tusk Hunters, an AI short film made with Flick for the Summer '26 Flick Residency Film Series. It features original music by Daniel Lindholm.
Beeble SwitchX 2.0 video-to-video update
SwitchX 2.0 lets filmmakers and VFX artists swap elements in real footage, now with native 4K, 20-second shots and 10-bit color. It also improves camera tracking and lip sync and adds an up-res finishing workflow.
OpenAI introduces ChatGPT Images 2.5
ChatGPT Images 2.5 brings sharper detail, more precise editing and faster image generation in ChatGPT. Developers get two API models, GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst.
Premiere gets a Generative Media Tool in the timeline
Adobe Premiere can now generate video and sound effects directly in the timeline, using Firefly or partner models such as Veo, Kling, Runway and Luma. After Effects also gets an AI Assistant in public beta.
Runway plugins for Premiere Pro and After Effects
A free Runway panel inside Premiere Pro and After Effects lets editors generate, restyle and upscale footage without leaving the timeline. Generations use the user's existing Runway plan credits.
Microsoft releases MAI-Image-2.6 and Flash
MAI-Image-2.6 is Microsoft's strongest image model so far, now available to developers in Microsoft Foundry. A Flash variant targets fast, high-throughput production work.
LipDub 2.0 speeds up short-form localization
LipDub 2.0 is a new lip sync model with no training step, making it 8x faster than LipDub 1.0. It handles single-speaker videos up to 2 minutes at 1080p in 80+ languages.
Use case: 'Xenogenesis' AI short film made with Flick
Xenogenesis is an AI short film by filmmaker @Gertywood5 made with Flick. It is part of the Summer '26 Flick Residency Film Series.
fal releases H3 Max video model
fal post-trained the open MiniMax H3 weights into H3 Max for text-to-video and image-to-video. fal says it runs faster than real time and costs $0.08 per second at 768p after the launch promotion.
OpenShot 4.0 adds color grading and ML subject isolation
OpenShot 4.0 is a major update to the free open-source video editor, adding color wheels, curves, LUTs and video scopes plus built-in screen and webcam recording. It can also isolate subjects using machine learning models that run locally.
h3-metal: MiniMax H3 running natively on Apple Silicon
h3-metal (antirez/h3.c) is an open-source C and Metal inference engine that runs MiniMax H3 on Mac computers. It supports prompt-to-video with audio, first and last frames, and image, video and audio references.
MiniMax H3 open weights run locally in ComfyUI
MiniMax H3 is MiniMax's first open-weights video model, making clips with stereo sound up to 2K and 15 seconds. ComfyUI supported it on day one with pruned, quantized files that can run on a 12 GB RTX 3060.