What's new in AI film & video tools

A directory of AI tools and releases for filmmakers and video creators, each linked to its primary source.

124entries8categoriesUpdated 2026-11-02

New here? Start with these 4

  1. Script to Storyboard: Plan Your AI Short FilmAI film pipeline · Vixel
  2. AI Shot List: How to Build a Camera-Ready PlanPre-production · Leah Retta (Runway)
  3. AI Camera Prompts: Movement, Angles and Shots in AI VideoShot prompting · Leah Retta (Runway)
  4. Reference Sheets Are a Cheat Code for AI VideoConsistent characters · Jerrod Lew (@jerrod_lew)
Filter: categories and tools

Categories

Tools

Show all 93 tools
All tools with counts >

Newest entries

  • H3 Continuity Studio gives MiniMax H3 a visual memory, so a room keeps its details when you cut away and come back

    H3 Continuity Studio is a free, open-source ComfyUI node pack for building multi-shot sequences with MiniMax H3 on a timeline. Continuing clips carry context from the clip before, and a visual memory brings chosen moments of an earlier clip into a later shot, even after an exterior scene or a new camera angle.

    H3 Continuity Studio4 sources · github.com

  • H3MD Director renders 2 to 3 minute MiniMax H3 videos in one ComfyUI run and re-renders only the segments that went wrong

    H3MD Director is a free, open-source ComfyUI node pack that plans long, multi-shot MiniMax H3 videos in a visual editor and renders them segment by segment, joined seamlessly, at up to 1080p. Every finished segment is kept on disk, so a bad one can be re-rendered on its own, locked to its neighbours.

    H3MD Director2 sources · github.com

  • Production Desk runs a one-person film production next to ComfyUI, from first key frame to a Nuke comp with EXR passes

    Production Desk is a free, open-source browser panel and set of scripts by Nuke compositor Erman Zambak that runs a short-film production next to ComfyUI the way a VFX house would. It makes key frames and LTX-2.5 shots, queues and reviews them, builds the DaVinci Resolve edit and hands the cut to Nuke as EXR plate, matte, depth and normal passes.

    Production Desk2 sources · github.com

  • Use case: "Wanderer", a 17-shot short film made locally with Pika H3 Flex, YuE2 and Qwen Image 2.1 on one RTX 4090

    ReAiLity Labs (@CreaitiveLabs) made "Wanderer", about 96 seconds of cinematic footage in 17 shots, entirely inside ComfyUI on a single RTX 4090, with 50 minutes of video render time. Their write-up shares the benchmark of Pika H3 Flex's 6 steps against 30-step MiniMax H3, the settings for each stage and every prompt.

    Pika H3 Flex2 sources · x.com

  • ComfyUI-AnyToBokeh adds a scripted rack focus and real-looking bokeh to a finished video clip

    ComfyUI-AnyToBokeh is a new, unofficial node pack that brings the Any-to-Bokeh research model (ICLR 2026) into ComfyUI as a refocusing pass on video you already have. You script where the focus sits frame by frame, or lock it to a masked subject, and the model paints depth-aware, steady background blur without regenerating the shot.

    Any-to-Bokeh4 sources · github.com

  • H3 Audio Retake Studio changes a line of dialogue in a finished clip and repairs the lip-sync, locally with MiniMax H3

    H3 Audio Retake Studio is a free, open-source browser studio for ComfyUI that fixes a wrong line in a video without regenerating the shot. You select the words on a waveform, make a new take of the line, and MiniMax H3 with a Pika H3 Flex LoRA repaints the speaker's mouth to match.

    H3 Audio Retake Studio3 sources · github.com

  • ComfyUI adds a MiniMax H3 Context Windows node for longer clips with matching sound

    A change merged into ComfyUI on 30 October adds context window support for MiniMax H3 and a new MiniMax H3 Context Windows node. The sampler then works on a long clip in overlapping slices of picture and sound instead of all frames at once, which keeps memory tied to the window size.

    MiniMax H32 sources · github.com

  • ComfyUI-Dials turns two prompts into a 14 KB motion or mood slider for MiniMax H3 and Pika H3 Flex

    ComfyUI-Dials is a free, open-source ComfyUI node pack that makes a strength slider for MiniMax H3 from pairs of prompts that differ in one quality, such as standing still versus running. Each dial is a file of about 14 KB, needs no training or reference images, and also works on Pika H3 Flex.

    ComfyUI-Dials4 sources · github.com

  • H3 Motion: drag a subject out of a photo along a path, or mask part of a video and change what happens, with MiniMax H3

    H3 Motion is a free, open-source app that lets you lift a subject out of a photo and drag it along a path, then has MiniMax H3 animate the shot with sound. It can also regenerate a masked part of an existing video, and it runs on your own ComfyUI with open-weights models only.

    H3 Motion4 sources · github.com

  • OneShot MV turns a song and one line about its world into a lip-synced music video, locally in ComfyUI

    OneShot MV is a free, open-source ComfyUI pack and Studio app that takes a song and one line describing its world, invents a cast, plans every shot around the song's energy and renders a lip-synced music video. It runs locally on open-weights models: Krea 2 for the portraits and LTX-2.3 for the singing clips.

    OneShot MV3 sources · github.com

  • Open Video Relight relights real video in ComfyUI from a 3D light stage, with Qwen-Image 2.1 and MiniMax H3

    Open Video Relight is a free, open-source ComfyUI workflow that changes the lighting of real footage while keeping the people and their performance. You place and move lights on a sphere around the subject, Qwen-Image 2.1 relights a few keyframes, and MiniMax H3 carries that light through the clip.

    Open Video Relight3 sources · github.com

  • Paint a plate: paint on the first frame to erase, add or relight, and the edit carries through the whole shot in ComfyUI

    Paint a plate is a free, open-source ComfyUI node pack from AInVFX: paint on the first frame of a shot to erase an object, add something or relight it, and the edit carries through every frame. It exports in the plate's own format, so an EXR comes back as an EXR with untouched pixels identical to the original.

    Paint a plate4 sources · github.com

  • Reshoot 4D gives an existing video a new camera move in ComfyUI, locally on open weights

    Reshoot 4D is a free ComfyUI node pack that turns a video of a person into a 4D capture, lets you keyframe a new camera path over it, and renders the same moment from that new camera. It runs locally on open weights, with LTX 2.3 rendering the final frames.

    Reshoot 4D3 sources · github.com

  • Tessera upscales video to 2K, 4K or 8K in ComfyUI with MiniMax H3 on one consumer GPU

    Tessera is a free, open-source ComfyUI node pack that upscales video with the open MiniMax H3 model, refining it tile by tile and chunk by chunk into 2K, 4K or 8K. Its streamed encode and decode mean the full-resolution clip never sits in memory, so it runs on a single consumer GPU at any clip length.

    Tessera2 sources · github.com

  • ZenTrainer 0.4 trains image and video LoRAs inside ComfyUI, with live previews, on NVIDIA or AMD

    ZenTrainer is a free, open-source app that trains LoRAs inside ComfyUI, using the models and GPU it already has, and shows samples and loss while it learns. Version 0.4.0, released on 30 October, trains ten models end to end, including MOSS Image, Krea 2, FLUX 3, Qwen Image 2.1, SDXL and MiniMax H3 video.

    ZenTrainer4 sources · github.com

  • Audio8-TTS-Realistic: a small open-weights voice cloning model in 23 languages

    Audio8-TTS-Realistic is a free, open-weights text-to-speech model from Edge0 with 0.6 billion parameters that clones a voice from one short reference recording. It speaks 23 languages, including Polish, and the Apache 2.0 licence allows commercial use.

    Audio8-TTS4 sources · huggingface.co

  • Overtone makes word-timed captions cut separately for 9:16 and 16:9 video, free and offline

    Overtone is a free, open-source command-line tool that turns subtitles, audio or video into word-timed captions and cuts them separately for each screen shape, such as 16:9 and 9:16. It can run fully offline with whisper.cpp and Ollama, or use paid speech APIs for higher accuracy.

    Overtone3 sources · github.com

  • Pika H3 Flex: open-weights six-step video with sound, built on MiniMax H3, runs in ComfyUI

    Pika H3 Flex is Pika's post-trained version of the open MiniMax H3 model, released with open weights on day one. It makes video with synchronized stereo audio in six sampling steps, takes up to 12 image, video and audio references, and runs locally in ComfyUI and diffusers.

    Pika H3 Flex5 sources · huggingface.co

  • OpenField is a free, local Higgsfield-style studio for video, stills and 3D on ComfyUI

    OpenField is a free, open-source studio that runs on your own NVIDIA GPU and puts screens for video, stills, 3D objects, effects, upscaling and talking shots on top of ComfyUI workflows. It was built for Comfy's Open Source Higgsfield challenge and uses only open-weights models such as LTX-2.5 and Qwen-Image, with no credits or paid account.

    OpenField3 sources · github.com

  • Perfect Looper makes seamless looping videos with sound from MiniMax H3 in ComfyUI

    Perfect Looper is a free ComfyUI workflow by CoolBreeze_ that uses MiniMax H3 to make a video, with its audio, loop back onto itself without a visible cut. It works from a prompt, an image or reference pictures, and it can also turn an existing 24 fps clip into a loop.

    MiniMax H33 sources · civitai.com

  • previz2shot turns a Blender grey playblast into a finished shot with LTX-2.5

    previz2shot is a free, open-source pipeline that takes a grey playblast rendered from Blender, paints a styled first frame with FLUX.2 Klein and renders the shot with the LTX-2.5 Layout-to-Render IC-LoRA. The result goes straight back into Blender's video editor, and it runs on Comfy Cloud or your own ComfyUI.

    LTX-2.5 Layout to Render3 sources · github.com

  • Seam renders AI video transitions between two shots on your own GPU with LTX-2.5

    Seam is a free, open-source app that takes two shots (images, short clips or 3D models) and renders a generated transition between them, such as smoke, water bend or a whip pan, with the open LTX-2.5 model in ComfyUI. It comes as a web studio, a command-line tool and a plugin that lets an AI agent plan and refine transitions.

    Seam3 sources · github.com

  • Comfy Pro Alpha Generator: click one person in a shot and get a native 4K alpha matte with hair-level edges in ComfyUI

    Comfy Pro Alpha Generator is a free, open-source ComfyUI node by VFX compositor Rikin Patel that pulls a matte of one chosen person from a video. SAM 3 picks and tracks the subject and LTX-2.5's alpha-gen IC-LoRA draws the edge, so single hair strands survive, at native 4K and with 12 GB of VRAM.

    Comfy Pro Alpha Generator3 sources · github.com

  • ComfyArt is a free, local Lovart-style design canvas with an AI agent, running on ComfyUI

    ComfyArt is a free, open-source infinite design canvas that installs into ComfyUI and generates images, video, music, voiceovers and 3D models on your own GPU. A design agent such as Claude Code, Codex or a local LLM can work on the canvas through MCP, with no account or credits.

    ComfyArt3 sources · github.com

  • ComfyUI v0.43 adds MOSS-Image layout control and edit LoRAs, ACEScg EXR output and Kling 4.0 nodes

    ComfyUI v0.43.0, released on 27 October, is the first stable build with native support for MOSS-Image color and depth control and for MOSS-Image edit LoRAs, which until now needed a nightly build. It also adds ACEScg colour space to EXR saving, new Pad Image and Branch nodes, and partner nodes for Kling 4.0, PixVerse V6-Max, FLUX 3 Max and Ideogram Vector.

    ComfyUI3 sources · github.com

  • Henge is an open-source studio for motion transfer and character swaps on MiniMax H3

    Henge is a free, open-source video studio that takes one guide video and a few pictures and gives the performance to a new character, swaps a character or only a head, or sharpens the footage. It runs MiniMax H3 through ComfyUI on your own GPU, on a rented RunPod machine, or from your PC with RunPod Serverless doing the rendering.

    Henge4 sources · github.com

  • Krea 2 BBOX puts comic panels, characters and exact text where you draw the boxes

    Krea 2 BBOX is a free community fine-tune of the open Krea 2 image model that follows a layout you write as boxes: panels, people, objects and word-for-word text each get their own position. It is built for multi-panel comics and storyboard pages, with speech bubbles that stay in their panel and characters that stay the same across panels.

    Krea 24 sources · huggingface.co

  • Pruna's free 6-step LoRAs make Qwen-Image 2.1 generate and edit images in 6 steps instead of 40

    PrunaAI released v0.2 of P-Q-Image-2.1, two free LoRA adapters that let the open Qwen-Image 2.1 model make or edit an image in 6 steps instead of about 40, with no CFG. The same day it published an official ComfyUI pack with ready workflows, so no custom nodes or community conversions are needed.

    Qwen-Image 2.14 sources · huggingface.co

  • Use case: a five-shot interior design film from still images with local MiniMax H3

    The Astravisualize team turned AI concept images of one bathroom into a 24-second vertical film by animating five separate views with MiniMax H3 running locally in ComfyUI, then cutting them in FFmpeg. Their write-up shares the exact camera prompts, the local settings and the takes they threw away.

    MiniMax H32 sources · astravisualize.com

  • OpenFill brings Photoshop-style generative fill to GIMP and PhotoCraft, running locally through ComfyUI

    OpenFill is a free, open-source plug-in that adds generative fill, object removal, generative expand and text-to-image to the GIMP 3 and PhotoCraft image editors. It runs everything on your own graphics card through a hidden ComfyUI backend, so you select an area, type a prompt and get the result on a new layer.

    OpenFill3 sources · github.com

  • Use case: a 3-minute 1920s jazz music video made fully locally on one RTX 4070

    Creator @Jeuner made a 3-minute, 23-shot music video for Bibi Beyond, an invented 1920s jazz singer, entirely on one 12 GB RTX 4070. MiniMax H3 animated each shot from a keyframe and the song, including the lip-sync, and Claude Code ran the direction, editing and pipeline.

    MiniMax H31 source · x.com

  • Zip Ties tidies messy ComfyUI workflows by bundling wires into ribbons

    Zip Ties is a free ComfyUI extension that gathers the wires crossing a line you draw into one ribbon, held by small zip ties you can move, rotate and recolour. It makes big video and image workflows easier to read without hiding any connections.

    ComfyUI3 sources · github.com

  • MOSS-Image Edit HQ, a second community edit LoRA, reaches v6 with workflows and a demo

    MOSS-Image Edit HQ is a free, community-made LoRA that teaches the open MOSS-Image-SFT model to edit a photo from a text instruction, with a focus on keeping everything you did not ask to change. Its author publishes every training round, including the ones that lost, plus two ComfyUI workflows and a demo you can try in the browser.

    MOSS-Image3 sources · huggingface.co

  • WanGP v18.81 lets you build a world from text or an image and walk through it with the keyboard

    WanGP v18.81 adds a Worlds gallery where you create a world from a text description with Hy-Nomas or from an image with EVOKE, then explore it with the keyboard, save your progress and record the walk as a video. The developer says 8 GB of VRAM is enough to start, thanks to the new MMGP v4 memory manager.

    WanGP2 sources · github.com

  • Ollama can now turn text into speech on your Mac, starting with Gradium's Phonon

    Ollama, the popular app for running AI models on your own computer, now does text-to-speech, starting with Gradium's small open Phonon models. It runs on Apple Silicon Macs today, with Windows and Linux support announced as coming soon.

    Ollama3 sources · ollama.com

  • FreeField: open-source Higgsfield-style video effects as ComfyUI workflows

    FreeField is an open-source library of one-photo video effects, such as Earth Zoom, Product Freeze and Comic Bonk, where each preset is a ComfyUI workflow plus a small preset file running on open-weight models. It comes with a gallery app and a command line, and most presets run on an 8 GB NVIDIA graphics card.

    FreeField3 sources · github.com

  • Licon VBVR LoRA helps MiniMax H3 keep objects consistent and actions in order

    Licon Studio's free VBVR LoRA for MiniMax H3 image-to-video targets three common failures: objects that suddenly duplicate, bodies clipping through props, and actions that happen in the wrong order. It is a 149 MB file that loads in ComfyUI's stock LoRA loader with any H3 first-frame workflow.

    MiniMax H32 sources · huggingface.co

  • Seedance 2.5 4K brings native 4K and 10-bit color for film work, in enterprise trials

    ByteDance's Volcano Engine announced Seedance 2.5 4K on 23 October, a film and TV tier of its Seedance 2.5 video model with native 4K output, 10-bit color and extra training on facial expressions, subtle movement and fine detail. For now only selected enterprise customers can try it, and no price has been published.

    Seedance3 sources · technode.com

  • TARA node keeps MiniMax H3's motion reference native and lines it up in time with the output

    TARA is a new, experimental ComfyUI node in the free ComfyUI-BFSNodes pack that changes how MiniMax H3 copies motion from a driving video. Instead of injecting the clip as guide latents, it feeds it as a normal video reference and nudges each output moment to look at the matching moment of the guide.

    MiniMax H33 sources · github.com

  • Tencent Hy-Nomas: an open interactive video world model, now with ComfyUI nodes

    Hy-Nomas is Tencent's open-source video world model: it builds a scene one video segment at a time, and you steer each new segment with text and camera moves while it remembers what came before. Community nodes published the same day run it natively in ComfyUI on 12 GB graphics cards, starting from your own image.

    Hy-Nomas6 sources · huggingface.co

  • Canva Craft: Canva's first in-house image model plans a layout, then renders it at 2K

    Canva launched Craft, its own image model that first plans a picture and then renders it, at 2K in more than 30 aspect ratios with text in 55 languages. It is in Canva's Dream Lab and public API, handy for posters, title cards and key art.

    Canva Craft1 source · canva.com

  • ComfyUI v0.42.2–v0.42.4: FLUX 3 Max, Fibo 2.0, HeyGen Voice, VEED Fabric and Mirelo SFX 2.0 nodes

    Three ComfyUI releases from October 20 to 22 add partner nodes for voice cloning, talking photos, lip-sync on real footage, sound effects and video upscaling. On the open side they add MOSS-Image color and depth control and native Fibo 2.0 nodes.

    ComfyUI3 sources · docs.comfy.org

  • Higgsfield Content Factory turns a "What if" idea into a finished faceless video

    Content Factory is a new Higgsfield tool that takes one "What if…" question and writes the script, picks characters, generates scenes with voiceover and edits a finished faceless video. Videos run from 15 seconds to 5 minutes, with one to four episodes per run.

    Higgsfield Content Factory2 sources · higgsfield.ai

  • Ideogram Vector 1.0 turns a prompt or an image into an editable SVG

    Ideogram Vector 1.0 is Ideogram's first vector model: describe a logo, title lettering or flat illustration and get back an SVG, or upload an image and convert it to vectors. Every shape can be recoloured, moved and resized on its own, which suits title cards, posters and merch for a film.

    Ideogram Vector 1.03 sources · ideogram.ai

  • MiniMax H3 PTD LoRAs cut reference-to-video generation to 8 or 4 steps

    Taobao Content AI released free acceleration LoRAs that make MiniMax H3's Ref2VA mode, which turns reference images into video with sound, run in 8 or 4 steps. A third, 8-step version trained with a method called RefAsymOPD claims sharper, more believable backgrounds at the same speed.

    MiniMax H34 sources · huggingface.co

  • MOSS-Image Edit LoRA adds text-instruction image editing to MOSS-Image, with native ComfyUI support

    MOSS Image Edit is a free community LoRA that lets the open MOSS-Image model edit an existing picture from a text instruction, changing colors, outfits, lighting, seasons or style while up to four reference images keep the rest. ComfyUI merged native support for it the same day, so it runs with stock loaders and samplers on your own GPU.

    MOSS-Image5 sources · huggingface.co

  • OpenKatana is a free, open-source video editor that your AI agent drives with code

    OpenKatana is a new open-source video editor where an AI coding agent plans the shots and builds the edit, with cuts, music, motion graphics and 3D scenes, while the project stays editable on your own computer. It works from a command line, through MCP, as ComfyUI apps and nodes, or in a local Studio app where you review and fix the result by hand.

    OpenKatana3 sources · github.com

  • Pruna P-Video-2-Pro-Avatar: talking, lip-synced videos up to 3 minutes from one portrait

    Pruna launched P-Video-2-Pro-Avatar, a MiniMax H3-based model that turns one portrait and a script or audio track into a lip-synced talking video up to 3 minutes long. It costs from $0.02 per second at 480p and $0.04 per second at 768p.

    P-Video-2-Pro-Avatar2 sources · pruna.ai

  • FastH3-Omni on local hardware: 62 seconds on an RTX 5090, 17 minutes on a 3090

    Hao AI Lab timed FastH3-Omni on desktop machines: a 5-second 480p clip from one image took 61.9 seconds on an RTX 5090, 192 seconds on a DGX Spark, 16.9 minutes on an RTX 3090 and 19.5 minutes on an M5 Max Mac. Community users report it running on an RTX 4090 and, with a GGUF workflow, on an 8 GB card.

    FastH3-Omni3 sources · haoailab.com

  • HeyGen Voice Turbo tops the TTS leaderboard and is free on the API until November 30

    HeyGen made Voice Turbo, the fast mode of its in-house voice model, free to use through its API until November 30, 2026, including voices you clone from one recording. HeyGen says Turbo ranks first on the Artificial Analysis text-to-speech leaderboard, just ahead of its full model.

    HeyGen Voice2 sources · heygen.com

  • MiniMax H3 Singularity V2: a community fine-tune against plastic skin and floating slow motion

    Minimax-h3_Singularity V2 is an open community fine-tune of MiniMax H3 for ComfyUI that aims at H3's shiny skin, blurry distant faces and slow, floating motion. V2, released in INT8, adds wider action coverage, closer prompt following and a fix for audio reference conditioning.

    MiniMax H3 Singularity V24 sources · huggingface.co

  • UniAVC: an open model that edits a video and its soundtrack from one instruction

    Researchers at HKUST and Huawei released UniAVC, a 19B open model built on LTX-2.3 that generates and edits video and audio together, so removing a person also removes their sound. It needs about 39 GB of VRAM even with offloading, which rules out consumer GPUs for now.

    UniAVC3 sources · github.com

  • Use case: how the Nexus series halved its Seedance bill with Blender storyboard frames

    PJ Ace's studio and director Theo Dudley make the Nexus series by blocking out scenes in Blender and feeding them to Seedance as references, and were spending about $2,000 a day on generations. Exporting each scene as 30 storyboard images instead of a 30-second Blender video roughly halved the cost, though they still spend at least $1,000 a day.

    Seedance2 sources · x.com

  • WanGP v18.21 "No GPU Left Behind": auto memory mode, faster RTX 20xx, FastH3 Omni and Fibo 2.0

    WanGP, the low-VRAM app for running open video and image models locally, added an automatic memory profile and big speedups for older RTX 20xx cards. The same update adds FastH3 Omni, Fibo 2.0 with Edit Turbo and the FlashVSR+ Turbo upscaler.

    WanGP1 source · github.com

  • Bria Fibo 2.0: an 8B open-weights model that generates and edits images, with a 4-step Edit Turbo

    Bria released Fibo 2.0, an 8-billion-parameter image model that both generates and edits pictures and was trained only on licensed data. Its Edit Turbo mode edits in 4 steps, and the weights run in ComfyUI and WanGP.

    Fibo 2.03 sources · huggingface.co

  • FastH3-Omni is out with open weights and a ComfyUI template: 8-step reference video with sound

    Hao AI Lab and NuvaLab released the weights and training code for FastH3-Omni, a distilled MiniMax H3 that turns images, video clips and audio references into a video with synced stereo sound in 8 steps instead of 50. ComfyUI supports it on day one with a built-in template and FP8 and NVFP4 files for smaller GPUs.

    FastH3-Omni4 sources · haoailab.com

  • Gradium Phonon: open text-to-speech models small enough to run on a laptop CPU

    Gradium released Phonon, a set of tiny open-weight text-to-speech models, the smallest of which is 31 million parameters and a 64 MB download. Gradium says it starts speaking in 7 milliseconds on a laptop CPU, with no GPU needed.

    Phonon3 sources · gradium.ai

  • Mistral Voxtral 3: directable text-to-speech in 35+ languages and cheaper transcription

    Mistral released Voxtral 3, a family of speech models with a text-to-speech model you can direct for emotion, pacing and emphasis, and new transcription models with speaker labels. The TTS costs $13 per million characters and can adapt a custom voice from 30 minutes of audio.

    Voxtral 31 source · mistral.ai

  • Use case: a Squid Game-style scene from self-hosted FastH3-Omni at about 14 cents a clip

    Creator Mickey Friedman ran FastH3-Omni on one rented B200 GPU and had Opus 5.5 storyboard, review and stitch the clips into a Squid Game-style scene. He worked out about $0.14 per 15-second 768p clip, against about $11.31 for a 15-second 1080p clip on the Seedance API.

    FastH3-Omni1 source · x.com

  • A free MOSS-Image adapter places people and objects from a flat colour map and a depth map

    MOSS-Image Color-Depth Control is a free community adapter that makes the open MOSS-Image model follow a colour map of where each object sits and a depth map of how far away it is. You can block out a scene with 3D mannequins and get a finished picture with the same placement, scale and pose, while the look and background come from your prompt.

    MOSS-Image3 sources · huggingface.co

  • FLUX 3 Max: fal and Black Forest Labs' video model renders a 10-second clip in about 11 seconds

    FLUX 3 Max is a new video model from fal Research and Black Forest Labs that makes 5 to 20 second clips with native sound from text, an image or keyframes. Its makers say a 10-second clip renders in about 11 to 12 seconds, matching Seedance 2.5's score in a preliminary blind test while being far faster.

    FLUX 3 Max3 sources · fal.ai

  • LTX Desktop 1.3.1 adds 4K, an assistant and new IC-LoRA tools for local LTX video

    LTX Desktop, Lightricks' open-source app for making video with LTX models on your own GPU or Mac, reached version 1.3.1 with 1440p and 4K output and an assistant that plans and queues jobs from plain words. It also adds four IC-LoRA tools, including AlphaGen, which cuts a subject out of a video without a green screen.

    LTX Desktop2 sources · github.com

  • Use case: a real dance restaged in a palace by MiniMax H3 from a depth video, on one RTX 5090

    A creator turned an 8.7-second dance clip into a grey depth video and had MiniMax H3 restage the routine with a new character in a palace, natively at 768×1376 on one RTX 5090. The write-up gives the full ComfyUI wiring, the settings and the mistakes that turned the palace into a white studio.

    MiniMax H32 sources · ai-muninn.com

  • ComfyUI v0.42 adds native MOSS-Image support and a Krea 3 node

    ComfyUI v0.42.0 adds native core support for the open MOSS-Image text-to-image model and a Krea 3 Image partner node. A v0.42.1 fix followed a few hours later for crashes some users hit with the MOSS-Image template workflow.

    ComfyUI4 sources · github.com

  • EdenLive turns a rough graybox scene into real-time video at about 25 fps on one RTX 5090

    VAST released EdenLive, an open model that takes a simple grey 3D proxy or a recorded video plus a first frame and a prompt and renders it as realistic video live. The 5B version runs at about 24.7 frames per second at 1280×704 on a single RTX 5090.

    EdenLive2 sources · github.com

  • H3-LongTakeNoCuts: ComfyUI nodes for a three-minute MiniMax H3 shot without cuts

    H3-LongTakeNoCuts is a free set of ComfyUI nodes and a workflow that fight the quality loss when you extend a MiniMax H3 shot with motion context. Its 18 October update adds character masking to the refine pass, and the author shows a three-minute one-take made with it.

    H3-LongTakeNoCuts2 sources · github.com

  • ReAngle re-shoots a video from a new camera angle, locally with LTX-2.5

    ReAngle is a free, open-source app that takes a clip and gives you the same performance back from a new camera angle, such as an orbit, crane, push-in or freeze-time move. It runs on ComfyUI with LTX-2.5 on a 16 GB NVIDIA card, and you can draw your own camera path in a 3D preview before anything renders.

    ReAngle3 sources · github.com

  • 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.

    MiniMax H32 sources · x.com

  • ComfyUI-SceneCut finds every cut in a video and exports the scenes as clips

    ComfyUI-SceneCut is a free node pack that detects every shot in a video, labels hard cuts, jump cuts and transitions, and double-checks each cut so a flash or whip pan doesn't split a shot. A viewer inside the node lets you fix cuts that are a frame off, pick scenes and export them as frame-exact clips with their sound.

    ComfyUI-SceneCut3 sources · github.com

  • 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.

    Kling 4.03 sources · digitaltoday.co.kr

  • 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.

    mflux3 sources · github.com

  • 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.

    drama3-h3-talking3 sources · github.com

  • 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.

    MOSS-Image7 sources · github.com

  • 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.

    Pika4 sources · pika.art

  • 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.

    Plenio Music Production System3 sources · github.com

  • 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.

    MiniMax H32 sources · ai-muninn.com

  • 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.

    WanGP2 sources · x.com

  • 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.

    FLUX 3 Image Klein4 sources · bfl.ai

  • 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.

    ComfyUI3 sources · github.com

  • 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.

    ElevenLabs Instant Soundtracks3 sources · elevenlabs.io

  • 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.

    Krea 33 sources · krea.ai

  • 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.

    Mago3 sources · mago.studio

  • 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.

    Bulbul V4 Flash1 source · sarvam.ai

  • 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.

    SolarRenderer3 sources · github.com

  • Topview open-sources 3D Builder, an agent-built 3D stage for planning AI video shots

    Topview 3D Builder is a free, open-source tool where your AI agent builds an editable 3D scene from a description or a reference image, so you can set blocking, props and camera moves before generating any video. Everything up to the video step runs on your own computer with no account, and you can take the renders to Topview or any video model.

    Topview 3D Builder3 sources · github.com

  • 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.

    BACH 1.0 Pro2 sources · runtimewire.com

  • ComfyUI-MangaScreen turns an illustration into a printed black-and-white manga page

    ComfyUI-MangaScreen is a free custom node pack that converts any image into a clean black-and-white manga or comic panel with crisp lines, flat tones and real halftone screentone. It keeps the line art on its own layer, so lines stay sharp instead of turning to mush like a plain halftone filter.

    ComfyUI-MangaScreen3 sources · github.com

  • 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.

    FastH3-Omni1 source · nuvalab.ai

  • 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.

    Gnani Timbre v3.01 source · gnani.ai

  • H3 Anti-Burn LoRA keeps faces from burning in chained MiniMax H3 clips

    H3 Anti-Burn LoRA is an open LoRA for MiniMax H3 that you load only on continuation clips, so a long chained video keeps the look of its first clip instead of getting crunchier faces with every segment. On a turbo model it was never trained on, its author measured face burn after ten clips drop from +61% to +4%.

    H3 Anti-Burn LoRA3 sources · huggingface.co

  • 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.

    Mirelo SFX 2.03 sources · mirelo.ai

  • 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.

    ComfyUI2 sources · motionengineeringtoday.com

  • 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.

    Concat3 sources · github.com

  • 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.

    Fish Audio Drama 33 sources · fish.audio

  • 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.

    HeyGen Video1 source · civitai.com

  • 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.

    NVIDIA VFX Nodes1 source · blog.comfy.org

  • 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.

    WanGP1 source · github.com

  • 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.

    Vidu Q41 source · news.aibase.com

  • 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.

    Kandinsky 6.0 Video1 source · arxiv.org

  • 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.

    WanGP2 sources · creativeainews.com

  • 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.

    Sync Lipsync v31 source · help.artlist.io

  • 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.

    Whistle2 sources · cactuscompute.com

  • 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.

    LTX-2.5 Layout to Render3 sources · huggingface.co

  • 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.

    MAI-Voice-2.11 source · microsoft.ai

  • 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.

    Eleven v41 source · elevenlabs.io

  • 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.

    MiniMax H32 sources · chrisbeach.co.uk

  • 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.

    Gemini 3.8 Flash TTS2 sources · blog.google

  • 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.

    Google Vids (Gemini Omni 1.1)1 source · blog.google

  • 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.

    ComfyUI3 sources · comfyui-wiki.com

  • 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.

    Boreal1 source · creatify.ai

  • 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.

    Nvidia AI for Media1 source · digitaltoday.co.kr

  • 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.

    Nari Qwen3-TTS2 sources · narilabs.com

  • 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.

    Flick1 source · x.com

  • 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.

    Beeble SwitchX 2.01 source · provideocoalition.com

  • 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.

    ChatGPT Images 2.51 source · openai.com

  • 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.

    Adobe Premiere Generative Media Tool1 source · blog.adobe.com

  • 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.

    Runway Plugins1 source · runway.com

  • 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.

    MAI-Image-2.61 source · microsoft.ai

  • 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.

    LipDub 2.01 source · lipdub.ai

  • 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.

    Flick1 source · x.com

  • 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.

    H3 Max1 source · prnewswire.com

  • 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.

    OpenShot 4.02 sources · openshot.org

  • 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.

    h3-metal2 sources · github.com

  • 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.

    MiniMax H39 sources · blog.comfy.org