SpaceXAI released the voice-to-voice model Grok Voice 3.0, ranking first on the Artificial Analysis Speech-to-Speech Index with a score of 86.9, higher than GPT-Live-1 (Astra, medium) at 83.2 and Gemini 3.8 Live Extended Thinking at 82.6.
Apple's closing market cap on October 14 stood above $5 trillion for the first time, making it the second company globally after Nvidia to reach that milestone. Seres has sued a Douyin user, alleging the user used AI to synthesize a false image of an AITO vehicle's body collapsing, and is seeking 1 million yuan in damages. Doubao mobile assistant announced it is gradually opening screen automation operation permissions for third-party apps, while apps from Alibaba, Tencent, and others have explicitly stated they refuse.
VIBE is an AI video generation app that supports generating explainer videos from text or images using models such as Kling, Sora, and Veo. Its process is: write a 60- to 120-second script, break it into 4- to 10-second shots, generate each shot, and add voiceover and export.
The author developed the ComfyUI-Drama3 custom node for Fish Audio Drama 3, a 35B text-to-speech model that supports voice cloning, dialogue, and inline prompts like [sighs] and [whispers], with a minimum requirement of 8GB VRAM.
Why it matters: The author integrated a 35B voice model into ComfyUI with quantization, making it runnable on 8GB VRAM, which can be used to evaluate local voiceover workflows.
HeyGen releases a new speech model, HeyGen Voice, which it calls the first speech model built for video adaptation and which ranks first on the Artificial Analysis speech leaderboard; it is already included in every HeyGen video at no extra charge.
Why it matters: HeyGen launches its first speech model for video adaptation, allowing creators to assess the room for replacing dubbing and multilingual lip-sync workflows.
An unofficial int4 quantized version, drama-3-int4, has appeared for Fish Audio's 35B (3B active) multi-speaker TTS model Drama 3, with a download size of 22.1 GB, lower than the original 41.5 GB. The expert layers, attention, and deltanet layers are quantized to int4, while the shared expert MLP remains in bf16.
Apple researchers have proposed the first tri-modal masked diffusion model (MDM) pretrained from scratch, handling text, image-text, and audio-text data simultaneously, and systematically analyzed multimodal scaling laws, modality mixing ratios, noise schedules, and batch size effects.
Google DeepMind's Gemini 4 Argon has launched globally, with priority access for Pro and Ultra subscribers. The model uses a sparse Mixture-of-Experts architecture, supports native multimodal input of text, images, and audio, has a context length of up to 1 million tokens, and is priced at $2 per 1 million input tokens and $10 per 1 million output tokens.
On October 13, Google expanded the availability of Gemini 4 Argon; the model supports a 1,048,576-token context and can accept text, images, video, audio, and PDFs, and is available to Pro and Ultra users of the Gemini app as well as paid API and AI Studio users.
Higgsfield, Runway, HeyGen, Invideo AI, Synthesia, and Kling AI are listed as six alternatives to Adobe Firefly in 2026, covering different workflows such as scene generation, editing, AI presenters, script-to-video, and cinematic camera movement.
To address the problems of unclear narration, abrupt sound effects, and overly loud BGM in AI videos, you can use the "three-layer sound" organizing method: first set the lead voice (narration or AI music), let BGM and ambient sound serve as support, and use sound effects only at transitions.
ComfyUI-Drama3 packages Fish Audio's 35B mixture-of-experts speech model Drama 3 with Qwen3-ASR into a ComfyUI custom node with no extra dependencies to install; expert weights are kept in system memory and streamed through a VRAM cache, so it runs on 8GB of VRAM.
Why it matters: Squeezing a 35B speech model onto an 8GB VRAM card, with four quantization versions and frame-by-frame benchmark data, so you can judge whether local output is feasible.
HeyGen's enterprise edition this month adds member usage limits and request features; administrators can set each member's usage quota, and members can directly request more quota within HeyGen with one-click approval. Video Agent supports team collaboration in the same session, with multiple people able to send prompts simultaneously while only one person edits at a time. Brand Glossary has been rebuilt, supporting per-language pronunciation and translation rules and CSV/Excel import and export; ElevenLabs is integrated as a single engine with a new v4 model option.
HeyGen released its October enterprise API update: the HeyGen Video 1.0 API is live, supporting short video generation from prompts, images, or reference materials. ElevenLabs v4 can be used for speech synthesis with per-request model selection, and Professional Voice Clone trains studio-grade voice clones based on HeyGen's proprietary speech model, supporting full-file or streaming output.
Fish Audio releases the open-source speech generation model Drama 3 pre-release version, which can convert text into performed speech. The model uses a 35B-A3B hybrid linear attention MoE backbone (initialized from Qwen3.5) and a new codebook-MTP audio decoder; pretraining covers about 80 languages and dialects and hundreds of billions of tokens, and it supports multi-reference voice cloning, natural language instructions, multi-round revisions, multi-character scenes, speech infill editing, and long-audio and streaming synthesis.
Why it matters: Open-source speech models put multi-character scenes, multi-round script revisions, and voice editing into the same model, allowing dubbing workflows to assess replacement potential accordingly.
UK House of Lords member Baroness Campbell of Surbiton used ElevenLabs' voice cloning and text-to-speech technology to deliver a speech in her own AI-reconstructed voice during the second reading of the Health Bill, the first time a member of either house of the UK Parliament has spoken using AI-generated voice.
An IT engineer with a psychology background used Grok's video generation feature and AI tools like Google Flow to recreate Japanese nursing care scenes, finding that the visuals are already quite natural but sound has become the biggest challenge. Even after specifying Japanese dialogue, the model may still output English, and in loud scenes like "I want to go home!" the AI voice may hit the right volume but struggles to convey emotions like anxiety and confusion, sounding exaggerated or mechanical. The author believes that the more realistic the visuals, the greater the responsibility for sound expression, and that in the future, video and audio may need to be generated separately.
pyvideotrans released v4.16, fixing the Edge-TTS 403 issue and adding two new dubbing channels: audio.cpp and fish.audio API. This version also adds the Atlas Cloud translation channel, a French WebUI interface, and includes "use LLM to re-segment sentences instead of LLM error correction" in the advanced options for subtitle translation.
GenMagic released its first version on ComfyUI, including five modules: GenMagic Image (text-to-image, image-to-image), GenMagic Video (text-to-video, image-to-video), GenMagic Speech, GenMagic Music, and GenMagic Text. This version is provided as ComfyUI nodes, with the source page noting coverage of 450+ image, video, audio, and text models.
Dingjiao One analyzes how AI editing is changing video production payment models, with money originally paid to editors and software companies now flowing to models, computing power, and automation services. In September 2026, Jianying launched the AI Creation Workbench Hub, Smart Editing Assistant, and AI Ultra membership. Adobe expanded Creative Agent to Premiere, while Descript, Runway, and other companies are extending from their respective segments into complete production workflows.
On October 12, Moxi Intelligence announced the completion of a Pre-A round exceeding 1 billion yuan, led by Huakong Fund, with participation from China Development Bank Science and Technology Innovation, Shanghai Science and Technology Innovation Fund, and New Micro Capital. Existing shareholders Yuanhe Holdings, Shanghai Science and Technology Innovation Group, and Xinglian Capital continued to follow on. This is approximately half a year since the angel round announcement.
Descript released its October update, adding a search bar and a full results page to the entire cloud drive, and supporting Underlord in finding files, folders, projects, and layout packs. Enterprise cloud drives can send asynchronous recording requests, letting guests record on their own schedule and read scripts with a teleprompter; AI Speech has been upgraded to ElevenLabs v4, making voice clones closer to the real person.
AITuber released a Lumen5 alternatives guide recommending tools by reason for leaving: choose AITuber for AI-generated visuals, Synthesia if you need an on-camera digital human, Pictory to keep the stock-library style at a lower price, and Kapwing and InVideo AI if you need team seats.
Why it matters: The 35B open-source speech model supports same-track generation of 16 characters and natural-language performance instructions; its pronunciation robustness data can be used to assess dubbing and audiobook workflows.
Jeff T. Thomas won the $2.5 million Future Vision XPRIZE with a three-minute AI trailer, The Gifted, unanimously selected by judges from about 2,500 entries. The prize is $100,000 in cash plus $2.5 million in equity, for a roughly $10 million narrative feature that will not be generated by AI.
A creator used ChatGPT to generate about 25 storyboard images, free voiceover from Irodori-TTS, a rented GPU on Runpod to run miniMax H3 for video generation, and capcut for editing to make a one-minute AI animation, at a cost of about ¥90–¥150 and taking two days.
A creator shares real experience making online course videos with AI: dubbing with a voice clone of their own voice, but after recording 4 versions with emotional ups and downs, no difference could be heard, so they ultimately kept the original voice and instead created a lively feel through writing techniques such as ending with questions.
DramaClip releases v0.2.1, adding a visual track that uses local Qwen3-VL models in three tiers (2B/4B/8B, GGUF quantized + llama.cpp, runnable on CPU) to uniformly sample 16 frames per episode and generate timestamped visual evidence for script prompts to reference by episode.
Hongguo Short Drama's DAU reached 168 million, a year-over-year increase of 107%. AI short dramas have become the most watched and interacted content on its platform. Traditional video platforms continue to lose users. The combined daily active users of Tencent Video, iQIYI, Youku, and Mango TV are only 155 million, with MTS declining year-over-year by 31%, 27%, 47%, and 24% respectively. The article argues that AI short dramas are the most fitting result of the "inertia" logic, and the convergence of short video consumption habits and AI technology drives their industrialized production.
Synthesia starts at $19/month for Starter and ranks first overall on slide import fidelity, AI narration quality and editability, supporting 140+ languages.
AITuber releases a guide to choosing AI content repurposing tools, matching tools to source material: for articles or PDFs, choose AITuber to generate narrated videos with visuals and subtitles; for long videos, choose Opus Clip; for podcasts, choose Castmagic; for distributing finished clips, choose Repurpose.io.
Stable Diffusion Art publishes a tutorial on how to build a MiniMax H3 text-to-video workflow in ComfyUI, generating a 5-second 24fps clip with stereo sound in one pass.
The research team proposes SyncReward, a reward model for evaluating and optimizing synchronization in audio-video generation, and releases the SyncReward-Data-80K dataset with 80K professional synchronization ratings and the SyncReward-Bench benchmark of 1K clips. The model achieves the highest agreement with human ratings in evaluations and can improve generation synchronization through inference-time best-of-N selection and reinforcement learning post-training. Data, benchmark, model and code will be open-sourced.
AI filmmaker Frank Houbre disclosed his 2026 creative tool stack, divided into six stages according to the filmmaking process, with only one main tool kept for each key stage. The writing stage centers on ScreenWeaver, which he co-created, with Claude and Claude Code handling review and repetitive tasks; image and video generation center on Higgsfield as the main tool because it aggregates multiple models under the same account.
OpenBMB released VoxCPM2, which removes the discrete token quantization step of mainstream TTS and instead uses a diffusion autoregressive architecture to generate speech in a continuous audio latent space; the repository has 38.7k Stars, an Apache-2.0 license, and fully open weights and code for commercial use.
Why it matters: VoxCPM2 removes discrete token quantization and instead generates speech in a continuous latent space, and offers multiple deployment paths from local to production.
Sound designer and re-recording mixer Charles Maynes responds to Rene Coronado's article, arguing that AI cannot replicate human metaphorical creative thinking, but the real threat is not at the creative level. He points out that cost-driven clients, AI tools that are rented rather than owned, copyright uncertainty, and erosion at the bottom of the market are the risks the sound industry needs to face.
By drawing frames with Python, stitching with ffmpeg, and using the latest speech model only for narration, the cost of a 3-minute learning video can be reduced to about 50–90 yen, far below the 570–1,700 yen of video generation AI. The author recommends Gemini 3.8 TTS or Eleven v4 for narration, and says erroneous sentences can be re-read individually for easy correction.
Kapwing published a step-by-step guide on using AI to generate Vox-style collage animation documentaries, broken into six steps: writing the script, breaking down visual beats, generating collage images, frame-by-frame animation, generating voiceover, and editing the final cut.
Why it matters: A complete breakdown of the six-step AI workflow for Vox-style collage documentaries, including reusable image and animation prompts, ideal for creators looking to replicate the visual style.
Author Akita Ryosuke published his four-step workflow for using Claude desktop with Claude in Chrome to operate Seedance 2.5 for video generation: first, use ChatGPT to create a character sheet and prepare props and sound materials; then write out clear details for characters, scenes, actions, camera, dialogue pronunciation, negative prompts, aspect ratio, quality, and duration; next, let Claude automatically operate the Yapper platform to generate; and finally, provide feedback for revisions on each item.
Why it matters: The author published a complete four-step workflow for using Claude to operate Seedance 2.5 for video output, including character sheets, lip-sync details, and cost breakdowns — transferable to your own generation workflow.