Skip to content
Date and source

Dates follow the news timeline in Beijing time and include both endpoints.

TodayNov 19Thu1 item
  1. Google DeepMind68

    Genie 4 is now available.

    Genie 4 is now available to Google AI Ultra subscribers, supporting real-time world editing during sessions and real-time native audio. Genie 3 has been updated in sync, with single generations up to 3 minutes long, supporting 360-degree panorama seeds and real-time world events; new Project Genie agent and community gallery have been added.

    Why it matters: Genie 4 is available to Google AI Ultra subscribers, supporting real-time world editing during sessions and real-time native audio, allowing creators to evaluate its interactive world generation, sharing, and re-creation capabilities.

Nov 18Wed
  1. 可灵官方博客65

    Kling releases Kling Image 4.0, focusing on realism and native 8K generation.

    Kling has released its next-generation image generation model, Kling Image 4.0, now available on Kling Creative Studio and open to enterprises and developers via API. The model is based on a native unified understanding and generation architecture, supporting native 8K image generation, text rendering of up to 2000 characters per image, generation of up to 10 consistent-style images from a single prompt, and long-image generation such as e-commerce detail pages. It also offers 110+ preset color grading styles, 70+ preset commercial fonts, and support for seven languages.

    Why it matters: The new model unifies visual understanding and image generation in the same autoregressive framework, and its native 8K and long-text rendering capabilities are worth creators evaluating as a replacement for existing image generation workflows.

Nov 16Mon
  1. Zyphra · Zonos 语音模型66

    Zyphra releases ZONOS3-Mini multilingual TTS model

    Zyphra releases compact multilingual TTS model ZONOS3-Mini, supporting 43 languages, with approximately 100 million active parameters per token, and provides a native transformers port. The model uses ZAC-1 codec as audio tokenizer, includes 27 built-in release voices, supports voice cloning from short audio, and is licensed under Apache-2.0.

    Why it matters: ZONOS3-Mini is released as a native transformers port, supporting 43 languages and voice cloning, facilitating evaluation of its suitability as a multilingual TTS baseline.

Nov 13Fri
  1. 美团 LongCat · 创作模型70

    Meituan LongCat releases LongCat-Video-Avatar 1.6, an audio-driven human video generation model.

    Meituan LongCat releases the open-source audio-driven human video generation framework LongCat-Video-Avatar 1.6, which improves lip-sync, coordinated facial and body movements, and long-video consistency compared to 1.5, supports video generation over 10 minutes, and optimizes inference speed.

    Why it matters: The new version improves long-video stability and inference efficiency, and in human preference evaluation it ranks second only to OmniHuman-2.0, providing a reference for selecting digital human video generation solutions.

Nov 10Tue
  1. MiniMax · 音视频模型58

    MiniMax releases MiniMax-H3-Image-PE prompt expansion model

    MiniMax releases MiniMax-H3-Image-PE, fine-tuned from Qwen3.8-27B, providing prompt expansion and aspect ratio recommendations for MiniMax-H3-Image. It supports two workflows: text-to-image and image editing. The image editing PE is still under development; only its prompt is released this time. The model runs with FP4 weights and can be deployed via SGLang or Diffusers, requiring approximately 19 GiB of GPU memory.

  2. Zyphra · Zonos 语音模型65

    Zyphra releases open-source speech model ZONOS3-Mini

    Zyphra releases open-source compact speech model ZONOS3-Mini, with approximately 100M active parameters, capable of real-time operation on phones, laptops, and browsers. The model supports cloning voices from audio samples, designing characters by age/accent/timbre, and word-by-word control of emotion, speaking rate, and expression. Weights are open-sourced under Apache 2.0, and it can be served via SGLang-Omni on NVIDIA GPUs with an OpenAI-compatible interface.

    Why it matters: With 100M active parameters, it can synthesize speech in real time in the browser, providing a locally runnable open-source solution for localized dubbing and character voices, supporting voice cloning and word-by-word emotion control.

Nov 9Mon
  1. 字节 Seed 官方博客63

    Seedream 5.5 released: supports 100 rounds of stable editing and native RGBA output.

    ByteDance released the next-generation image generation model Seedream 5.5, which supports 100 rounds of multi-round editing, native RGBA transparent background output, and improved image realism and text rendering. Compared with Seedream 5.0 Pro, the usability rate across six production scenarios has improved, with film/TV and office scenarios exceeding twice the previous version; the severe text corruption rate dropped to 0.4%.

    Why it matters: Seedream 5.5 focuses on 100 rounds of stable editing and native RGBA output. Evaluations show the severe text corruption rate dropped to 0.4%, providing more reliable image assets for professional visual production.

Nov 6Fri
  1. Google AI Blog82

    Gemini Omni 1.2 Flash: Faster, more cost-effective, supports 15-second clips.

    Google releases Gemini Omni 1.2 Flash, which natively supports 15-second video generation; compared to 1.1, the price drops 30% to $0.07/second (720p). The new model supports multi-turn conversational editing, enhanced reference following, and can be upscaled to 1080p or 4K after 360p/720p drafts. Gemini Omni is now generally available in Gemini API and Gemini for business.

    Why it matters: The new model increases native video length to 15 seconds and cuts prices by 30%, making it suitable for evaluating batch video production and ad creative production costs.

Nov 4Wed
Nov 3Tue
  1. MiniMax · 音视频模型66

    MiniMax releases H3-Image, an open-source image generation and editing model

    MiniMax releases open-source image generation and editing model MiniMax-H3-Image, unifying text-to-image and image editing in a single checkpoint. The model is a 33.1B-parameter DiT, distilled to 6-step sampling, directly generating approximately 16-megapixel images (4096×4096 to 6272×2688) without super-resolution.

    Why it matters: The new model unifies text generation and image editing in a single checkpoint, directly outputs 16-megapixel images with only 6-step sampling, allowing evaluation of its suitability for existing image generation workflows.

Oct 30Fri
  1. ElevenLabs 产品与研究60

    ElevenLabs launches SFX v3 sound effects model, now available on ElevenCreative and ElevenAPI

    ElevenLabs launches SFX v3 sound effects model, now available on ElevenCreative and ElevenAPI. SFX v3 handles prompts describing multiple sounds more reliably, produces more believable music and vocal generation, and delivers cleaner, more natural audio quality than SFX v2. Creators can use it in Flows, the Sound Effects tab, Studio, and Music Editor, while developers can integrate it via ElevenAPI.

    Why it matters: SFX v3 improves multi-sound prompt following, music and vocal generation, and audio quality, allowing creators to evaluate whether to replace their existing sound effects generation workflow.

Oct 29Thu
  1. 可灵官方博客71

    Ami discusses how Kling 4.0 brings subtle emotions to life.

    Kling invited creator Ami to use Kling 4.0 to produce the short film "Tea When You Have Time". She shared her experience using new capabilities such as 30-second generation, Omni Reference, and multi-keyframe. Kling 4.0 supports single-run 30-second generation, 21:9 aspect ratio, up to 10 keyframes, consistent 12-character assets, good lip-sync for Cantonese dialogue, and supports multiple dialects and English accents.

    Why it matters: Creator hands-on testing shows Kling 4.0 performs stably in 30-second long takes, 12-character consistency, and Cantonese lip-sync, providing a basis for evaluating an upgrade.

Oct 26Mon
  1. 可灵官方博客67

    Kling Releases Kling 4.0, Supporting 3–30 Second Generation and Omni Reference

    Kling officially released Kling 4.0 on October 15, 2026, supporting single generations of 3–30 seconds, more stable large-scale motion and complex camera movements, and up to 10 keyframe inputs. Omni Reference can combine up to 15 reference assets in a single generation, including 10 images, 5 videos totaling no more than 30 seconds, and 7 elements, and prompts can be used to make targeted modifications to existing assets' subject, background, expression, camera position, and style.

    Why it matters: The official post reviews Kling 4.0's reference asset, keyframe, and targeted editing capabilities, and offers workflow suggestions from storyboard to delivery.

Oct 21Wed
  1. FastVideo 视频模型 · 模型发布80

    FastVideo releases ComfyUI version model weights for FastH3 Omni

    FastVideo released ComfyUI-usable weights for the final version of FastH3 Omni, providing two diffusion models, BF16 and INT8 ConvRot, which can generate video and synchronized audio from prompts and ordered image, video, and audio references.

    Why it matters: FastH3 Omni releases ComfyUI-usable BF16 and INT8 weights and provides the way to connect reference images, video, and audio, making it easier for creators to evaluate local video production pipelines.

  2. Krea 官方博客60

    Krea 3 appears on three image-editing leaderboards: a breakdown of each ranking and interval

    Krea 3 Ultra ranked 4th in Arena multi-image editing, 8th in single-image editing, 5th in Artificial Analysis AA-Image-Editing v2.1, and 4th in Design Arena (out of 51 models) on three image-editing leaderboards published on October 19-20.

    Why it matters: Krea officially breaks down the votes, confidence intervals, and ranking intervals of the three leaderboards item by item, explaining why the same model appears in four positions.

Oct 20Tue
Oct 17Sat
  1. FastVideo 视频模型 · 模型发布60

    FastVideo releases FastH3-QAD-v2 video-audio model

    FastVideo releases FastH3-QAD-v2, an 8-step video and audio model trained with NVFP4 weights and quantization-aware distillation (QAD), runnable on a single Blackwell GPU (RTX 5090, RTX PRO 6000) or DGX Spark.

    Why it matters: The 8-step video and audio model lands on a single Blackwell GPU with NVFP4 quantized weights, and error data versus the BF16 teacher model is provided.

Oct 16Fri
  1. Krea 官方博客71

    Krea releases the image editing model Krea 3, at $0.06 per 2K image and most edits completed within 8 seconds

    Krea releases the image editing model Krea 3; the company says most edits complete end-to-end within 8 seconds at native 2K, at $0.06 per 2K image and $0.03 per 1K image, with up to 6 reference images per edit.

    Why it matters: The official comparison provides per-image processing time and pricing for Krea 3 versus Nano Banana Pro and GPT Image 2, which can be used to estimate batch image editing costs.

Oct 15Thu
  1. Black Forest Labs 官方博客82

    Black Forest Labs releases FLUX 3 Image Klein open-weight model

    Black Forest Labs releases FLUX 3 Image Klein, a 5.6B-parameter open-weight model that supports text-to-image, image editing, and multi-reference generation with up to 10 reference images, and features native 4K generation, pixel-level editing, and bounding-box control.

    Why it matters: An Open-Source Image Model with 5.6B Parameters Brings 4K Generation and Pixel-Level Editing Down to 8GB VRAM, Allowing Evaluation of Local Image Generation and Editing Solutions.

  2. Black Forest Labs · FLUX 模型62

    Black Forest Labs open-sources FLUX 3 Image Klein Prompt Upsampler

    Black Forest Labs releases FLUX 3 Image Klein Prompt Upsampler, a 9B vision-language model that expands brief image requests into detailed prompts better suited to FLUX 3 Image Klein.

    Why it matters: The official release open-sources a 9B prompt expansion model that can expand brief requests into complete prompts including lighting, composition, and subject details, and supports reference images and bounding boxes.

  3. 可灵官方博客62

    Kling breaks down the 5M+ view short film "The Beat," detailing five capabilities of Kling 4.0

    Kling's official blog breaks down the production process of its short film "The Beat," which has surpassed 5 million views on social platforms. The article explains Kling 4.0's capabilities item by item: native 30-second continuous shot generation, Omni Reference supporting up to 15 multimodal references in a single task, Multi-Keyframe supporting up to 10 keyframes, 10-bit HDR output, and upgraded stereo audio with lip-sync.

    Why it matters: The official blog breaks down the production process of a short film with over 5 million views, explaining Kling 4.0's 30-second native generation, Omni Reference, and 10-bit HDR usage in photorealistic shots.

  4. HeyGen Updates78

    HeyGen releases the HeyGen Voice speech model

    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.

Oct 14Wed
  1. OpenAI News60

    Higgsfield uses GPT-6 Astra to extend AI video from 1 minute to over 10 minutes

    Higgsfield extends AI video generation length from 1 minute to over 10 minutes, powered by GPT-6 Astra. At the same time, it uses GPT-Image-2.5 to improve image precision and GPT-Live to provide a path toward voice-driven filmmaking.

    Why it matters: Higgsfield uses GPT-6 Astra to push AI video length from 1 minute to over 10 minutes, which can be used to assess its usability for feature-length creation.

  2. Fish Speech 官方发布72

    Fish Audio releases the open-source speech generation model Drama 3 pre-release version

    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.

Oct 13Tue
  1. Black Forest Labs · FLUX 模型78

    Black Forest Labs releases FLUX 3 Image Klein open-weight model

    Black Forest Labs Releases FLUX 3 Image Klein, a 5.6B Open-Weight Model That Simultaneously Supports Text-to-Image, Single-Reference Image Editing, and Multi-Reference Image Combination Generation, with Qwen3-VL-4B-Instruct as the text encoder.

    Why it matters: The 5.6B open-weight model unifies text-to-image, single-image editing, and multi-image combination into a single model, and provides quantized versions with VRAM usage details, making it easier for self-deployers to assess implementation costs.

  2. Black Forest Labs · FLUX 模型76

    Black Forest Labs Open-Sources FLUX 3 Image Klein Base, 5.6B Weights Support Text-to-Image and Multi-Image Editing

    Black Forest Labs has released FLUX 3 Image Klein Base on Hugging Face, an open-weight model with 5.6B parameters. Text-to-image, single-reference image editing, and multi-reference image composition share the same model, with the text encoder being Qwen3-VL-4B-Instruct.

    Why it matters: The 5.6B open weights cover text-to-image, single-image editing, and multi-image composition simultaneously, and provide three tiers of VRAM usage—BF16, FP8, and NVFP4—making it easy to evaluate self-hosting costs.

Oct 12Mon
  1. 腾讯 ARC · 创作模型60

    Tencent ARC Open-Sources Multi-Person Video Generation Model Multics

    Tencent ARC released Multics on Hugging Face, achieving scalable multi-person video generation through layered communication. The repository includes the complete 4-player Multics DiT checkpoint, configuration, Wan2.2 VAE, UMT5 text encoder, and tokenizer; the same checkpoint supports 6-, 8-, and 10-person generation without retraining. The weights are distributed under the Apache 2.0 license.

    Why it matters: Tencent ARC open-sources the multi-person video generation model Multics, where the same weights can scale to 10 people, letting creators evaluate generation options for multi-person scenes.

Oct 11Sun
Oct 10Sat
  1. 可灵官方博客58

    Creator hands-on test of Kling 4.0: character consistency, multi-image transitions, and style switching performance

    Korean visual creator 서오키 (@oasis.seo.ai) was invited by Kling to hands-on test Kling 4.0, using three creative experiments to examine Omni Reference, style transitions, and character consistency. She said Kling 4.0 can coherently handle subtle movements such as eye contact, expressions, and gestures, preserves the original image's color tone and texture when connecting multiple images, and keeps character form and movement stable during repeated switching between live action and animation, reducing repeated generation and post-production fixes. She also hopes Sound Effects, character consistency, and fine control will continue to improve.

Oct 9Fri
  1. 通义千问 · 图像与音频模型62

    Alibaba releases Qwen-Image-2.1-Turbo accelerated image generation model

    Qwen released Qwen-Image-2.1-Turbo, an accelerated checkpoint of Qwen-Image-2.1 for text-to-image and image editing, requiring only 8 denoising steps.

    Why it matters: 8-step sampling and KV cache reuse of text and reference image context let creators assess the speedup potential of their image generation and editing workflows.

  2. Midjourney 官方更新50

    Midjourney tests Thinking Mode for image generation

    Midjourney is testing a new Thinking Mode for image generation on the Alpha website (alpha.midjourney.com), which users can run by clicking "Rerun (Thinking)" in the lower right corner of a job's lightbox. Tests show the mode helps improve prompt accuracy, text layout, and image coherence. Midjourney says it may in the future make it a general thinking mode, or "add more thinking" after a job is completed.

Oct 6Tue
  1. FastVideo 视频模型 · 模型发布62

    FastVideo Releases FastH3 Trim Lightweight Model and ComfyUI Repackaged Files

    FastVideo released the experimental lightweight model FastH3 Trim, reducing transformer blocks from 50 to 42, using rank-16 timestep conditioning, 8-step sampling, with a 4.2x smaller size than the base H3 and faster than FastH3 V2.

    Why it matters: FastH3 Trim offers a lighter video generation solution with 42 layers, 8-step sampling, and 4.2x size compression, along with ComfyUI files and parameters for various precisions.

Oct 3Sat
  1. FastVideo 视频模型 · 模型发布62

    FastVideo releases the FastH3 Trim 8-step audio-video generation model

    FastVideo released the experimental model FastH3 Trim, which removes 8 of the least impactful blocks from FastH3 V2's 50 MiniMax H3 transformer blocks and replaces each block's AdaLN timestep projection with a shared rank-16 basis, trained with 8-step DMD2 (checkpoint 300), enabling 8-step text-to-video generation with synchronized audio.

    Why it matters: FastH3 Trim uses pruning plus shared AdaLN to bring 8-step audio-video generation to consumer-grade hardware, and can be compared against V2 to weigh image quality against size.

  2. IndexTTS · 语音模型60

    IndexTeam Releases Index-Translate-35B-A3B-preview Multilingual Translation Model

    IndexTeam has released Index-Translate-35B-A3B-preview, a multilingual text translation model based on Qwen3.5 with 35B total parameters and 3B active parameters in an MoE architecture. It supports 150 languages and natively supports end-to-end image translation without requiring external OCR.

    Why it matters: The Index-Translate family adds a 35B-A3B preview version supporting 150 languages and direct image translation, allowing evaluation of its low-resource language performance against report data.

Oct 2Fri
  1. 可灵官方博客62

    Kling 4.0 enters closed beta: creators test 30-second long takes and multi-shot consistency

    Kling 4.0 has entered closed beta, with an official release planned for October this year; upgrades include stable dynamic motion, high-quality stereo audio, more accurate lip sync, up to 4K resolution and 10-bit HDR output.

    Why it matters: Kling 4.0 enters closed beta; creators test 30-second long takes, motion stability and multi-shot consistency, which can help judge whether a commercial production pipeline is worth switching to.