Skip to content
Date and source

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

Nov 8Sun
  1. Luma News18

    Midjourney's Limitations: What It Cannot Do

    Midjourney has notable limitations in brand consistency, text rendering, enterprise integration, and professional video production. It lacks a public API, generates images publicly by default, and standard access requires a paid subscription. Text rendering may produce inconsistent characters, character consistency is difficult to maintain, and video output lacks cinematic-level controls such as multiple keyframes and professional export formats.

  2. Luma News16

    Synthesia's limitations: what it cannot do

    Synthesia can only generate digital human broadcast videos, cannot output professional post-production formats such as HDR/EXR, and cannot produce cinematic shots. Script changes require regenerating the entire segment, and it cannot make local edits to already approved elements, making it difficult to meet creative needs such as product launches and brand advertising. Luma's Ray 3.2 provides native HDR and 16-bit EXR export, and Layers allows changing the product, title, or background while preserving other elements.

  3. Luma News19

    HeyGen's Limitations: What It Cannot Do

    HeyGen cannot create cinematic scenes, product shots, or environmental content; it only supports preset gestures, with no custom animation or full-body motion control. During peak rendering times, delays can reach up to 60 minutes. Lip-sync accuracy decreases with tonal languages and regional accents, and it does not provide HDR or EXR professional color-grading output. Luma's Ray 3.2 supports 16-bit EXR export, Layers supports element-level editing, and Luma Agents can maintain project context.

Nov 6Fri
  1. PixVerse 中文官方博客69

    PixVerse's hands-on comparison of the best Serverless AI video inference service providers in 2026

    PixVerse releases a comparison of the best Serverless AI video inference service providers in 2026, covering five providers: fal, PixVerse, Baseten, Runpod, and Modal, and uses the same product image to test three models: PixVerse V6, H3 Max, and Wan 2.2.

    Why it matters: PixVerse uses the same product image to test three models, providing a comparison of per-generation cost, speed, and quality, and includes a reusable testing process for service providers.

Nov 5Thu
Nov 4Wed
  1. 可灵官方博客63

    Ash Koosha: Kling 4.0's long takes give emotions time to unfold

    Ash Koosha shares his experience using Kling 4.0, saying that long takes give emotions time to unfold, and targeted editing can fix individual details. He estimates that discarded footage will be reduced by at least 50%, and Kling 4.0 supports single generations of up to 30 seconds. He believes that more stable camera movement and more nuanced expressions are more important for filmmaking.

    Why it matters: Creators tested Kling 4.0's long take and targeted editing capabilities, estimating a 50% reduction in discarded footage, providing a reference for cinematic long take production and detail correction.

Oct 23Fri
  1. Synthesia · 官方博客22

    Synthesia rounds up 50 video consumption trends not to be ignored in 2026

    Synthesia summarized 2026 video consumption trends: internet users worldwide consume 11 hours 39 minutes of online video per week, exceeding traditional television's 10 hours 15 minutes, with short video accounting for 6 hours 42 minutes. Video already accounts for more than 75% of global cellular data traffic, and 92% of internet users stream at least once a month, but only 32% pay for subscriptions.

Oct 22Thu
  1. Pictory 官方博客25

    Agentic AI examples in marketing and video: how Pictory 3 uses AI video agents to generate videos

    Pictory 3's AI video agent can turn requirements such as topic, audience, purpose and duration into a script, scenes and a preview, and generate the final video after the user confirms. In marketing scenarios, HubSpot's Data Agent and Prospecting Agent handle customer research and buying-signal monitoring, Salesforce Agentforce generates marketing briefs and audience segments, and Sprinklr's agent monitors more than 30 channels and sets approval and circuit-breaker mechanisms.

  2. Pictory 官方博客25

    How generative AI and agentic AI differ for video creators: a comparative analysis of Pictory 3

    Pictory's official blog compared how generative AI and agentic AI divide the work in video creation: the former produces individual assets such as scripts, images or clips according to prompts, while the latter plans and executes multiple steps in a goal-oriented way. The article uses Pictory 3's AI video agent as an example; it can complete five settings—voiceover, visual sources, aspect ratio, layout presets and brand kit—and generate a script, scenes and a preview for human review. Most workflows use both together, with humans responsible for review and judgment.

  3. Luma News16

    Gemini Omni Flash vs. Wan 3.0: Which Is Currently the Best AI Video Model?

    Gemini Omni Flash excels at conversational editing and multimodal input. Wan 3.0 can generate clips up to 30 seconds long and supports document input. Both have their own advantages. Luma AI's Ray 3.2 offers multi-keyframe sequences, HDR output, and EXR export. Layers (powered by Uni-1) supports single-element modification, making it suitable for production workflows. In 2025, the AI video generation market size is approximately $700 million to $2 billion, with North America leading.

Oct 21Wed
  1. Krea 官方博客62

    Krea 3 vs Nano Banana Pro: leads on leaderboards, but still behind on 4K and long text

    Krea officially compared Krea 3 with Nano Banana Pro, stating that Krea 3 ranks higher on all three leaderboards—Arena Multi-Image Edit, Design Arena and Artificial Analysis—that most edits return within 8 seconds at native 2K, and that it is 3 times cheaper than GPT Image 2 and Nano Banana Pro.

    Why it matters: Krea officially compared its own Krea 3 with Nano Banana Pro using three public leaderboards and same-prompt tests, and listed the applicable scenarios for each.

Oct 20Tue
  1. Higgsfield 官方博客30

    Comparison of the best all-in-one subscriptions for AI image and video in 2026: Higgsfield, Runway, Krea, Magnific, Pika Art

    Higgsfield AI, starting at $15/month, covers image, video, character consistency and ad production within the same credit system, while Runway, starting at $15/month, offers image and video generation plus a mature post-editing layer including Aleph 2.0 and Edit Studio. Krea aggregates 150+ models from $9/month, Magnific focuses on a 250M+ asset library and design workflows, and Pika Art, from $10/month, focuses on short-video social content.

Oct 2Fri
Sep 8Tue
Feb 3Tue
Jan 27Tue
  1. Hugging Face Blog22

    Architecture Choices in China's Open-Source AI Ecosystem: How MoE, Multimodality, and Domestic Hardware Can Surpass DeepSeek

    Chinese open-source models have nearly unanimously shifted to MoE architecture, with Kimi K2, MiniMax M2, and Qwen3 all adopting this approach, aiming for sustainable deployment at controlled costs. Open-source activity expanded to multimodality and agents starting February 2025, with StepFun's Step-Audio-R1.1 voice model outperforming closed-source models, and Tencent open-sourcing Hunyuan Video and Hunyuan 3D.

Jan 31Fri
  1. Hugging Face Blog22

    AI Art Tools Monthly Report Issue 1: A Review of Key Advances in Image and Video Generation in 2024

    The Hugging Face blog launches the first issue of the AI Art Tools Monthly Report, reviewing key releases in the creative AI space in 2024. In image generation, diffusion models have shifted from U-Net architectures to DiT and flow matching. Flux.1 surpasses Midjourney v6.0 and DALL·E 3 (HD) on multiple benchmarks, while SDXL remains the mainstream foundation for personalized and controllable generation.

Jan 27Mon
  1. Hugging Face Blog60

    The Diffusers Team Surveys the State of Open-Source Video Generation Models and Optimization Approaches

    The Hugging Face Diffusers team publishes an article surveying the state of open-source video generation models and introducing how Diffusers supports these models. The article lists open-source models such as CogVideoX, Mochi-1, HunyuanVideo, Allegro, and LTX-Video, noting that high resource requirements, generalization ability, and latency are the main limitations.

    Why it matters: The Diffusers team systematically surveys the state of open-source video models and provides an optimization combination that reduces HunyuanVideo's VRAM usage to approximately 6.5GB.

Dec 9Mon
Oct 23Wed
Jun 20Thu
Jun 26Mon
  1. Hugging Face Blog22

    Hugging Face Newsletter Issue 4: Bias in text-to-image models

    Hugging Face released Issue 4 of its "Ethical and Social Newsletter," focusing on bias in text-to-image models. The article outlines five sources of bias: training data, pretraining data filtering, CLIP inference, the model's latent space, and post-hoc filtering, and points out that Stable Diffusion v2.1 and ERNIE ViLG produce culturally stereotyped results under the same prompt, "a house in Beijing." The team also introduced the Stable Bias project, a tool for exploring and comparing bias in generated images.

May 8Mon
  1. Hugging Face Blog30

    A deep dive into text-to-video models: development and challenges from GANs to diffusion models

    A Hugging Face blog post traces the development of text-to-video models: early GAN- and VAE-based methods were limited by low resolution and short duration, after which Phenaki, Make-A-Video, NUWA, VideoGPT, CogVideo, and others shifted to transformer architectures, while current mainstream approaches use diffusion models. Text-to-video must ensure spatiotemporal consistency across frames and faces three major challenges: high compute costs, scarce high-quality datasets, and vague video descriptions; models can typically generate only very short clips.

Mar 2Thu
  1. Hugging Face Blog12

    Hugging Face releases ethical guidelines for the Diffusers library

    Hugging Face released an ethical framework for the Diffusers library to guide maintainers in technical decisions about community contributions. The guidelines cover six values: transparency, consistency, simplicity, accessibility, reproducibility, and responsibility, and list safety mechanisms such as community discussion forums, content labels, and bias exploration, as well as deployment measures including Safe Stable Diffusion, staged releases on the Hub, and OpenRAILs licensing.

Dec 15Wed
Jun 17Wed
  1. OpenAI News20

    OpenAI's Image GPT: Generating Images with Transformers

    OpenAI found that a large Transformer model trained on language can also generate coherent image completions and samples when trained instead on sequences of pixels. The research also established a correlation between sample quality and image classification accuracy, showing that its best generative model contains features under an unsupervised setting that are comparable to those of top convolutional networks.

Apr 23Tue
Jul 28Thu