Skip to content
Date and source

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

Sep 30Wed
  1. DataEye数据 · 搜狐公开专栏30

    How can content companies hold their place at the AI short-drama table? A conversation with AI product R&D lead Zhong Huai on decoding the logic of industrialized hits

    Facing the second half of AI short dramas shifting from "tool competition" to "ecosystem competition," the content company's AI tool builds systematic engineering capability with "five core functions + one accompaniment service," including one-click delivery of hundred-episode finished films, one-click provisioning of global assets, intelligent overall scheduling, canvas Agent output, and ultra-fast production of hit scripts.

  2. ArchiGen AI · 图像生成与ComfyUI实战20

    Before trusting AI rendering tool rankings, first draw your input path

    ArchiGen points out that three 2026 AI rendering tool rankings from Chaos, Gendo, and Syntina give different winners, and the divergence stems from the ranking tables compressing the variable of how project information enters the tool and how review results return to the team. The article suggests filling out a seven-column "path card" for each tool (source, transfer, control, output, revision, return, loss) and counting the manual cost of the "conversion tax" brought by each boundary crossing. Native model integration does not equal better quality; it solves a different transfer problem.

Sep 29Tue
  1. 公众号:短剧自习室62

    Exclusive Interview with Xuebao: Zombie Apocalypse — 2026 Through the Eyes of an AI Short-Drama Producer

    Xuebao, founder of Xuebao Studio, described the first half of 2026's AI short-drama industry as a zombie apocalypse in an exclusive interview. After the release of Seedance 2.0, the production process was greatly simplified and output surged, but a DataEye report shows that 221,900 new AI short dramas were added on Douyin in the first half of the year, with a hit rate of only 0.47%.

  2. 公众号:短剧深水区58

    Exclusive Interview with DRAMAGO Founder Chang Jing: The Reality and Imagination of Overseas Interactive Dramas

    In an exclusive interview, DRAMAGO founder Chang Jing reviewed the company's April transformation from a traditional short-drama platform to an interactive drama platform. The first work went viral upon launch, with in-app data showing that AI interactive dramas have a three-episode completion rate nearly 2x that of live-action short dramas, and first paywall conversion rates nearly 30x higher than traditional short dramas. She believes interactive dramas are not an independent track but a content format, with the biggest bottleneck being the platform's product capabilities. She revealed that the next step will be creating 'story worlds' based on existing drama universes, with a beta internal test expected to launch at the end of September or early October.

Sep 28Mon
  1. 深响 · 搜狐公开专栏36

    What New Brand Collaboration Possibilities Is AI Comic Drama Bringing?

    According to DataEye, from January to May 2026 the domestic AI short drama market exceeded 22 billion yuan, with users surpassing 600 million in the first half of the year; brands such as Tmall and La Mer have already incorporated AI comic drama into their marketing plans. Brand collaborations on Hongguo Short Drama follow two paths: entering the storyline for content co-creation, or co-creating with virtual characters; La Mer and "Shock!

Sep 27Sun
  1. note · zou_aidouga · 图像视频创作分享34

    Don't take up an AI video side hustle if you've never made videos: the reasons edits get stuck and what to do before starting

    An independent video creator advises that people without video production experience should not rush into treating AI video as a side hustle right now; the core reason is that AI-generated video cannot be changed in just one place—every revision requires regeneration, causing visuals to change and costs and schedules to spiral out of control. In his September tests, rough-quality 1-minute video cost about 100 yen, but two videos generated from the same prompt and same image diverge visually from a certain second onward. He suggests first practicing with free editing apps to change only one thing at a time, cultivating judgment, and recording the number of redos and the cost.

  2. ArchiGen AI · 图像生成与ComfyUI实战18

    ArchiGen: perfection is not a render instruction—AI architectural rendering needs four acceptance gates

    ArchiGen proposes that the "definition of done" for AI architectural rendering should be split into four acceptance gates—project fidelity, communication, image quality, and evidence—rather than a single phrase like "looks good, resembles the reference image." The first gate requires listing protected features before generation (massing, floor and roof edges, window openings, facade rhythm, columns, stairs, etc.) and verifying them against the approved base image with overlay comparisons; any feature exceeding the agreed tolerance is sent back for rework.

Sep 26Sat
  1. ArchiGen AI · 图像生成与ComfyUI实战22

    AI relighting images are not daylighting analysis: ArchiGen proposes a five-sentence caption test

    ArchiGen AI points out that AI-relit interior images can present daylight effects with credible direction, softness and color temperature, but the model may not know the building's location, date, glass transmittance, surrounding obstructions and year-round weather data, so they cannot serve as evidence of daylighting performance.

Sep 23Wed
  1. Digiday · 微短剧营销与商业36

    Brands rush to bet on micro-drama marketing, but missing ROI may render it meaningless

    Micro-drama marketing is attracting brands such as Crocs, Nuuly, and Native (under Procter & Gamble) to enter the space. A Snap and Ipsos survey shows that 42% of daily active social media users watch micro-dramas every day. However, marketers have not yet solved the ROI problem; a dentsu executive says traditional metrics are misleading, and eMarketer data shows that in-app advertising revenue for ReelShort and DramaBox grew 31% and 29% respectively in Q1 2025.

Sep 22Tue
  1. ArchiGen AI · 图像生成与ComfyUI实战13

    2026 AI Rendering Software Rankings May Be Based on 2025 Evidence

    ArchiGen points out that AI rendering software comparison rankings titled '2026' may have pricing, version, and integration information sourced from earlier checks. The article recommends labeling every key field with a 'verified date' and a 'review due date,' and recording version, billing method, and host software support separately according to three different-speed clocks — product, business, and workflow — using four status markers: Current, Review due, Superseded, and Unverified, to avoid a blanket 'updated' label masking outdated facts.

Sep 21Mon
  1. 公众号:短剧深水区38

    How Does a 3-Year-Old AI Cute Baby Beat Cultivation Grandmasters? Breaking Down the Cute Baby Genre with "Listening Water Little Koi" and "Nuonuo Descends the Mountain"

    The AI realistic-human short drama "Listening Water Little Koi" accumulated 1.2 billion views on Douyin's paid tier in 4 days, with a single-day peak of 700 million, and 56.01 million heat on Hongguo's free tier. The AI comic drama "Nuonuo Descends the Mountain, All the Senior Brothers Are Panicking" surpassed 1 billion Douyin views after launching on August 25, with over 10 million pre-orders on Hongguo before Season 2 went live.

Sep 20Sun
  1. ArchiGen AI · 图像生成与ComfyUI实战22

    Same-Model AI Rendering Comparisons Still Need to Disclose Inputs, Prompts, Attempt Counts, Selection, Editing, Time Spent, and Failure Records

    A four-tool AI rendering comparison starting from the same SketchUp model is inherently unfair: plugins can read the real-time viewport and camera, browser tools only receive exported JPEGs, and real-time engines rebuild materials and scenes, so each product starts with different information. The article proposes a seven-item disclosure checklist—actual inputs, preparation work, instructions, attempt counts, selection rules, editing and failure records—and suggests listing preparation, waiting, correction, and export time along with credits consumed. The author also advocates distinguishing between "input-controlled image tests" and "workflow-native tests," arguing that scores from the two should not be combined.

  2. ArchiGen AI · 图像生成与ComfyUI实战12

    Why Browser Canvases, BIM Plugins, Real-Time Renderers, Image Editors, and Feasibility Engines Can't Share a Single Leaderboard

    ArchiGen analysis points out that putting browser-based collaborative canvases, BIM plugins that read modeling viewports, real-time rendering engines, and general-purpose image generators on the same leaderboard is a classic case of category mismatch. The article suggests first defining categories along four boundaries—source, task, deliverable, and revision path—then setting up four tracks: model-linked concept rendering, image-based visual development, real-time scene production, and dedicated tools, each with its own acceptance criteria and hard exclusion items, rather than continuing to stack star ratings.

Sep 18Fri
Sep 17Thu
  1. Creator Animation · AI影片制作教程12

    Why AI filmmakers are leaving ad-supported platforms: Creator Animation's independent distribution approach

    The core reasons AI filmmakers leave ad-supported platforms are ads interrupting viewing, algorithmic gatekeeping, and fragmented revenue splits. Creator Animation offers an independent distribution channel where creators keep 80% of revenue and the platform takes 20%, supports content up to R-rated, and prohibits pornographic content.

Sep 16Wed
  1. 壹娱观察 · 搜狐公开专栏34

    AI video meets world models: will interactive film-games finally reach maturity?

    ByteDance is rumored to be advancing a real-time spatial video model based on Seedance, with Zhang Yiming personally paying attention; games, short dramas, livestreaming, and PICO could all become application scenarios. Shengshu Technology's Vidu S2 demonstrated real-time character responses and instant changes at the same time; its VR spatial video version plans to generate binocular frames with depth for VR devices, but it is still at the teaser stage. Although world models can create the first impression of "being inside a world," they have not yet solved the problems of long-term consistency, memory, and narrative order.

  2. 公众号:短剧深水区32

    Why the "Shengdai CP" of Lin Daiyu and Sun Wukong took the internet by storm: AI fan works and the evolution of the ideal type in women-oriented narratives

    In the summer of 2026, AI fan-made videos pairing Lin Daiyu and Sun Wukong as the "Shengdai CP" went viral across major platforms; on September 15 the related topic reached No. 17 on Weibo's entertainment trending list, and netizens called them "Miss Internal Friction and her all-purpose mister." Previously, an AI fan work of Xue Zhenzhu from "The First Half of My Life" exceeded 3.4 billion views on Douyin, and China Literature, Douyin, and Bilibili have opened AI fan-work licensing for classic IPs and set up dedicated creation funds. The article argues that the Sun Wukong-style protective male lead who is "strong but not oppressive" is replacing the domineering CEO as the new template for the ideal type in women-oriented short dramas.

Sep 12Sat
  1. 壹娱观察 · 搜狐公开专栏38

    The race for the first AI video IPO: who will list first among Shengshu Technology, Aishi Technology, and Kling

    Shengshu Technology has completed its conversion to a joint-stock company, Aishi Technology and Yanyu Technology have successively been reported to be heading for Hong Kong listings, and Kling, after completing an independent financing round, has been given very high expectations for an independent IPO—the race for the first AI video IPO is becoming increasingly concrete. Over the past year, AI video financing has risen noticeably: Aishi Technology's Series C reached $439 million, Shengshu Technology completed $500 million in financing, and Kling's subscription amount in this round totaled about RMB 20.447 billion. But listing first does not automatically bring better models or user renewals; companies ultimately still need to answer whether paying-user retention, customer renewals, and operating cash flow can improve.

Sep 8Tue
Feb 18Wed
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