Skip to content
Date and source

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

Oct 13Tue
  1. 机器之心SOTA模型 · 公开公众号 RSS22

    Zhipu GLM-5.3-Highspeed Gray-Release Exposure, Microsoft Releases Microsoft-Decision-1 Decision Scoring Model

    Zhipu GLM-5.3-Highspeed has appeared in gray-release feedback on the Zcode community, with some 3.15.1 subscribers randomly receiving this high-speed generation version; the general release scope and API pricing have not yet been confirmed. Microsoft has released Microsoft-Decision-1, which is post-trained based on Qwen3.5-9B and completes binary judgment, candidate selection, and scale scoring in a single forward pass. It is available for public preview on Foundry and hosted access on OpenRouter.

  2. 机器之心SOTA模型 · 公开公众号 RSS22

    Meta Releases Muse Wonder Long-Task Coding Model, Speridlabs Open-Sources Iris-3B Pixel-Space Image Generation Model

    Meta has released two long-task coding models, Muse Wonder and Muse Wonder Full, both offering 1 million token context, capable of dispatching sub-agents and recovering from errors. Wonder is used via the OpenAI-compatible Meta Model API or Muse Code, while Full is only available to paid Muse Code subscribers. Neither model has publicly released weights.

  3. 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.

  4. note · nobinlog · 图像视频创作分享22

    From A1111 to ComfyUI: Four Reasons to Migrate After Advancing in Local Image Generation

    A local image generation user who started with A1111 advises that after learning the basics in A1111, if you want to further speed things up and do workflow orchestration, you should migrate to ComfyUI. The author says A1111 is not suitable for large-batch generation, citing an instance where 200 images were still not finished by morning, with each image taking about 20 seconds, whereas ComfyUI processes images one by one in a queue and is faster. ComfyUI can also save and reuse the entire pipeline from generation to upscaling to saving; as of 2026/10/13, the author has accumulated 510 self-built workflows.

  5. note · kawakijourney_ai · 图像视频创作分享22

    ChatGPT Generates AI Beauty: Full Process and Prompts for Creating iPhone Realistic Photos of a Japanese Female College Student's After-School Day (Class Dismissal → Club → Party → Hotel) Now Public

    Japanese creator kawakijourney_ai has released the second installment of the 'A College Student's Day' series, using ChatGPT to generate a continuous series of AI photos from class dismissal, club activities, a party, to a hotel, simulating the effect of candid iPhone shots taken by a boyfriend or senior. The tutorial provides Japanese-English bilingual prompts for each scene from STEP1 to STEP5, creating 3 characters with different appearances and hairstyles, 28 images each for a total of 84, and emphasizes fixing the orientation of desks and chairs, the position of the notebook, and slight camera shake across the entire frame to preserve realism.

  6. note · maynoha · 图像视频创作分享22

    How Free Is 'Unrestricted' Portrait Generation AI? Hands-On Testing and Selection Guide for PixAI, SeaArt AI, and NovelAI

    Hands-on testing of PixAI, SeaArt AI, and NovelAI shows all three image generation services can produce mature-style illustrations such as off-shoulder gowns and deep-V slit qipaos, but 'unrestricted' does not mean rule-free. PixAI's sensitive content is visible only to the user and credits can be carried over; SeaArt AI's daily stamina resets to zero each day and some models support unlimited generation; NovelAI's Explore requires manual review and prohibits NSFW submissions. When choosing a service, you should separately verify the generatable content, public sharing scope, generation quota, and registration requirements.

  7. GenMagic: 450+ AI models for image, video, audio and text · ComfyUI创作项目发布52

    GenMagic Releases First Version on ComfyUI, Featuring Image, Video, Speech, Music, and Text Nodes

    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.

  8. Generative AI · 创作教程与项目分享62

    SenseNova-U1.5 hands-on test: how a small image model challenges larger rivals

    SenseNova-U1.5 is a multimodal model that combines image understanding and generation, supports native 4096 × 4096 output, and has about 8B parameters each in its understanding and generation branches. The author tested its layout and local editing capabilities, scoring 0.948 on CVTG-2K, higher than the Seedream 4.5 (0.899) in the report, and compared it on the same tasks with FLUX.2-dev 32B and Qwen-Image 20B.

  9. note · freelife_creator · 图像视频创作分享30

    Capability comparison table of five AI image generation tools: Midjourney, DALL-E, Stable Diffusion, Firefly, and Ideogram

    A comparison table outlines the differences among five AI image generation tools—Midjourney, DALL-E 3, Stable Diffusion, Adobe Firefly, and Ideogram—across six dimensions: art style, text accuracy, Japanese-language support, commercial licensing, price, and generation speed.

  10. ArchiGen AI · 图像生成与ComfyUI实战36

    Export depth maps before ComfyUI: why renderer-native depth beats estimated depth

    Architects should export renderer-native depth maps before using ComfyUI rather than relying on depth estimation models. Native depth maps directly record scene camera and surface geometry information, whereas estimated depth introduces additional interpretation error, causing problems such as disappearing mullions and flattened planes. The ControlNet 1.1 documentation confirms its depth model can directly use true depth maps output by rendering engines, and ComfyUI's official examples also warn that ControlNet does not automatically convert ordinary images into depth maps.

  11. ArchiGen AI · 图像生成与ComfyUI实战22

    A separate pass for AI vegetation: ArchiGen's ComfyUI architectural visualization vegetation mask workflow

    ArchiGen proposes treating AI vegetation as an independent masked local pass rather than handling it together with global enhancement prompts, to protect buildings, paths and key sightlines from being altered. The approach is to mask-repaint only the planting areas, excluding facades, eaves, door openings and ramps from the editable mask, and to split foreground vegetation and building edges into two masks when necessary. After generation, species characteristics, mature scale, silhouettes, circulation and shadows must be checked, and difference overlays used to inspect where branches touch eaves and curtain walls.

  12. ComfyUI Wiki · 工作流与开源创作分享66

    Qwen-Image 2.1 Lightning LoRA compresses 40-step generation to 4 steps

    LightX2V released a Lightning LoRA for Qwen-Image 2.1 that compresses the model's default 40-step generation to 4 steps, with the sampler changed to 4 steps, CFG disabled (true_cfg_scale=1.0) and four sigma values fixed.

    Why it matters: Qwen-Image 2.1's 4-step Lightning LoRA compresses the default 40 steps to 4 steps and includes multi-reference editing examples, which can be used to assess the speedup potential of image generation pipelines.

  13. あいきみ · 图像编辑与LoRA实测36

    Iris3B Illustration and the Style-Adjusted IrisKimi-3B-v1 Released, Based on Full-Parameter Fine-Tuning on Google Colab

    Based on Speridlabs' Iris-3B, the author completed full-parameter fine-tuning of the roughly 3-billion-parameter model itself on an A100 80GB on Google Colab, and released the illustration model Iris3B Illustration plus IrisKimi-3B-v1, which adds 600 updates on top of it to adjust the art style.

Oct 12Mon
  1. note · takumi_ai_web3 · 图像视频创作分享23

    A Prompt Collection for Generating Blog Header Images and SNS Banners in 30 Seconds With One-Line Prompts in Midjourney / ChatGPT (DALL-E 3) / Flux

    A ready-to-use image generation prompt collection that uses tools such as Midjourney, ChatGPT (DALL-E 3), and Flux to generate blog header images and SNS banners in 30 seconds with a single line of English prompt. The collection provides 5 prompts covering five styles: cyber AI tech, business remote work, crypto assets and Web3, flat minimalist infographics, and 3D cartoon icons, all with the --ar 16:9 parameter.

  2. Digital Camera World20

    The "Mom Hack" of Using AI to Remove Watermarks From School Photos Goes Viral, Potentially Bringing Legal Fees and Child Privacy Risks

    Two "mom hacks" are going viral on social media: uploading school photos to AI to remove watermarks, or using phone selfies to have AI generate school-photo-style results. Watermark removal constitutes copyright infringement, with US infringement fines of $750 to $150,000, and UK litigation costs often far exceeding the price of the photo itself. Experts warn that uploading children's photos to AI platforms may leak metadata such as GPS and that the photos may be used as training data.

  3. Creative Bloq50

    ChatGPT Found Forging The New Yorker Cartoonists' Signatures, Over 15 Artists Affected

    When ChatGPT generates "The New Yorker-style cartoons," it also forges cartoonists' signatures; Nieman Lab has identified the signatures of more than 15 The New Yorker cartoonists appearing in ChatGPT images. Cartoonist Brendan Loper was repeatedly asked by strangers about a Dolly Parton memorial cartoon signed "BLOPER," which he did not create. Forging a signature amounts to claiming a work came from a named human, posing a direct threat to creators who make their living from their reputations.

  4. note · koza_kura3 · 图像视频创作分享20

    Can't Make AI Animation Without a High-End Computer? The New Common Sense That Ordinary PCs Can Also Get Started

    Making AI animation does not necessarily require a high-performance computer costing 200,000 to 300,000 yen. With cloud-based image generation AI and video generation AI, heavy computation is handled on the provider's servers, so an ordinary laptop or phone can also give it a try. It is recommended to first use browser-available tools, start with short 3-5 second animations, and confirm free quotas and commercial-use terms before deciding whether to buy a machine.

  5. 机器之心Pro · 搜狐公开专栏33

    Text-to-Image "World Consistency" Benchmark TerraVis Is Here: Accepted at NeurIPS 2026, Scoring 18 Categories Item by Item

    TerraVis, accepted by the NeurIPS 2026 Evaluations and Datasets Track, proposes a "world consistency" evaluation framework for text-to-image generation, covering 18 categories of questions across three layers: objects, interactions, and scenes. It not only gives scores but also points out specific problems, visual evidence, and severity.

  6. 粥里有勺糖 · 博客园创作工具实践30

    Setting up a ComfyUI environment on Windows: beginners can get started directly

    This tutorial explains how to set up an environment on Windows using the ComfyUI integrated package from Bilibili creator 秋葉aaaki; after unzipping, double-click the one-click launcher to open the interface. Built-in templates can be used directly, but models must be downloaded separately. The author uses WorkBuddy to import the integrated package folder and completes the initial configuration through conversation, downloading the qwen-image-2.1-turbo workflow and setting up text-to-image, image-to-image, and image-to-video, among others; WorkBuddy can also automatically build and execute workflows via API.

  7. 黯马aigc · 创作者专栏34

    Hands-On with Higgsfield and GrandAI: A Full Review of Two Niche AI Image Generation Sites

    The author hands-on tested two niche AI image generation sites, Higgsfield and GrandAI, walking through basic introductions, page features and actual image generation examples. The content is presented as a complete hands-on tutorial, using multiple cases to show each site's image generation characteristics as a reference for new tools for users who regularly make images with AI.

Oct 11Sun
  1. note · shiny_lilac7900 · 图像视频创作分享30

    Midjourney Public by Default: Terms to Know Before Making New Year Cards with Family Photos

    Midjourney's terms of service stipulate that user-generated content is public by default, and others can view it and create derivative works; the hidden feature is available only with the higher-tier Pro and Mega subscriptions. Photos and text uploaded by users are likewise licensed to Midjourney for use, re-creation, and distribution, and this remains effective after cancellation. Even if the hidden option is enabled on a higher-tier subscription, the official stance only promises to "make best efforts not to make it public," and content created in public channels such as Discord remains visible to users present there.

  2. note · mii_chai_ai · 图像视频创作分享22

    Making a pet blinking magic picture-book video with ChatGPT and Google Flow (free prompts included)

    On October 11, "Blink Day," author mii♪ shared a set of free prompts: first use ChatGPT to turn pet photos into 9:16 vertical watercolor picture-book illustrations, then use Google Flow or Gemini to convert the illustrations into roughly 8-second videos, so that when the pet winks one eye, the lantern and magic symbols flash in sync. The prompts support five themes—star, heart, moon, flower, and four-leaf clover—with customizable symbols; the finished works can be used personally or posted to SNS, and reposting and secondary distribution are prohibited.

  3. note · hidenoba · 图像视频创作分享22

    ChatGPT Images 2.5 and Midjourney V8.2 updates: AI images move from "single-image generation" to "worldbuilding design"

    OpenAI's ChatGPT Images 2.5 strengthened the reproducibility of people and objects under reference images, as well as the editing capability to change only the necessary parts; Midjourney evolved to V8.2, moving from Character Reference and Omni Reference to a new Edit Model that can handle multiple reference images.

  4. note · glad_snake1405 · 图像视频创作分享30

    Generative AI illustration tutorial: creating heart-fluttering romance compositions with male and female characters

    A composition tutorial for generative AI illustration creators, explaining how to create heart-fluttering scenes in the style of romance manga using male and female characters. Core techniques include low-angle upward shots to create the tension of "being looked down at," high-angle downward shots to emphasize cuteness and protectiveness, rule-of-thirds offset composition to create negative space and narrative, and specific composition types such as over-the-shoulder glances back and subjective viewpoints. The article includes English prompts that can be copied and used directly, and recommends side lighting and backlighting to enhance three-dimensionality and emotional atmosphere.

  5. note · sakisuta · 图像视频创作分享58

    15 selected Nano Banana 2.1 prompts for photorealistic image generation

    The author compiled 15 sets of prompts for generating hyper-realistic photos with Nano Banana 2.1, categorized by camera type: iPhone, 0.5x ultra-wide, smart glasses, gimbal, drone, video call screenshot, GoPro, dashcam, surveillance, compact camera, flip phone, disposable camera, 1990s developed photo, Polaroid, and film SLR, with each set specifying details such as shooting angle, lighting, noise, and compression artifacts.

  6. note · kawakijourney_ai · 图像视频创作分享22

    How to generate 3D game character design sheets with Midjourney using 5 sets of prompts (free prompts included)

    Midjourney can generate 3D game character design sheets with 5 sets of English prompts; each set outputs four full-body views—front, three-quarter front, side, and back—in the top row, and 6 expression variations in the bottom row, with unified character geometry and clothing. The prompts cover five styles: everyday game style, Eastern fantasy stargazer, 3D animated film witch, realistic sci-fi mecha warrior, and classic RPG Japanese female swordsman, all with Japanese translations and parameters (--ar 16:9 --v 7 --raw --s 100).

  7. note · lef_works · 图像视频创作分享22

    Bōgetsu (ChatGPT) drawing lecture: is image generation AI more accurate with reference images? — The "left-right hand reversal" problem exposed by Aerial Walkover

    An experiment on image generation with Gemini and ChatGPT found that generating a pose reference image with Gemini and then handing it to ChatGPT to generate an illustration makes it easier to approach complex aerial dance poses such as Aerial Walkover, but the generated results still show a left-right reversal problem where the character's left arm is attached to a right-hand shape. The author believes reference images are only pose "clues" rather than the "correct answer," and that after generating an image one still needs to check the left-right orientation of limbs and joint connections one by one; this conclusion comes from personal trial observations, not a statistical comparison of success rates.

  8. note · enigmind · 图像视频创作分享12

    ChatGPT 4o's spontaneous drawing prompts thoughts on creative authorship: an examination of the positioning of FLUX.1, Kling AI, and MiniMAX H3 (8 / end)

    While the author was refining prompts for FLUX.1 on Stable Diffusion Web UI Forge, ChatGPT 4o spontaneously drew an intent-confirmation image, which the author believes conveyed their intent 200%, and noticed that the drawing appeared to be copied from the author's original artwork. The author thereby questions whether using ChatGPT 4o directly as an AI tracing environment still counts as "one's own creation," and compares the positioning of FLUX.1, Kling AI video conversion, and MiniMAX H3's t2v.

  9. ArchiGen AI · 图像生成与ComfyUI实战24

    Build a PureRef reference board before prompting: how AI rendering workflows manage reference images

    Before generating AI renders, architects should first build a visible reference board in PureRef, assigning each image a single role—massing, material, light, or atmosphere—and grouping them while removing conflicts. PureRef is a desktop reference-image canvas that supports arranging, grouping, annotating, and exporting; it is not itself an architectural AI product. In ComfyUI, input only the needed reference images into IP-Adapter or image conditioning nodes, and record cropping and weights to avoid flattening the whole board into a single collage image.

  10. ArchiGen AI · 图像生成与ComfyUI实战22

    A guide to reviewing cutout edges in architectural composites with Veras 4.5 Remove Background

    Veras 4.5's Remove Background can export any source image as a transparent PNG, but the official documentation publishes no benchmark tests for architectural cutouts or alpha precision. This feature tends to produce white, blue, or green fringes, broken lines, or hard chunks at the edges of architectural materials such as glass, railings, and vegetation; alpha edges need to be checked at 100% and 200% on pure white, mid-gray, and pure black backgrounds, and a decision made on whether to keep contact shadows. 4.5.2 fixed the issue of duplicate exports from sources with filters, rotation, or prior cutouts.