Skip to content
Date and source

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

Oct 16Fri
  1. Reddit comfyui · 创作与工作流分享38

    Template site prompts perform poorly when migrated across models: LTX-2.5 requires rewriting in the model's own style rather than direct copying

    Copying prompts from template sites to test LTX-2.5 consistently yielded poor results until the user compared them with the reference prompts built into the official workflow. The built-in prompts are approximately 200 words and include explicit camera movements, pacing nodes, and on-screen text typography details (e.g., 'pure white sans-serif, wide spacing, no glow, no shadow'), whereas most shared prompts are tag-list style written for Midjourney/Ideogram. Rewriting the same scene in the model's own expected style produced dramatically different results.

  2. Reddit comfyui · 创作与工作流分享62

    ComfyUI tutorial: Using RefMod with MiniMax H3 to maintain character consistency without LoRA training, tested with 6GB VRAM

    The author released a ComfyUI workflow that combines RefMod with MiniMax H3, enabling the generation of new images of the same character without training a traditional character LoRA. The process involves preparing 8 character photos, using the RefMod custom node to generate a RefMod file, then loading the MiniMax H3 workflow and entering prompts for scene, clothing, pose, or environment.

  3. note · miyuu_2026 · 图像视频创作分享22

    Lakeside girl generation test: English vs. Japanese prompt comparison across Z-Image-Turbo, Flux.2-Klein, Qwen Image 2.1, Krea-2 Turbo, and Flux1-dev

    The author used the same set of lakeside girl prompts in ComfyUI to compare English and Japanese generation results across five models: Z-Image-Turbo, Flux.2-Klein 9B, Qwen Image 2.1, Krea-2 Turbo, and Flux1-dev. Overall, English prompts produced more realistic results, Krea-2 Turbo performed best, Qwen Image 2.1 exhibited obvious AI faces with poor results, and Flux1-dev could not generate with Japanese prompts.

  4. postPerspective36

    Nvidia Editor's Day: New Windows RTX Spark computers unveiled

    Nvidia unveiled the RTX Spark line of laptops and desktops at Editor's Day in San Francisco, emphasizing local AI processing and agentic AI. The SparkN1x chip integrates a 20-core Grace CPU with a Blackwell GPU, offering up to 128GB of unified memory, 6,144 CUDA cores, and 1 petaflop of NVFP4 compute, roughly six times the performance of Qualcomm's fastest solution.

  5. Monster社区 · 中文ComfyUI实测与工作流62

    Image Tagger 1.3.0 released: video prompt reverse-engineering, translation, and batch export

    Image Tagger V1.3.0 chains prompt reverse-engineering from reference videos, bilingual editing, batch export, and long-video segmentation into a single workflow, with this release focusing on prompt configuration for target models such as Seedance. To use it, first create a project and import online or local videos, then select the model series and conditional modes (T2V, I2V, full-modal, video extension, video editing) based on the intended generation use case. After reverse-engineering, prompts can be translated between English and Chinese for side-by-side editing and saving.

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 Releases ComfyUI Repackaged Model Files for Flux 3 Image Klein

    Black Forest Labs has released ComfyUI repackaged model files for Flux 3 Image Klein on Hugging Face. The original model repository is black-forest-labs/flux-3-image-klein.

    Why it matters: The official documentation provides the file directory and multiple quantized versions of Flux 3 Image Klein in ComfyUI, making it easy for local deployers to choose as needed.

  3. pixaroma · YouTube 创作视频22

    ComfyUI Nodes for Beginners: Essential Basics Explained

    The first video in the ComfyUI node explanation series is released, introducing the basic concepts of nodes, including inputs, outputs, settings, and the meaning of connection colors. The video demonstrates switching between Nodes 2.0 and the classic interface, four ways to add nodes, connecting nodes and adjusting parameters, and uses Flux 2 Klein 9B to build a complete text-to-image workflow step by step from a blank canvas. The model uses a non-commercial license, but generated images can be used for any purpose, including commercial use.

  4. Reddit comfyui · 创作与工作流分享22

    ComfyUI 0.40.0 update reported to have severe bugs: broken interface, unusable

    After the ComfyUI 0.40.0 update, severe bugs appeared: the interface is broken, images are displayed repeatedly, and normal navigation is impossible, making the software almost unusable. The user said they encountered the problem while using ComfyUI Desktop on a Mac M1 Ultra, and mentioned it may be related to the custom nodes Impact Pack and latent Switch node.

  5. AI-Canvas-tauri · 公开创作工具发布36

    AI Canvas v0.10.8 released: adds ComfyUI model deduplication and run console

    AI Canvas v0.10.8 adds ComfyUI model deduplication and a run console; it can find duplicate models by file content and replace selected copies with hard links or soft links, preserving the original path and file name. The run console supports starting and stopping the app-hosted ComfyUI and viewing real-time output. The toolbars of image, video, text, and audio nodes are unified into a quick command bar, and material references support searching by node number, type, and name.

  6. Reddit StableDiffusion · 创作者与工作流分享60

    Testing Krea 2 Turbo: 77 ethnicities × 7 prompt sets, same seed to see how ethnicity changes the output

    The author conducted a controlled experiment in ComfyUI with Krea 2 Turbo using 77 ethnicities × 7 prompt sets and a fixed seed. When the full-body prompts did not specify clothing, 34 of the 77 women were nude or topless, of which 17 of the 21 former Eastern European ethnicities were among them; after specifying concrete clothing, the 231 images had zero nudity. The 45 European ethnicities converged to only about 2 faces under full descriptions; when only the ethnicity name was written, all had brown updos and none were blonde, with hair color and eye color in the prompt carrying almost all of the differentiation.

  7. Reddit StableDiffusion · 创作者与工作流分享12

    Recommended Minimax H3 video upscaling workflow: how to complete generation and upscaling within 24GB VRAM

    A user on Reddit asked about the best upscaling workflow for Minimax H3 (MH3), hoping to complete both video generation and upscaling in an environment of AMD RX 7900xtx (24GB VRAM), 48GB RAM, Linux ROCm + ComfyUI. The user currently uses quantized GGUF variants to avoid OOM, with a regular generation spec of 0.2MP and about 10 seconds duration.

  8. 沃垠AI · 公众号公开订阅62

    CapCut launches new AI features including CapCut Assistant and CapCut Hub

    At its AI new creation launch event in late September, CapCut comprehensively upgraded the PC version and APP, launching CapCut Assistant (called Xiaoying on the APP) and the canvas-based product CapCut Hub. The author tested using CapCut Assistant to edit 8 video clips into a vlog with voiceover, subtitles, transitions, and filters, taking about 5 minutes in total, and said that repetitive operations such as trimming voiceover scripts, slicing long videos, and commentary videos can all be handed over to it.

  9. note · tsugiito_design · 图像视频创作分享40

    ComfyUI uses Animagine XL 4.0 Opt to generate a kimono character reaching out a hand: tags and workflow JSON

    The author used ComfyUI with Animagine XL 4.0 Opt to generate a kimono character reaching toward the camera, with black hair and long sleeves fluttering in the wind, without using any LoRA, completing the output on an RTX 4060 8GB at 1344×768, 28 steps, CFG 5, Euler a.

  10. Creative Bloq42

    I used Nvidia RTX Spark's AI agent to operate Blender, and here's how it went

    On an ASUS ProArt P16, the locally running Hermes agent (Qwen3.8 27B) could drive Blender to complete simple tasks such as warming the lighting, changing hair color, and exporting before-and-after comparison images, but struggled with complex multi-step edits: adding clown makeup to the model took about half an hour and only produced an approximate effect; moving an object across scenes, estimated to take 5-10 minutes after 25 minutes of preparation, was still unfinished after an hour and a half.

  11. openscreen · 公开创作项目发布22

    OpenScreen v2.1.0-rc.1 released

    OpenScreen released v2.1.0-rc.1, a release candidate for the v2 series, adding a live demo to the homepage and publishing a v2 announcement. This version fixes multiple issues in preview, recording, timeline, and export; MP4 export now supports mid-way cancellation and only publishes complete files; and a new MCP checkpoint feature allows rolling back an entire agent turn in one step.

  12. note · kamiryu_ai · 图像视频创作分享36

    Making a 15-second coffee product ad with AI video: generated with Higgsfield, edited with Claude Code, but the final product bag couldn't be fixed correctly by AI

    A creator used Higgsfield to generate a 15-second vertical 720p coffee drip bag ad in one go, consuming 68 credits, but the product bag appearing in the last approximately 2.4 seconds was replaced by AI with a large stand-up pouch, and the text on the bag became meaningless letters.

  13. note · toshia_fuji · 图像视频创作分享22

    Generative AI news roundup: DigUp, Fish Audio Drama 3, Mirelo SFX 2.0, OmniHuman 2.0 and other new tools released

    Fish Audio released the voice AI "Drama 3," supporting up to 16 speakers, emotional performance, and voice editing, with the model data and code open-sourced. Mirelo SFX 2.0 can automatically generate synchronized sound effects for videos and remove unwanted music, processing about 10-minute clips per run. BytePlus's OmniHuman 2.0 supports generating talking and singing videos up to 300 seconds long from a single image plus audio, and can be called via API.

  14. 娱乐资本论 · 搜狐公开专栏62

    AI video canvas feature comparison: 10 platforms diverge by comic drama, film and TV, and e-commerce going global

    Entertainment Capital Review conducted a feature comparison of 10 AI video canvases—AniShort, Jurilu, LibTV, Nadu Pro, OiiOii, Seko, ShotLab, TapNow, Wondershare Drama Factory, and Xiaoyunque—with data as of October 5, 2026. The review organizes each platform's focus according to three user types: comic drama serials, professional film and TV, and e-commerce going global, and lists the standard features all 10 have as well as the exclusive capabilities held by only one or two. Similar features such as consistency anti-face-swapping, script storyboarding, audio-video separation, and team collaboration are also compared item by item.

  15. Reddit StableDiffusion · 创作者与工作流分享50

    MiniMax H3 with Fizgig Nodes Enables Stylized Illustration Generation: Native 3072x1536 in One Minute

    MiniMax H3 combined with Fizgig Nodes can perform text-to-image generation of stylized illustrations, natively generating a 3072x1536 image in 16 steps on a 5060 Ti 16GB in about 1 minute. Its prompt comprehension is close to Krea 2, with fewer out-of-the-box limitations and good performance on concepts and world knowledge, but it requires a LoRA trained on static high-resolution style images.

  16. note · loyal_owl5720 · 图像视频创作分享22

    AMD XDNA 2 NPU speeds up Anima image generation by about 40%, bringing a single image down to the 35-second range

    When running Anima image generation on the AMD XDNA 2 NPU, by adjusting the timing of the MIMALLOC_PURGE_DELAY=0 setting, generation speed improved from about 60 seconds to an average of 35.81 seconds per image, a speedup of about 40%. This optimization only takes effect when the NPU is used; generating on the CPU alone still takes an average of 55.47 seconds per image. It is currently unclear why this setting is only effective for the NPU, and power consumption will be quantitatively evaluated later.

  17. note · seisei_surumono · 图像视频创作分享22

    ComfyUI virtual human node development (Part 1): using Live Chat stream to generate and maintain character consistency

    A developer has begun building a virtual human node for ComfyUI, combining several previously self-made custom nodes and using Live Chat stream to let the character generate its own appearance during conversation. Although features such as silver hair, sailor uniform and blue eyes are not fully fixed, the character still maintains roughly the same identity, with T2I-generated images used as input in the first round. The node combines a vrm model with voicevox to complete conversations, and connects to ollama and embedding gemma2 to handle memory and personality; it is currently at a stage where it runs successfully.

  18. note · kiryu_mizuki · 图像视频创作分享12

    Runs on 6GB of VRAM too: generation practice with Krea2, Qwen-image2.1 and MiniMax H3

    A creator with only 6GB of VRAM shared that, after being told at first that they could only use SD1.5, they can now generate images with Krea2 and Qwen-image2.1, and use MiniMax H3 for video. They noted that Qwen-image2.1 has very strong reference consistency, while Krea2 excels at character skin texture and illustration quality; they currently mainly use these two models for image generation.

  19. Ciaro · 中文AI电影制作教程22

    Professional video editing software: acceptance checkpoint for ingesting shot footage

    For ingesting AI-generated shot footage, editors must first set acceptance criteria such as frame rate, delivery aspect ratio and planned edit points, then assign each candidate version one of four outcomes: pass initial review, normalize, regenerate, or defer. Check items cover shot identity, playback, action and editing headroom, aspect ratio cropping and audio; failed items must be turned into clear actions and decision records kept.

  20. Production Expert39

    Why audio professionals need agentic AI, not generative AI

    DaVinci Resolve 21 has integrated agentic AI; Claude or ChatGPT can control the software via API with prompts. In testing, it automatically synced four clips from two cameras without timecode and one audio track without timecode on the timeline, generated a transcript, picked out the best two minutes, and performed loudness normalization to YouTube standards.