Skip to content
Date and source

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

Oct 15Thu
  1. 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.

  2. 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.

  3. note · tuwabuki1211 · 图像视频创作分享22

    AI Image Generation: The Trial-and-Error Process of Using Double Exposure to Create an Underwater City and Glowing Jellyfish

    When a creator used AI image generation to create "underwater city and glowing jellyfish double exposure," the results repeatedly went off track: specifying neon first produced a street scene with pink hotel signs, keeping only the jellyfish turned it into a monochrome glowing jellyfish, and reorganizing the prompt also had no effect. In the end, the creator changed to directly specifying the phenomenon "a neon street scene reflected inside the jellyfish's bell," and only then was the target image generated.

  4. 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.

  5. 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.

  6. 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.

  7. note · promptdesign · 图像视频创作分享22

    New AI beauty image prompt collection ⑫: 100 templates of spectacular scenery around the world × characters, ready to copy directly into ChatGPT and Gemini

    This prompt collection contains about 100 image generation templates of "spectacular scenery around the world × adult women," all in JSON format that can be copied directly into tools such as ChatGPT and Gemini. In addition to the basic 1:1 version, each theme also provides a 9:16 vertical version, a 1280×670 note header image version, and adapted versions with changed background, lighting, clothing, and expression, covering scenes such as the Charles Bridge in Prague and Rio beaches, and can be used for note header images, SNS posts, and vertical short video materials.

  8. note · singulatelier · 图像视频创作分享10

    Paying for X Premium but only using Grok to generate videos: one user rediscovers Grok's research and data analysis uses

    A paying X Premium user previously used Grok only as a simple video generation tool, but now has begun trying it to research trending topics and user reactions on X, and hands the results to their 3 AI employees—Theresa in charge of investing, Rebecca in charge of Rakuten ROOM product selection, and Penelope in charge of news intelligence—to assist with analysis and product selection. How much X information Grok can reference and whether its answers are accurate remain to be verified, and the AI employees do not automatically coordinate with each other; information still has to be moved by manual copy-paste.

  9. note · anpontan_note · 图像视频创作分享22

    nano banana pro Generated Title Image Japanese Text Wrapped into Two Lines: Fix It by Adding One Instruction to the Generated Image Without Changing the Prompt

    When using nano banana pro to generate a title image, the prompt specified that 「15分は手作業だった」 should be on one line, but the output image wrapped it into two lines. Instead of rewriting the prompt, the author added a short instruction, 「一行にして」, once to the generated image, and it was corrected back to one line in one try. The specified section of header-designer.md currently has no line-count specification, and whether to add one has not yet been decided.

  10. note · myloose · 图像视频创作分享8

    AI Image Generation Study: Using Coordinate-Based Local Editing to Remove the Double Eyelid Line from a Monolid Male Character

    The author found that the image generation tool supports local editing of areas specified by coordinates, and tried to remove only the separate crease on the upper eyelid of a monolid male character while not changing the eye contour, eyeball, eyelashes, eye-corner angle, eyebrows, or hair; the first attempt succeeded. However, in a wide-shot composition, the arrows and white boxes were too thick, causing the specification to be ineffective; in a two-person close-up, the character's double eyelid line instead became fainter, and remnants were left at the split points of hair strands. The author ultimately considered switching to Astra, ChatGPT's current top model, for image generation.

  11. note · hagiri1312 · 图像视频创作分享12

    The AI Illustration Was Cute, but I Didn't Use It This Time: A Record of Seeing My Own Preferences Clearly After One Trial

    On October 10, I asked AI to generate a trial illustration, gave feedback midway that "I want a cute style rather than a refined style" and regenerated it, but ultimately still decided not to use it on the grounds that it was "cute but too AI-flavored." From this, the author distinguishes "thinking the finished product is good" from "wanting to continue in this direction" as two different things, and suggests that before commissioning next time, first clarify what you like, the lingering sense of incongruity, and what to judge in the next image, then decide whether to generate more.

  12. note · pixolm_ai · 图像视频创作分享22

    6 Halloween AI Image Prompts: Photo Processing and Icon Creation

    Pixolm used Nano Banana 2.1 under Custom Create with 1K and Standard settings to generate 1 image each, completing 6 Halloween examples: 3 photo-processing images based on the same fictional adult female reference image, and 3 icons generated purely from text, at 8 Credits per image and 56 Credits in total. The photo processing preserved the person's facial features and 4:5 composition, replacing only clothing and background; the icons used no character reference, instead specifying the character, color scheme, and circular negative space.

  13. GitHub:AI视频开源项目发现38

    StreamEval Releases an Interactive Streaming Video Generation Evaluation Benchmark

    StreamEval is the first evaluation benchmark for interactive streaming video generation, covering sessions from 15 to 300 seconds and testing a system's ability to respond to new requests and maintain context during video generation. The benchmark includes five tasks—digital human dialogue, virtual human performance, world generation, instruction editing, and reference editing—and ten metrics, each scored on a 0–100 scale, with no combined total score across tasks. The data, evaluation code, protocol, and scoring rules are all public, and the models and leaderboard are updated monthly.

  14. note · genkidesu2021 · 图像视频创作分享18

    How to Write Image Generation AI Prompts: 3 Major Elements, Specific Examples, and Tool Recommendations for DALL-E 3 and Midjourney

    This tutorial breaks image generation AI prompts into three major elements—subject, composition/angle, and art style—and gives specific rewriting examples such as "a field of fully bloomed large yellow sunflowers" and "a black cat curled up asleep in the sunlight." For tools, it recommends DALL-E 3 (which can be given instructions directly in Japanese in the paid version of ChatGPT and supports local edits) and Midjourney (runs on Discord, high image quality, English input recommended). The author also reminds readers not to pile on too much information, to choose according to what each tool is good at, and to fine-tune repeatedly rather than expecting a one-shot result.

  15. 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.

  16. 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.

  17. 可灵官方博客15

    How Kling IMAGE 3.0 Omni uses AI for virtual try-on of wedding veils

    Kling IMAGE 3.0 Omni supports multi-image reference generation, combining a user's wedding photo with veil reference images to preview different veil lengths, lace and wearing positions while preserving the face, hairstyle and wedding dress, and outputs 2K or 4K images. This feature only shows appearance; it cannot reflect fabric weight, the drape when walking, or wearing comfort, and is suitable for narrowing down choices before a fitting.

  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. note · bold_oxalis9933 · 图像视频创作分享10

    Wanting the cat to strike a JOJO pose but running out of AI image generation quota: lessons from Canva, Google AI Mode and English prompts

    A user tried to use AI image generation to make their cat strike a JOJO pose, repeatedly adjusting Japanese prompts without success, used up their free quota, and ultimately failed to complete it. During the experience, Canva's generated image quality was excellent, but free services are generally limited by quotas; after switching to Google Search's AI mode, they felt its image generation restrictions were more lenient than Gemini's.

  21. note · tidaa · 图像视频创作分享42

    How to use Gemini Omni: Generate a CM video with 5 images and a storyboard, then edit it into a finished piece with Canva (Part 2: Video Generation and Editing)

    Using 5 key visual images and a storyboard generated with Midjourney, the author generated 5 vertical 9:16 video clips in Gemini Omni, each approximately 3 seconds long, corresponding to five shots: hook, opening the bottle, drinking, whip pan to the sea, and beach afterglow. The author then assembled the 5 video clips into a single CM in Canva, originally planned for 15 seconds, with a final cut of 20 seconds.

  22. note · koheii_miaa · 图像视频创作分享65

    How an AI musician uses a single Mac to produce an MV with lyric sync and lip-sync, 17 minutes per track

    The author published a workflow for locally generating MVs on a Mac: input a song and cover image, output a 30-second video with synchronized lyrics and a cover character singing, in both 16:9 and vertical formats, approximately 17 minutes per track, without using cloud-based paid generation.

    Why it matters: The author published the complete local workflow from vocal separation to lyric sync with time breakdowns, allowing for an assessment of the feasibility of producing MVs on a single machine.

  23. note · lumo_ai · 图像视频创作分享70

    How MiniMax H3's 360-Degree Orbit LoRA Achieves a Full Rotation Without Breaking

    MiniMax H3's 360-degree orbit LoRA generates a video from a single photo where the camera orbits around the subject and returns to the starting point. The approach involves placing the same image in both the first and last keyframes, resulting in approximately 3 seconds and 73 frames. Recommended settings are 768×768 resolution, 28 steps, and LoRA strength of 1.0. It works in ComfyUI without requiring custom nodes, and the file is approximately 148MB.

    Why it matters: A breakdown of MiniMax H3's 360-degree orbit LoRA usage and training data sources, along with an explanation of back-face inference and applicable boundaries.

  24. Stable Diffusion Art 教程78

    Using ComfyUI to Generate Audio-Enabled Videos from a Single Image with MiniMax H3 (Includes Workflow)

    Stable Diffusion Art published a tutorial explaining how to use MiniMax H3 (i.e., Hailuo 3.0) in ComfyUI to generate 24fps multi-shot videos with synchronized stereo audio from a single image, with visuals and sound effects generated in the same sampling pass.

    Why it matters: The tutorial provides the complete workflow for image-to-video generation with MiniMax H3 in ComfyUI, including start/end frame interpolation and a time-saving approach using 8-step Turbo LoRA.

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

    Using MiniMax H3 to Create a Vertical Short Film About Link and Zelda from The Legend of Zelda Who Don't Want to Be Toys Anymore

    A creator used MiniMax H3 to generate a vertical short film telling the story of Link and Zelda who don't want to be toys anymore. The author made several upgrades to the standard workflow, using KDEnlive to add Gaussian blur and film grain to live-action scenes, eliminating the common 'oily skin' texture often seen in AI-generated videos.

  26. note · seisei_ai_1st · 图像视频创作分享38

    100 Ways to Request AI Portrait Composition: 25 Composition Prompt Templates with Examples

    This AI image creation guide breaks composition down into 100 reusable prompt templates, organized into 25 groups of 4 each by categories such as 'placement' and 'lines,' with example images generated using the same model. Below each composition is a code block that can be directly copied to the end of a prompt, such as '3×3 grid layout,' '1:1.6 golden ratio balance,' and 'diagonal cross layout.'

  27. note · fuckdesune · 图像视频创作分享20

    Using ChatGPT to make original characters do "expressions they would absolutely never make": sharing a 2×3 collage project prompt

    A Japanese creator shared a set of prompts for using ChatGPT to generate a 2×3 collage project of original characters making "expressions they would absolutely never make," requiring that core features such as the character's facial features, hairstyle, and clothing remain unchanged, with only the expression, gaze, gestures, and acting deviating from the everyday persona. The author lists 6 forbidden expressions each for characters such as いの, エアリア, 灰音, 玲, 一臣, and 駄文ちゃん, and provides both a general version and a fill-in-yourself template, and also suggests first briefly introducing the character to ChatGPT and then asking about "解釈違いの表情" to get inspiration.

  28. Wireflow · AI短剧与视频制作教程32

    The current state of the CapCut API: what options and alternatives are available in 2026

    CapCut still has no official editing or rendering API, its terms of service prohibit automated script access, and developers.capcut.com has no public DNS record. CapCutAPI-type projects on GitHub (such as VectCutAPI and capcut-mate) can only generate draft files, which still require manual export in the app, and newer versions of CapCut and Jianying have encrypted drafts or rejected old templates.

  29. note · palealloy · 图像视频创作分享12

    Krea2 generates AI illustration "Autumn Sky Whale Ship Festival 2026": ComfyUI direct-output workflow and Detailer detail trade-offs

    The author used Krea2 in ComfyUI to generate an illustration for the X platform event "Autumn Sky Whale Ship Festival 2026," saying that リアス was completely unusable for similar subjects while Krea2 performed outstandingly. Currently a direct-output workflow without upscaling (アプスケなし) is used, but the author found that too much detail description causes fine details to break down, and is considering adding processing such as a Detailer.

  30. AITuber · AI短视频与提示词29

    How to use AI to clone the format of a viral video

    AITuber released a guide explaining how to use AI to clone the format rather than the material of a viral video: record the hook, pacing, duration, and subtitle style, then rewrite the script with your own topic, and have AI generate entirely new visuals, voiceover, and subtitles. The article uses a 35-second science explainer Short as an example and demonstrates in four steps; for the talking-head version, you can upload a clear photo to generate an AI avatar, and for voiceover you can choose from AITuber's 1,300+ voices.