Skip to content
Date and source

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

Oct 9Fri
  1. Wireflow · AI短剧与视频制作教程26

    ComfyUI vs Automatic1111: which Stable Diffusion UI to use in 2026

    The core of the ComfyUI vs Automatic1111 comparison is the trade-off between control and ease of use: ComfyUI uses node graphs, is faster, and is still actively updated, with v0.39.0 released on October 5, 2026; Automatic1111 is easier to get started with, but has not been updated since v1.10.1 on February 9, 2025, and does not support Flux.

  2. Hugging Face Blog46

    Building six small models on Hugging Face with ML Intern: from a 0.8B prompt rewriter to a camera-angle LoRA

    Because the official Qwen-Image 2.1 prompt rewriter is a 9B model requiring about 20 GB of memory, the author used ML Intern in HuggingChat to build a 0.8B version that runs on CPU, achieves a 99.7% valid output rate, and consumes about one quarter of the teacher model's tokens, with a total compute cost of $16 including 8,797 annotations.

  3. note · honest_newt5115 · 图像视频创作分享12

    Opening an art shop with AI image generation at 49: a record of WONDER & DESIRE's creation and sales

    The 49-year-old author, finding increasing enjoyment in AI image generation, spent one week opening a digital art shop, WONDER & DESIRE | Art Studio, on Gumroad, STORES, Pinterest, Instagram, and X; so far there have been no sales at all. The author set up a planning department and a production department within ChatGPT and established a shared specification document, then used the new process to complete the first work, "黄金の月夜を舞う魔女と黒猫," which was finalized after two rounds of proofreading.

  4. Midjourney 官方更新50

    Midjourney tests Thinking Mode for image generation

    Midjourney is testing a new Thinking Mode for image generation on the Alpha website (alpha.midjourney.com), which users can run by clicking "Rerun (Thinking)" in the lower right corner of a job's lightbox. Tests show the mode helps improve prompt accuracy, text layout, and image coherence. Midjourney says it may in the future make it a general thinking mode, or "add more thinking" after a job is completed.

Oct 8Thu
  1. note · noble_mink9287 · 图像视频创作分享22

    College Student Creates "Future Tokyo" Video with AI: Generated with PixVerse, Edited with CapCut, and Submitted to TikTok

    A college student who is an AI beginner created a video themed "Tokyo 100 Years from Now" using AI. They first used AI to generate images blending Japanese castles, cherry blossoms, high-rise buildings, and aerial transportation, then used PixVerse to convert static images into dynamic footage, and finally completed editing with CapCut and submitted it to TikTok. The author emphasizes that AI is not fully automatic—prompt design and material selection still require personal judgment.

  2. 机器之心SOTA模型 · 公开公众号 RSS22

    Anthropic Releases Claude Haiku 5.5; Mistral Opens Mistral Large 4 Public Preview API

    Anthropic released Claude Haiku 5.5 on October 7, the first Haiku model to support adjustable reasoning intensity, targeting high-frequency tasks such as summarization, classification, and coding sub-agents. When prompts do not exceed 100,000 tokens, pricing is $0.10 per million input tokens and $0.50 per million output tokens.

  3. 张小闲 · 创作者专栏38

    Tutorial on 3 Methods for AI Poster Replication: Reference Image Prompt Reverse-Engineering, Color Block Sketch Control, and Partial Replacement

    Addressing the need to replicate poster effects with AI, the tutorial presents 3 methods: reference image with reverse-engineered prompts, color block sketch control, and partial replacement. The first step is to upload the reference image to AI to reverse-engineer prompts for composition, color scheme, lighting, texture, and style, then select a method based on image complexity. The partial replacement method allows directly specifying products, characters, or scenes to replace, and is the fastest.

  4. ArchiGen AI · 图像生成与ComfyUI实战30

    Reference Images Clashing? Assign Each Image a Clear Role

    To address the issue of reference images conflicting with each other in AI rendering, ArchiGen recommends assigning each reference image a unique role before generation—such as lighting, material, or occupancy—and comparing them pairwise to identify contradictions. If an image's harmful features outweigh its useful instructions, it should be removed from the generation set. Tools such as Veras, ComfyUI's IPAdapter, and Midjourney's image and style reference can be used, but conflict review must be completed before adjusting weights.

  5. ArchiGen AI · 图像生成与ComfyUI实战29

    Which Control Map Should Architects Use in ComfyUI: Don't Infer Geometry That the Project Can Directly Export

    ArchiGen proposes a four-tier source priority for architects selecting control maps in ComfyUI: prioritize geometry exported from the project model, followed by native renderer channels, and only use preprocessors like Depth Anything V3 to reverse-engineer from rendered images when original depth or Canny channels are unavailable. Each step down loses project information; inferred depth maps are merely "plausible-looking" and do not equal the project's actual geometry. Control maps must be matched with compatible ControlNets, and outlines, openings, and details should be checked at the final delivery scale.

  6. SuSu_酥酥12

    Dario achieves a comprehensive phased victory in AI models

    Dario has substantively achieved a comprehensive phased victory in AI models! Evidence: 🔸 With Opus 5.5, Sonnet 5.5, and Haiku 5.5 delivering a precise, high-quality triple strike, plus optimization features like the Claude Code Mods plugin, Dario has claimed the throne of AI dominance. 🔸 Sam Altman halted GPT-6.1 Astra, crippling his own efforts; GPT-6.1 Sol lost to Opus 5.5 and Sonnet 5.5, and GPT-6 Luna was defeated by Haiku 5.5. 🔸 Elon Musk announced that Grok Bot has officially integrated Opus 5.5, indirectly admitting that Grok 4.7 is inferior to Opus 5.5. 🔸 Sundar previewed Gemini 4 Argon as very strong, but it still hasn't opened for public testing—still just talk on paper—and is also missing the best window for model iteration. #Dario #SamAltman #ElonMusk #Sundar #Haiku55 #Opus55 #GPT61Sol #GPT6Luna #Gemini4Argon #Grok47

Oct 7Wed
  1. Ars Technica · AI62

    Google launches improved SynthID detector and opens it globally

    Google announced that its SynthID detector now supports watermarks from all partners and has launched a standalone website, SynthID.com, for anyone to use. SynthID is an invisible signal encoded in the pixels of AI-generated images and videos and in audio waveforms. Google says that Gemini alone has integrated this watermark into over 180 billion images and videos and 240,000 hours of audio.

  2. note · aicreatorpathnet · 图像视频创作分享22

    Grok Bot to integrate Claude Opus 5.5, Midjourney, and Suno? | AI trending topic

    Elon Musk previewed on X that Grok Bot will call the most suitable backend model for each task, naming Claude Opus 5.5, Midjourney, Suno, and other major APIs. How it will be used, which plans it applies to, pricing, and launch timing have not been announced, and the article does not say the feature is already available.

  3. note · prime_puma3624 · 图像视频创作分享24

    Hands-on test of ComfyUI's Detail Daemon: can AI illustrations really be "stacked" with it?

    Detail Daemon was retested in ComfyUI with the Anima environment, setting detail_amount to OFF / 0.20 / 0.30 / 0.40 / 0.50 to compare character and background generation results. 0.30 balanced detail and naturalness relatively well, while 0.50 caused image breakdown in both the character and background conditions. The author considers it not a high-quality magic node, and since Anima itself can add information through prompts, does not plan to keep it in the workflow for now.

  4. note · eraser107 · 图像视频创作分享22

    After putting past Nijijourney works into Midjourney v8.2, the world changed

    Combining Midjourney v8.2 with past Nijijourney works produced notable results; the author believes it makes up for Nijijourney's shortcomings and is clearly better than v7. Using exactly the same prompts and reference images while adjusting stylize and weird values for verification, it produced an artistic style with an acceptable subject. The recommended approach is to feed historical works generated by Nijijourney directly into v8.2.

  5. SuSu_酥酥22

    Hands-on test of OpenCode anonymous model Exo Free on a convenience store scene

    OpenCode anonymous model Exo Free ran a convenience store rainy night scene test, and the results look a bit stronger than GPT 6.1 Sol. The model ID is opencode/exo-free, completely free for a limited time, with the interface listing 1M context, up to 131K output, support for text+image and high reasoning; data during the free period may be used to improve the model.

  6. X搜索:AI视频开源项目发现18

    List of 30 AI tools that "seem illegal" but are actually legal

    A list of 30 AI tools covering scenarios such as image, video, voice, coding, and writing, including Raphael.ai for free unlimited image generation, ElevenLabs 5-second voice cloning, Suno for generating songs from lyrics, Runway Gen-3 free trial, Kling AI video generation, D-ID for generating digital humans from photos plus audio, and Cursor AI coding.

  7. Oimi · 中文图像视频教程67

    Google releases Nano Banana 2.1: half the price, panoramic aspect ratios, and 14-image fusion

    Google officially released the image generation and editing model Nano Banana 2.1 (model ID: gemini-nano-banana-2.1) in the Gemini API, at about $0.034 per 1K image, roughly half that of Nano Banana 2.

    Why it matters: Google's image model 2.1 cuts the per-image cost by about half and adds panoramic aspect ratios and 14-image fusion, which can be used to evaluate whether to migrate an existing image-generation pipeline.

  8. DesignerBox · 商品图与视频工作流22

    3D product rendering: renders, photos, or AI images? (2026)

    3D product rendering computes images from a digital model without a camera; a single model can output all colorways, any angle, configurators, and AR files. For furniture or home brands with 20 to 500 products, rendering wins on the variant matrix, whereas AI images only produce pictures and no 3D files, requiring manual checks of materials and proportions. The workflow has five steps—modeling, materials, lighting, rendering, and post-production—and requires CAD files, reference photos, and physical color swatches as inputs.

  9. DesignerBox · 商品图与视频工作流22

    AI image generator comparison: test with your own product (2026)

    A method for testing AI image generators with your own products: pick three hard-to-draw products (one with label text, one with a reflective surface, one with a fabric pattern), fix the size and aspect ratio, and run three prompts twice on each model for 18 images total, scoring each image on five criteria—shape proportions, label text and logo, material color, lighting, and whether anything was added from thin air—with a pass requiring all five.

  10. ArchiGen AI · 图像生成与ComfyUI实战29

    How to test the consistency of AI multi-angle renders: the second camera will expose the first

    ArchiGen proposes rendering the same model with two cameras that overlap by about one-third of the frame, comparing 8-12 countable facts such as the number of openings, edge lines, and material boundaries to verify multi-angle consistency. A third-party leaderboard says ReRender AI offers multi-angle rendering, but this is only a restatement of vendor claims, not actual test results. Each camera generates 4 times with all outputs kept, one local repair is allowed, and defects are recorded in four columns: invariant, results from the two cameras, and repair outcome.

  11. Slop Reel · AI 视频制作简报60

    India's advertising regulator requires labels on AI ads, Italy investigates Suno's terms

    India's Advertising Standards Council (ASCI) published AI advertising rules on September 29, effective three months later, requiring labels on synthetic influencers, consented voice or face clones, fictional scenes, and demonstrations of nonexistent products when they could mislead buyers; false testimonials and non-consensual deepfakes are outright banned, with responsibility falling on advertisers.

  12. TechCrunch · AI32

    After abandoning their first product, former Ramp engineers raise $20 million for AI ad generation platform Melius

    Melius, an AI ad generation platform founded by former Ramp engineers, closed a $20 million Series A led by CRV, plus a $5 million seed round led by General Catalyst, for $25 million in total funding. The team abandoned its entire codebase after half a year developing its first performance-marketing tool and pivoted to a "creative work intelligence agent lab" that generates ad creative assets and marketing campaigns; within two months of ending stealth in July, annualized revenue had already exceeded $1 million.

  13. SuSu_酥酥78

    Google Releases Nano Banana 2.1 Image Generation and Editing Model

    Google has officially released the Nano Banana 2.1 image generation and editing model, which the company says comprehensively surpasses the previous generation in visual design, precise mask editing, and subject consistency, with more natural and realistic images. This version has rolled out across Gemini, Search AI Mode, AI Studio, Flow, Stitch, Google Ads, and enterprise platforms.

    Why it matters: Google's image generation model gets an upgrade, calling out changes in three capabilities—visual design, mask editing, and subject consistency—and has rolled out across multiple product lines.

Oct 6Tue
  1. note · life_good_life · 图像视频创作分享22

    2026 AI Image Generation Tool Use-Case Quick Reference: Recommendations by Use Case and 20 Copy-Paste-Ready Prompts

    This 2026 quick reference provides AI image generation tool recommendations by use case: choose Midjourney for visual aesthetics, ChatGPT or Gemini for Japanese text in images or conversational editing, and Adobe Firefly for commercial images where copyright compliance matters.

  2. note · lofty_eagle5366 · 图像视频创作分享12

    Generating Images with 150,000 PixAI Points Saved Up

    The author spent all 150,000 points saved in their PixAI account on image generation in one go, with none of the works using reference images. The generation process is shown in order, with one failed image costing 7,000 points while the daily login reward is 10,000 points. Since relying solely on prompts had limited results, the author spent another 100,000 points training a LoRA model, feeding it 10 semi-realistic images they had on hand, and felt the results improved noticeably.