Skip to content
Date and source

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

Oct 30Fri
  1. Hugging Face Daily Papers29

    SpecSlot: Read, Allocate, Route — Role-Coupled Tokenization for Pixel-Space Generative Transformers

    SpecSlot is a pixel-space rectified flow Transformer that decouples output detail from global attention sequence length via role-coupled tokenization. On ImageNet-512, SpecSlot improves FID by 2.84 over the baseline while reducing FLOPs by 3.9%; in text-to-image generation, the DPG-Bench score exceeds JiT-B/8 by 1.06, with forward cost reduced by 40.6%.

  2. Reddit comfyui · 创作与工作流分享65

    Phaino: An Open-Source AI Image and Video Studio Based on ComfyUI

    NoobGamerZaid has released Phaino, an open-source AI image and video studio based on ComfyUI. It supports open-source models such as FLUX, Wan 2.1/2.2, and HunyuanVideo, offering one-click camera movement, character replacement, video tools, and a built-in timeline editor. Every result is an editable ComfyUI workflow that can connect to Comfy Cloud or local ComfyUI.

    Why it matters: The open-source studio encapsulates ComfyUI workflows into editable complete pipelines, enabling creators to evaluate localized AI film and video production and output efficiency, and reuse its model integration methods.

  3. note · miyuu_2026 · 图像视频创作分享27

    10/30 MOSS-Image int8 Trial Experience

    MOSS-Image int8 generates 1024x1024px images on ComfyUI. With default settings (prompt enhancer enabled), it takes 426 seconds; after disabling prompt enhancer, it drops to 122 seconds, and after adjusting the age parameter (15-19 years old), it takes about 117-121 seconds. The generated images have high realism, and compared to Krea2, the style is more rugged.

  4. Reddit StableDiffusion · 创作者与工作流分享57

    What does a character LoRA actually change: the same girl in the same scene vs. a new face every run

    The author compares the generation results without LoRA and with a character LoRA: without LoRA, each run yields a different girl, while the character LoRA maintains the same face, curly hair, and body across different scenes. The author suggests placing 20 thumbnails in a phone-sized grid to judge consistency as a whole, and points out that hair and body can also reveal inconsistencies.

  5. pixaroma · YouTube 创作视频62

    ComfyUI Prompt Parsing: CLIP Text Encode, Negative Prompts, and Dynamic Prompts

    pixaroma released a ComfyUI node explanation video, explaining how CLIP Text Encode converts prompts into model instructions, the working mechanism of positive and negative prompts, and why negative prompts fail under CFG 1. The video also explains that prompt weights are ignored on Z-Image and Flux 2 Klein, and demonstrates two techniques: random selection {red|blue|green} and comments, using Z-Image Turbo throughout.

Oct 29Thu
  1. Reddit StableDiffusion · 创作者与工作流分享73

    Krea 2 Turbo Distill 3-Step LoRA: 8 Steps Reduced to 3 Steps

    The creator released a 3-step distilled LoRA for Krea 2 Turbo, reducing denoising steps from 8 to 3, with denoising speed improved 2.6x and detail quality reaching 1.15x that of the teacher model. Training used 12,000 samples, about 41 hours on a single H100; a ComfyUI workflow is provided, compatible with diffusers, ComfyUI, and MLX.

    Why it matters: This 3-step distilled LoRA reduces Krea 2 Turbo's denoising steps from 8 to 3, with detail quality exceeding the teacher model, giving creators a faster image generation option.

  2. note · comfylab · 图像视频创作分享58

    [ComfyUI] Hands-on test of Fibo 2.0 DEV and Fibo 2.0 Edit Turbo

    The author tested Bria AI's Fibo 2.0 Dev and Edit Turbo, verifying their VGL structured prompts and PE prompt enhancement in ComfyUI. Edit Turbo completed editing in about 4.7 seconds after PE was fixed on an RTX 4070 Ti SUPER, while Dev took about 92 seconds; the author finds Qwen Image 2.1 more convenient for editing.

  3. Reddit comfyui · 创作与工作流分享63

    Director's Cut: A Film and Video Production Application Based on ComfyUI

    The author open-sourced Director's Cut, a film and video production application based on ComfyUI that supports preset character creation and a video studio. The application uses Qwen-Image 2.1 to generate and edit images, MiniMax H3 to generate video with optional OmniRef, and also integrates an MCP server for AI assistants to call.

  4. Reddit StableDiffusion · 创作者与工作流分享19

    Training a LoRA character with tattoos on MiniMax H3

    The author trained a LoRA character with tattoos on MiniMax H3 for a week without success; the tattoo positions were correct but the results were chaotic and blurry. They tried regular H3 and ref model, linear rank 8-128, learning rate 0.0003, resolutions 256-1280, 30 high-resolution close-up photos plus 3 videos, training up to 20,000 steps, but still couldn't get the model to remember the tattoos.

  5. Reddit StableDiffusion · 创作者与工作流分享56

    Retro Anime Style LoRA for MOSS

    The author trained a retro anime style LoRA, recreating the cel texture, film grain, and hand-painted background style of 80s/90s OVAs. Recommended weight is 0.7-1.0, trigger word is retro anime style illustration, and it can be downloaded at https://civitai.red/models/1970260/retro-anime-style.

  6. Reddit StableDiffusion · 创作者与工作流分享27

    Ultimate Digital Artwork v2 - MOSS lora

    MOSS lora v2 released. The author cleaned low-quality images from the dataset and re-annotated all images, no longer using converted Ideogram prompts. The new model shows significant improvements in color depth, prompt adherence, and high-detail image generation. Due to performance improvements, the model is less adventurous than v1, and prompts need to be more specific.

Oct 28Wed
  1. note · kongo_jun · 图像视频创作分享41

    ComfyUI tests MOSS-Image's MAX ULTRA hires LoRA

    ComfyUI tests MOSS-Image's MAX ULTRA hires LoRA for high-resolution image generation, which can prevent generating too many people when no LoRA is used. On an RTX 4090, generation took 140 seconds and used 20GB of VRAM. The author thinks the generation speed is slow and needs to be paired with a speed-up LoRA, but that would reduce text and detail quality.

  2. Reddit StableDiffusion · 创作者与工作流分享32

    AuroraIMG2-50M-Preview released!

    AuroraIMG2-50M-Preview has been released. It is a text-to-image model with a denoiser of only 50M parameters and approximately 90M end-to-end inference parameters. On a 2016 Intel i5 7200U laptop CPU, generating with 50 steps takes 27 seconds, and with 20 steps takes 11 seconds. It is released under the Apache 2.0 license.

  3. note · tsugiito_design · 图像视频创作分享60

    Creating a Grim Reaper Shouldering a Large Scythe with ComfyUI: Tags and JSON for Giant Blade, Red Eyes, and Unruly Smile

    The author shares a complete workflow for creating a Grim Reaper character shouldering a large scythe using ComfyUI, using the Animagine XL 4.0 Opt model, preserving the satisfactory pose through inpainting, only fixing the face and clothing, and includes 6 JSON files and input images. The article compares the effects of different tags on pose, expression, and lighting, and provides detailed steps for partial corrections.

  4. Clip设计夹 · 创作教程与项目分享24

    No budget for marketing? RoboNeo uses AI to cover IP short dramas, viral remakes, and emoji packs.

    RoboNeo uses AI to help small and medium-sized businesses do IP marketing at low cost. A single sentence can generate an exclusive IP character, which can be used for short dramas, viral remakes, and emoji pack production. Users need no design or editing experience; through a four-step process, they can complete everything from IP creation to content output. With one investment, they get a full team of "IP designer + screenwriter + animator + editor", and all works automatically become brand assets.

  5. note · akb428 · 图像视频创作分享51

    Internet café real-world test: RTX 5060 running ComfyUI image generation — how fast are Anima and Qwen-Image 2.1?

    The author tested RTX 5060 (8GB VRAM) running ComfyUI image generation: Anima took 57 seconds per image, while Qwen-Image 2.1 reference generation took 47 seconds per image. The local LLM (Ollama's gemma4:12b-it-qat) managed only 14 tok/s, becoming the speed bottleneck when used together. The author also tested their own novel-generation tool Inkubus: about 14 minutes for 10,000 characters, but the 12b model's quality is limited.

  6. ComfyUI Wiki · 工作流与开源创作分享63

    Krea 2 Turbo 3-step distilled LoRA released.

    lvladikov released Krea2-Turbo-Distill-3step-LoRA, compressing Krea 2 Turbo from 8 steps to 3 steps, about 2.6x speedup (from 65.4 seconds to 24.9 seconds), with fine texture reaching 1.11x that of the 8-step teacher. This rank-64 LoRA can be loaded directly in ComfyUI, diffusers, and MLX, without custom nodes, and comes with a pure stock node workflow.

  7. かみもと · 本地AI视频工作流56

    MOSS-Image generates manga single pages and spreads, sharing prompt techniques for maintaining character consistency

    The author used the local image generation model MOSS-Image to generate black-and-white manga single pages and spreads, and shared three prompt techniques for maintaining character consistency: fixing the character description text, declaring the characters overall, and embedding the same description in each panel. As a result, the two characters remained distinct to the end, and the large spread panel was also successful; the text was mostly accurate, but complex Chinese characters were occasionally miswritten, and unspecified sound effects were added.

  8. Reddit comfyui · 创作与工作流分享68

    PrunaAI releases a 6-step LoRA adapter for Qwen-Image-2.1, without requiring an upsampling step.

    PrunaAI releases a 6-step LoRA adapter for Qwen-Image-2.1, compressing image generation from 40 steps to 6 steps. The 6-steps-upsampled version removes the separate prompt upsampling step, saving 41 to 80 seconds depending on prompt style. The LoRA is loaded directly onto Qwen-Image-2.1, and the pipeline remains unchanged.

    Why it matters: The new adapter compresses image generation from 40 steps to 6 steps and removes the upsampling step, allowing creators to assess whether to replace their existing generation workflow.

Oct 27Tue
  1. note · quirky_hippo6791 · 图像视频创作分享60

    ComfyUI Verification Round 3: Claude prompt turns a yellow streetcar into a yellow creature

    The author used Claude to write prompts and ComfyUI to generate images, verifying 12 landscape themes over 3 days. They found that mixing in person-related descriptions caused monsters and hands to appear in the images. After adjustments, monsters decreased, but the "yellow streetcar" was generated as a yellow creature, showing that the association between color and object is difficult to control. Based on this, the author summarized seven inspection rules and compared them with the ChatGPT version.

  2. 可灵官方博客15

    ChatGPT review: Pros, cons, and real-world performance.

    ChatGPT excels at drafting, research, coding, file analysis, and image creation, but complex visual projects may require specialized tools. The review compared ChatGPT Images 2.5's conversational generation and editing with Kling IMAGE 3.0 Omni (supports 10 reference images, Image Series, native 2K/4K output) and Kling 4.0 (Omni Reference supports 15 reference assets, 3-30 second video generation).

  3. Runway 官方新闻74

    Runway launches App Studio, enabling no-code creation of custom creative apps.

    Runway launches App Studio, allowing users to describe their needs in natural language and build custom creative apps with standalone interfaces without code, generating images, videos, and audio. App Studio includes built-in apps such as Character Swap, AutoComp, and Product Portfolio, built on Brand Kits, Workflows, and Assets. It is included in all paid plans and consumes existing Runway credits.

    Why it matters: Runway's App Studio brings no-code app building into creative work, allowing teams to turn years of accumulated workflows and brand standards into tools anyone can use, reducing reliance on individual experts.

  4. Reddit comfyui · 创作与工作流分享67

    Moss Film Looks free workflow: select film and lens, enter your idea to generate.

    The author released the free MOSS-Image workflow, which can simulate film styles such as Kodachrome, Portra, CineStill 800T, Tri-X, and Aerochrome, and supports lens selection from 8mm fisheye to XPan panoramic.

    Why it matters: The free MIT-licensed workflow integrates film simulation, lens selection, and local LLM prompt generation into a single control panel, allowing creators to directly reuse or reference its design.

Oct 26Mon
  1. 可灵官方博客30

    Kling IMAGE 3.0 Omni Funny AI Image Creative and Prompt Guide

    The official Kling blog released a guide to creating funny AI images, offering 12 creative directions and 6 customizable prompts based on Kling IMAGE 3.0 Omni, covering scenarios such as animals at work, anthropomorphic food, and historical figures traveling to the modern era. The model supports up to 10 reference images, native 2K/4K output, and can maintain consistent style across a series of images via Image Series.

  2. note · cocoredo_2509 · 图像视频创作分享47

    Will Canva Craft change the production workflow? 8 things worth verifying before introducing composition-planning AI

    Canva released and launched its self-developed image generation model Canva Craft on October 22, using a two-stage structure that first plans the composition and then renders the image, supporting up to 2K resolution and more than 30 aspect ratios. The article proposes 8 items to verify before introducing it into production, including text accuracy, composition reproducibility, Magic Layers editing capability, consistency of people and small objects, stability of local edits, delivery dimensions, rights, and cost. The official documentation does not guarantee that a specific character will generate the same face every time, so actual output must be used to confirm.

  3. Stable Diffusion Art 教程78

    Multi-image editing, text-to-image, and native transparent background workflows for Qwen Image 2.1 in ComfyUI

    The author tested Alibaba Qwen team's Qwen Image 2.1 in ComfyUI, using a single 7B model to perform text-to-image generation, multi-image editing with up to 10 reference images, and native RGBA transparent cutouts.

    Why it matters: The author tested Qwen Image 2.1's multi-image editing and alpha channel output in ComfyUI, and provides directly reusable prompt formulas and parameter settings.

  4. note · myloose · 图像视频创作分享22

    ChatGPT keeps failing to generate monolid eyes? Author analyzes how conversational AI "betrays" prompts

    The author found that ChatGPT repeatedly failed to generate portraits of men with monolid eyes, because the conversational AI compressed the detailed "monolid eyes" condition emphasized by the user into the single word "monolid eyes," and also added unspecified clothing and performance details on its own. Switching to a model with higher reasoning ability improved the situation, but the image generation stage may still ignore this feature; the author recommends checking the instructions the conversational AI actually passes to the image generation AI.

  5. note · quirky_hippo6791 · 图像视频创作分享28

    ComfyUI hands-on test: letting Claude automatically generate 18 watercolor images overnight, auburn hair turns white

    The author used Claude to build an automated workflow for ComfyUI, automatically queuing up 18 watercolor test images at 1:30 a.m. on a 16GB Mac, all of which were completed by 7:41 a.m. The specified auburn-haired character, however, appeared with white and silver hair in large numbers in the tag version, and this remained the case even after weakening the LoRA; the cause has not yet been confirmed. The author ultimately adopted the tag version, wrote auburn hair into the character setting sheet, and added white hair and silver hair to the negative items, to be verified in the next generation.

  6. Aitrepreneur · YouTube 创作视频58

    Aitrepreneur tests the open-source model MOSS Image and releases a ComfyUI workflow

    Aitrepreneur tested the open-source 5B-parameter image model MOSS Image, saying it can run locally in ComfyUI, generate photorealistic photos and fantasy illustrations, and outperform Krea 2 in several respects. The author also provides a companion workflow supporting standard and Turbo generation, prompt enhancement, and LoRA experiments.