Video Automation v0.2.14 has been released, fixing security issues with certificate dependencies and desktop builds. This version was completed by @1076184145 in commit #21; see v0.2.13...v0.2.14 for the full changelog.
Using a fictional umbrella shop "Shizukudo"'s 20-second vertical video as an example, the article explains that before generating asset images shot by shot, you should first use an image generation AI to create reference images of the same product from various orientations, such as front and side, and then attach the reference image to generate each asset image one by one. The author recommends comparing each asset image side by side with the reference image and the first shot's asset image, correcting inconsistencies item by item in text, and confirming there are no frames, numbers, or explanatory text before handing it over to the video generation AI.
To address the blurry output issue of MOSS Image, a two-stage denoising process can be used: first denoise normally for 5 steps, then apply LatentOperationSharpen to the latent image, and then continue denoising. The author used the base model plus LoRA and showed comparison results of stacking the SFT model as a LoRA onto the base model during the second-stage denoising, as well as comparison results of only sharpening the latents.
Based on the Hugging Face file list and official documentation, the author outlines the purpose of each file in the ComfyUI repackaging of MOSS-Image: the 9GB version (10.16 GB on HF, 9.47 GiB on disk) is bf16, and the 5GB version (5.41 GB) is int8 ConvRot quantization; Base and SFT each have two versions.
François Chatelain independently completed the AI hip-hop MV "Lorem Ipsum Rap" under the name F&C ai Lab, turning the design-world placeholder text "Lorem Ipsum / Dolor Sit Amet" into the song's core hook.
ComfyUI_MiniMaxH3_Director is a MiniMax H3 video generation workflow component for ComfyUI that can be imported and used directly without manually building the node structure.
Blackmagic Design raised prices on 36 products citing memory costs, with the ATEM 4 M/E Constellation IP Plus going from about $13,000 to $20,000, while lowering the price of DaVinci Resolve Studio 21 and several editing panels by 20% through November 20.
Tencent recently launched Lüpaopao, a short drama mini program embedded in the WeChat ecosystem; WeChat's short drama content center consolidated previously scattered short drama mini programs, adopting a free-plus-ads model to target Hongguo Short Drama. The user base of WeChat short drama mini programs had grown to 508 million by May this year; in Q4, Lüpaopao offers a 7-day 100% revenue incentive for high-quality new dramas with per-capita watch time of 10 minutes within 7 days of release, including a 90% cash share and 10% ad credits. The AI simulated live-action short drama "Yinghou Yaya" surpassed 100 million views within 3 days of launching on Lüpaopao, with dual-platform ad revenue reaching several million yuan.
Why it matters: From the paid era of WeChat mini program short dramas to Lüpaopao's free approach, this outlines Tencent's defensive logic and incentive policies in targeting Hongguo.
AI Canvas v0.10.14 adds ComfyUI RefMod character reference pack management, which can associate character cards and bind them to H3 video workflow loading slots; video generation can directly reference video and audio files in the resource library. Custom model APIs support trial runs, stopping and waiting, and resuming asynchronous task checks, and protect unsaved drafts when switching models or protocols. After a ComfyUI result download or save failure, the original task is retained so it can be retrieved later, reducing duplicate submissions.
The open-source creative tool released v1.6.9.1, optimizing the canvas empty-node ratio toggle animation, floating layer drag positioning, and connection hit detection, and making quick creation results display stably by resource identity to avoid video reloads and image flickering. In the canvas Agent, Pi can perform one limited continuation run for unfinished plans, and video polling continues querying the original task when the connection is interrupted. Added Cheaper Inference Chat and MaxForAI video declarative protocol plugins; the database schema is upgraded to 59.
In the third installment of her AI image creation series, the author uses ChatGPT to help learn the relationships among the ribcage, pelvis, spine, and shoulder-arm joints, then writes prompts accordingly to generate two full-body action images: a roundhouse kick and a stepping-forward pose. She specifies the ribcage and pelvis as separate three-dimensional forms with their own directions and lets the pelvis rotate naturally with the kick direction, making the connection between the hip joints and knees more natural than before.
A user asked the AI "Astra-sama" to generate a self-portrait based on its own description, repeatedly emphasizing that it wanted monolid eyes (single eyelid); although the AI was not explicitly instructed, it proactively drew a monolid robot brother. The actual output, however, clearly had double eyelids, which did not match the "faintly" expectation; the author complains that continuing to play this way will burn through their Work quota.
Anthropic released the AI assistant Claude Duet, based on Claude Opus 5.5, which can proactively advance tasks in the background; it is currently available as a research preview to Pro and Max users in the United States via a waitlist.
The creator used OpenAI ImageGen to generate original images, ComfyUI + WAN to generate video, ACE-Step to create music, and RIFE / FlowFrames for frame interpolation and FFmpeg for final compositing to produce a complete music video titled Dancing in the Dark. All video generation, music production, and rendering were done locally on an AMD RX 7900 XTX.
A user shared their workflow in the ComfyUI community and explained that simply dragging and dropping the second image allows direct reuse of the workflow. The post did not disclose the models used, parameters, or specific generation results.
Kokon 0.10 adds a Prompt+ feature to the Qwen-Image 2.1 desktop editor, using the model's already-loaded Qwen3-VL 8B text encoder to expand brief ideas into detailed prompts without downloading a second model; it is off by default, and the system prompt can be edited in settings.
Twitch emotes should be designed at the smallest chat size of 28×28 and uploaded as square PNGs with transparent backgrounds; static images are limited to 1 MB, and animated images must not exceed 60 frames. Recraft Studio can first generate the initial emote using vector output, then use V4 Styles with up to 10 reference images to lock in a consistent style, and remove the background and export 112/56/28 sizes in one click. Custom emotes are available only to Twitch Partners and Affiliates and must pass platform review.
The Jiangsu-produced short drama on the Long March theme, "My Comrade-in-Arms Is a Soldier from 1935," launched on Migu Yuekan and Hongguo Short Drama for network-wide broadcast on October 22, jointly produced by Wuxi Guangxin Film & Television Animation Technology Co., Ltd. and Migu Digital Media Co., Ltd. The series tells the story of modern young man Zhang Xiaofan, who accidentally travels back to the Long March route in 1935, experiences rear-guard blocking actions and the life-and-death test of crossing Liupan Mountain together with a Red Army detachment, and gradually grows into a revolutionary comrade-in-arms.
An AI beauty image prompt collection for tools such as ChatGPT and Gemini, containing about 100 JSON-format prompts that can be copied directly, with themes combining spectacular scenery from around the world with adult women. Each theme provides a 1:1 basic version, a 9:16 vertical version, a 1280×670 note image-output version, and background and atmosphere adaptation versions, along with negative_prompt and aspect_ratio fields.
Nanami Ogawa released AI short film No. 4, "Morning Time Edition," generated using ComfyUI with the Wan2.2 14B model, at a resolution of 1024x576px and 73 frames, composed of 6 shots of 3 seconds each for a total of 18 seconds. The production environment was Windows 11 Home, an AMD Ryzen 7 7700 processor, an NVIDIA GeForce RTX 5060 Ti (8 GB) graphics card, and 32GB of memory.
Qwen-image3 currently has no local version and must be called via API. The author topped up 5 USD on ComfyUI to obtain 1,055 credits for img2img testing, with each generation consuming 10 credits.
PROMPICKS, a Windows application for ComfyUI, has been updated to 1.2, adding prompt weight adjustment (0.1–10.0), restoring settings from generation history, one-by-one continuous generation and aborting, and raising the per-generation limit from 4 images to 32. The new version also supports saving tag combinations, pinning tags, and automatically saving selection states, and retains 5,000 standard tags, priced at 1,500 yen.
AIArtEngine released version v8.4.2, with a full changelog covering the changes from v8.4.1 to v8.4.2. The original text did not disclose specific features, parameters, or performance data.
The scenedetect project for ComfyUI has updated its recommended video node, which can now choose OpenCV or PyAV decoding and selects representative frames based on decoding timestamps, improving detection reliability for variable frame rate videos and short scenes. The PySceneDetect compatibility requirement has been raised to version 0.7.2 or higher, and PyAV can be optionally installed when not built into ComfyUI. A usage guide and example workflow for using VHS images with native video nodes have also been added.
The author used the ChatGPT image generation model to create a Halloween-themed AI boyfriend CG and provided the complete prompt for "Halloween Lover," requiring a 3:2 horizontal composition and a slightly high-angle overhead view, with the male character retaining the facial features and physique of the reference image but with eye color changed to deep red, and a pumpkin-headed female monster on the right, with beauty and eeriness in a 7:3 ratio. The prompt needs to be used together with the character's "face close-up, standing illustration, three-view" trio to maintain consistency in clothing, hairstyle, and appearance when changing poses and compositions.
Krea 3's Ultra mode was merged into the ComfyUI template on October 22, and the author tested the model's editing capabilities via the ComfyUI API. Krea 3 supports native 2K editing, region editing that redraws only changed areas, compositing of up to 8 reference images, HEX color value specification, and layout text preservation; Ultra has a fixed 2K output and takes several minutes per image.
The author optimized MOSS text-to-image in ComfyUI using the Spectrum node with an 8-step LoRA, testing 2-megapixel image output on an RTX3060 with 6GB VRAM and 16GB memory.
When using Z_image_turbo_bf16 to generate full-body portraits, the user found that facial close-up details were excellent, but the eyes appeared distorted and asymmetrical. After trying the two upscaling models RealESRGAN_x4 and 4x-ultrasharp, the results remained the same, and the positive prompt already included "Detailed eyes, sharp focus, detailed iris." The user asks whether there is a way to improve eye quality in full-body images.
A Japanese creator has released a set of prompts for reproducing the full process of "an office lady walking into a hotel after work and changing into Halloween cosplay" using Midjourney, ChatGPT, Gemini, and other AI generation tools, broken down from STEP1 to STEP5: walking out of the office building wearing an employee badge, walking on a night street, taking off shoes in the hotel entrance, and cosplay poses on the sofa and bed.
The author used Animagine XL 4.0 Opt to separately generate the character and raindrop assets, then in ComfyUI used a mask of the transparent area of the window frame to composite the character and raindrops onto the inside of the glass, preserving the front-to-back relationship of the window frame being in front of the character.
Creator Mizuki shared their experience generating original character introduction posters with GPT, Grok, Meta AI, and Gemini; detailed step-by-step instructions repeatedly failed to meet expectations, and the final poster-style result was achieved through self-adjustment. Their daily generation environment is an RTX 3060 Laptop GPU (6GB VRAM) and 32GB memory, currently using a ComfyUI workflow built with ChatGPT and Grok, and they hope for a tool that can automatically generate high-quality video workflows suited to the local environment.
In the 2026 AI director software ranking, ScreenWeaver ranks first, followed by LTX Studio and Saga. The ranking evaluates six tools covering at least two production stages based on criteria such as "whether generated shots remain linked to the script text." ScreenWeaver integrates 7 video models including Seedance 2.5, Veo 3.1, and Kling 3.0; the Auteur version costs 39.99 USD/month, but the tool does not include full editing.
Moss Intelligence's open-source 5B-parameter text-to-image model MOSS-Image gained 1.3 million downloads in overseas communities within 4 days of launch, received 1.4k Stars on GitHub, and topped the Hugging Face text-to-image trending list.
Why it matters: The 5B-parameter open-source model gained 1.3 million downloads in 4 days of launch; the community has already produced accelerated and quantized versions and LoRAs, allowing observation of its actual performance in text rendering and layout control.
LTX production technical director Mohamed Oumoumad and Comfy's Doug Hogan explained in a livestream the professional VFX workflow LTX built for ComfyUI, aimed at real post-production pipelines and involving 16-bit EXR, ACES color, and delivery resolution output.
In the latest Office Hours, Midjourney revealed that the new Niji model is nearly complete and may be made public in about a week at the earliest, but details of new features have not yet been announced. V9 is being developed with the goals of improving overall image consistency and prompt understanding; the initial model has entered testing, with no release date set. The new Upscaler is being tested and may launch before V9 if quality meets standards; Thinking Mode involves a trade-off between accuracy and aesthetics, and the official team is considering enabling it on demand or having the model decide automatically.
Pictory 3's AI video agent can turn requirements such as topic, audience, purpose and duration into a script, scenes and a preview, and generate the final video after the user confirms. In marketing scenarios, HubSpot's Data Agent and Prospecting Agent handle customer research and buying-signal monitoring, Salesforce Agentforce generates marketing briefs and audience segments, and Sprinklr's agent monitors more than 30 channels and sets approval and circuit-breaker mechanisms.
Pictory's official blog compared how generative AI and agentic AI divide the work in video creation: the former produces individual assets such as scripts, images or clips according to prompts, while the latter plans and executes multiple steps in a goal-oriented way. The article uses Pictory 3's AI video agent as an example; it can complete five settings—voiceover, visual sources, aspect ratio, layout presets and brand kit—and generate a script, scenes and a preview for human review. Most workflows use both together, with humans responsible for review and judgment.
Pictory releases Pictory 3, adding an AI video agent: users describe their video requirements and answer five setup questions, after which the agent generates a script, scenes and a preview, and then generates the final video after revisions via conversation or the editor. Videos can start from an idea, script, webpage, document, PowerPoint or recording, with documents and URLs combined limited to at most 5 sources each time. New users will be moved into Pictory 3 with the rollout, while old projects remain in Projects and will not be converted automatically.
Pixolm used Nano Banana 2.1 to generate 6 vertical 9:16 Halloween wallpapers, including pottery monsters, paper pumpkin hills, a night-garden cat, a small dessert shop, pumpkins on a dark background and a watercolor witch's desk. The prompts uniformly required low-contrast clearance in the top 40%, the subject in the lower half and centered, and no text, fake clocks or phone frames drawn. Generation settings were 2K, Standard, 1 image per run, with no reference images or prompt optimization used.