Tagtelier v0.6.0 update: this desktop app generates prompts arranged in effective order by clicking Japanese categorized tags, and integrates with ComfyUI to complete image generation, video generation, and video editing.
MiniMax H3 RefMod was used in animation production, and in testing the generation quality has reached a level where it can be used fully as animation; both live-action and animated versions can be generated. Making about 55 minutes of video with Comfy-Cloud costs about $80USD per month; enterprise-level costs are about $630USD/month, and operation only requires training an operator. Compared with Google Colaboratory disconnecting every 30 minutes, Comfy-Cloud or the ComfyUI Desktop version is more worry-free and easier to use.
The author introduces RefMod, a new character-fixing technique for MiniMax H3 in ComfyUI; without LoRA training, using 8 to 20 images of the same character, it can generate a 1 to 2MB .safetensors character file within tens of seconds.
Why it matters: RefMod encodes 8 to 20 images into a 1 to 2MB character file, offering creators who do not want to run LoRA training a comparable alternative path.
ComfyUI releases the LivePair creative project, offering three types of nodes: text-to-image, image editing, and video generation. This is the first release; specific models, parameters, and availability information have not yet been disclosed.
The author used Wan 2.2 on Hugging Face, invoked via Claude Code's MCP, to hand large movements in 10 still-image storyboard frames to a video generation AI, while small movements were done by moving parts with a Python program. After inputting both the first and last frames, the backgrounds and lines in the videos generated by Wan 2.2 did not break down, and there was no case of the art style being overwritten like with Kling previously; however, the character still stands up when climbing rocks, the leaf-licking only repeats once, and text instructions only fixed the feet, water bottle, and eyes.
The author, who publishes a note daily using Claude Code subagents, breaks down the time per post: 19 posts were published from 9/22 to 10/10, at about 25 minutes each, of which /daily execution takes 10 minutes, image generation 5 minutes, and note pasting and publishing 10 minutes. In other words, 60% of the time is spent outside /daily, and image generation requires manually pasting the prompts in posts/ into nano banana pro, so it still cannot be fully unmanned.
InvokeAI 7.0.0 alpha 2 is released, adding multi-canvas projects, a MySQL/MariaDB high-performance database backend, GGUF transformers support for krea2/ideogram4/ernie-image, and the Atlas Cloud image service provider. After upgrading the database, this version cannot be rolled back to v6; the official recommendation is to create a new root directory for trial use, and a new reserve_vram_gb setting has been added to mitigate OOM.
The SparkH3 team releases Spark-H3, a block-sparse attention inference backend for MiniMax-H3; its ComfyUI package ComfyUI Spark-H3 v0.1.2 provides precompiled installers for RTX 40 and RTX 50 series on Windows and Linux.
Before generating AI renders, architects should first build a visible reference board in PureRef, assigning each image a single role—massing, material, light, or atmosphere—and grouping them while removing conflicts. PureRef is a desktop reference-image canvas that supports arranging, grouping, annotating, and exporting; it is not itself an architectural AI product. In ComfyUI, input only the needed reference images into IP-Adapter or image conditioning nodes, and record cropping and weights to avoid flattening the whole board into a single collage image.
The alien character Bonnie tested the AI video production platform SkyProduction, using the three features "Smart Canvas," "Storyboard," and "Video" to complete an approximately 40–50 second short film with 8 scenes. Video generation used Seedance 2.5 and image generation used Seedream 5.0; a single image costs 2 credits, and a 5-second video costs 50 credits, outputting 9:16 vertical 720P with audio.
To make a Product Hunt launch video, first identify the customer problem to demonstrate, then record the product solving that problem, and cut irrelevant details. Organize the video into a roughly 50-second story following "problem → product action → result → next step," and use a mood board to unify layout, pacing, and sound direction before motion design.
Figma Weave (formerly Weavy) pricing is: the free plan includes 150 credits; Starter is $24 per month, Professional $45, and Team $60 per user, including 1,500 to 4,500 credits respectively.
ROCm Ninodes released v2.3.14, reverting the h3_tile_size change. The reason is that 512px H3 VAE tiles produce visible seams and overexposure in actual rendering; the model default of 256 remains unchanged.
Design Direction v1.5 is released. It is a local HTML tool — unzip the ZIP file and open index.html in a browser to use it, no installation required. Priced at 4,980 yen.
AtomicChat released an AD-Q4_K quantized version of Qwen-Image-2.1-Turbo, sized at 4.2GB — roughly 70% smaller than the official BF16 version at 14.2GB. The author measured LPIPS to be 21% lower than the standard Q4_K.
A developer created a custom ComfyUI node that allows hand-drawing directly on the node, then feeding the result into ControlNet for composition control. The author notes that prompts like "from side" are difficult to control for character orientation and gun-holding poses; hand-drawing fixes the composition and reduces randomness, and even extreme biased compositions are achievable. The node is publicly available with a workflow, and only requires LineArt-level control capability.
Author Akita Ryosuke published his four-step workflow for using Claude desktop with Claude in Chrome to operate Seedance 2.5 for video generation: first, use ChatGPT to create a character sheet and prepare props and sound materials; then write out clear details for characters, scenes, actions, camera, dialogue pronunciation, negative prompts, aspect ratio, quality, and duration; next, let Claude automatically operate the Yapper platform to generate; and finally, provide feedback for revisions on each item.
Why it matters: The author published a complete four-step workflow for using Claude to operate Seedance 2.5 for video output, including character sheets, lip-sync details, and cost breakdowns — transferable to your own generation workflow.
An author used AI to adapt their note article about "selling things" into a 1-minute-28-second claymation, working from September 28 to October 9 and redoing it 9 times.
The author used MiniMax H3 in ComfyUI to fully locally generate a 2-minute-33-second Inkubus intro video. From conception to uploading to YouTube took about 6 hours, with GPU runtime of approximately 2 hours.
Sitor AI has open-sourced Frametell (帧言) on GitHub, which can turn a sentence, document, PDF, slide deck, or image into a 1080p explainer video with subtitles. In the workflow, a director agent first produces a storyboard for confirmation, then writes narration, dubs it with ElevenLabs (Volcano Engine TTS for Chinese), draws page by page and self-checks, and includes five built-in styles: warm-story, study, chalkboard, whiteboard, and tech-dark.
Why it matters: An open-source, locally run explainer-video generation tool that first produces a storyboard for confirmation and then dubs and renders, so creators can use it to assess whether to add it to their production workflow.
The author ran MiniMax H3 locally with ComfyUI on an RTX 5060 Laptop (8GB VRAM); generating one 576×1024, 10-second clip with audio at 20 steps took about 38 minutes, while the same task took 6 minutes 18 seconds on an RTX 5080.
awesome-claude-video-skills collects 268 open-source projects that let coding agents make videos, divided into 10 categories by use such as frameworks and toolboxes, product promos, knowledge explainers, editing, and vertical short videos. Each project is marked with a SAFE or CAUTION safety rating, audited by Agent Skills Hub, and the list refreshes every 8 hours.
The SunoMV editor breaks a 4-minute, 21-shot song into five tiers of visual cost: using only text PVs and existing storyboard images for motion effects both cost 0 credits, AI motion effects cost about 118 cr/shot, AI video costs about 700 cr for a 13.8-second shot, and shooting the whole film costs about 11,500 cr. Its "Balanced" option only shoots the 5-shot instrumental section and the 6-shot chorus, at about 5,600 cr, using full-screen text PVs for the verses and video overlaid with text PVs for the chorus.
Depth Anything 3 can estimate a usable relative depth map from a single flat architectural render, providing structural guidance for ComfyUI, but it cannot restore precise dimensions or occluded surfaces. The article provides a recovery workflow: keep the original image and the original depth map, connect a depth-compatible ControlNet and an image-to-image path, start from lower denoising and control strength, and first use a false-color preview to check depth boundaries at glass, thin rods, repeated bays, and large flat areas.
The process for making a GitHub README demo GIF is: record a single operation and its result, convert it to a GIF and store it in the repository or upload it as an attachment, then embed it with .
Chaos introduced Smart Selection in Veras 4.5, supporting one-click selection of targets by object, material, category, and custom prompts, driving Modify and Replace operations without mask brushes, and covering the Web version as well as integrations such as Revit, Rhino, SketchUp, and Enscape.
The core difference between Figma Weave (formerly Weavy) and ComfyUI is the contest between a hosted canvas and a local canvas: the former uses credits in the browser to call commercial models such as Google, OpenAI, Kling, and Runway, while the latter is GPL-3.0 open-source software that runs on your own GPU.
Higgsfield's Cinema Studio changes AI video camera movement, camera positions, and shots from pure text prompts to explicit parameters locked in before generation, reducing variation in results from the same prompt across generations.
LTX Desktop 1.3.0 is released, with the entire app switched to a new interface consistent with LTX.io, adding a Home page, sidebar, and quick search, and merging first-run setup into a single Setup screen.
When running MiniMax H3 + Veda Sparse Attention on an RTX 3060 Ti (8GB VRAM) to generate 1376×768 video, a 5.2-second video completed in about 5 minutes, but a 6.6-second video took 31 minutes 30 seconds, a 7.3-second video rose further to 38 minutes 34 seconds, and a 10.1-second video terminated abnormally due to a CUDA memory allocation failure.
Pixaroma releases the music video "Out of Memory," a whole song built around ComfyUI, with all footage recorded in real ComfyUI, including real workflows, live previews, out-of-memory errors, missing nodes, and the process of repairing a corrupted VAE from the logs.
sd-webui-infinite-image-browsing released v1.9.1, adding a unified preview popup for non-media files and adding video duration badges and tag filtering (AND) to fuzzy search. This version also fixes a first-launch error and improves the sidecar file anti-orphaning mechanism and caches the unpacking process.
PROMPICKS is a Windows image-generation assistant app containing 4,874 positive tags and 126 negative tags, 5,000 in total, and supports combining Japanese tags to produce English prompts. It can connect to ComfyUI to execute generation and preview directly, is compatible with checkpoint and diffusion model split structures, and supports up to 3 LoRAs in a standard split workflow. It is priced at 1,500 yen as a portable ZIP version, and the product itself does not include ComfyUI or generation models.
Toonflow 2.0.5 released, fixing the issue where clipboard images could not be pasted onto the canvas and unifying asset height validation errors at the provider layer. This version provides installers for Windows, M-series and Intel Macs, and Android (ARM64, Android 8 and above); the M-series Mac installer has completed Developer ID signing and Apple notarization.
liyupi open-sourced ai-promo-video-maker on GitHub; users only need one sentence describing the product to promote, and the AI completes research, questioning, screenshots, music, voiceover, editing, and verification, outputting a 4K commercial promo video.
Why it matters: This open-source workflow breaks promo video production into specifications and scripts that AI can execute, letting creators assess whether it can replace outsourced processes.
The author recommends two free-to-use AI video generation sites, Vheer AI and Hunyuan Video. Vheer AI's Image to Video feature, when using the Vheer Quality model, was measured to consume 0 credits for generation; Hunyuan Video allows setting the model, prompt, and aspect ratio, and is suitable for making vertical videos for TikTok and YouTube Shorts.
The ComfyUI scene detection project scenedetect fixed a Comfy Registry distribution issue by excluding development-only files from the release package. GitHub Releases and Comfy Registry release notes are now also kept consistent.