Skip to content

Creative Directions

ComfyUI & Workflows Latest

ComfyUI Ecosystem and Reusable Workflows: Nodes, process templates, batch production methods, and creator engineering practices.

199 picksLast 30 days 99 itemsCollected 1,541 items

updated

Pick archive · page 8

Oct 13Tue141–160
  1. ComfyUI Wiki · 工作流与开源创作分享78

    Fish Audio releases Drama 3 speech synthesis model, supporting 16 characters and natural-language performance instructions

    Fish Audio released its next-generation speech synthesis model Drama 3 on October 13, 2026, an open-weight 35B MoE model activating 3B per token.

    Why it matters: The 35B open-source speech model supports same-track generation of 16 characters and natural-language performance instructions; its pronunciation robustness data can be used to assess dubbing and audiobook workflows.

  2. ComfyUI Wiki · 工作流与开源创作分享66

    Qwen-Image 2.1 Lightning LoRA compresses 40-step generation to 4 steps

    LightX2V released a Lightning LoRA for Qwen-Image 2.1 that compresses the model's default 40-step generation to 4 steps, with the sampler changed to 4 steps, CFG disabled (true_cfg_scale=1.0) and four sigma values fixed.

    Why it matters: Qwen-Image 2.1's 4-step Lightning LoRA compresses the default 40 steps to 4 steps and includes multi-reference editing examples, which can be used to assess the speedup potential of image generation pipelines.

Oct 12Mon
  1. Higgsfield 官方博客62

    Higgsfield launches Katana: replicating edits inside Claude based on a reference video

    Higgsfield released Katana, an editing skill that runs inside Claude, calling Higgsfield models through the Higgsfield MCP connector; using a reference video or a preset from the preset library plus up to 5 photos of people, products, or clothing, it reconstructs a new video that preserves the reference clip's shot order, duration, and audio track, with results appearing in the conversation and in Higgsfield Assets.

    Why it matters: Hand the reference clip's shot order, rhythm, and audio-track structure to Katana inside Claude to replicate it, so creators can judge whether manual matching cuts can be skipped.

  2. note · wise_azalea8300 · 图像视频创作分享78

    Comparison of 44 ComfyUI video upscaling and restoration methods, with an all-in-one comparison workflow

    The author compared 44 video upscaling and restoration methods on ComfyUI, all running locally for free, and distinguished between two scenarios: restoration and upscaling. Restoration targets low-quality footage from small cameras, surveillance, and the like, while upscaling targets upgrading HD to FHD, 4K, and 8K and recovering skin texture and eyelash detail. The article releases a large-scale comparison workflow that can run all 44 methods in one go, making it easy to pick suitable methods for different content such as people, landscapes, or animation.

    Why it matters: The author separates the two needs of restoration and upscaling, uses the same footage to compare 44 methods side by side, and releases a ComfyUI workflow that can run in one click.

  3. note · comfylab · 图像视频创作分享75

    Hands-on test of MiniMax H3's RefMod node: keep the same character appearing repeatedly without training a LoRA

    The author tested MiniMax H3's RefMod custom node in ComfyUI. It compresses character images, motion videos, or audio into reusable reference files, requires no LoRA training, is used much like a LoRA loader, allows adjusting strength, and can combine multiple references to put characters in the same frame.

    Why it matters: The author tested MiniMax H3's RefMod custom node and provides the complete parameters and prompt-writing approach, from installation and building a reference library to integrating with the official T2V workflow.

Oct 11Sun
  1. z-ab · 开源短剧与创作项目分享74

    Lanshu open-sources Agent Skill lanshu-create-ai-presenter-video: a digital human video from one authorized photo, with 9 built-in styles

    Lanshu open-sourced Agent Skill lanshu-create-ai-presenter-video; in Claude Code, Codex, Gemini CLI, Cursor, and OpenCode, a single sentence runs the entire process of script, voiceover, digital human generation, lip-sync, subtitles, editing, rendering, and quality check. The repository has 2.6k Stars, 404 Forks, and an MIT license.

    Why it matters: An Agent Skill that breaks AI video production into computable states and paid gates; its workflow design approach can be transferred to other content pipelines.

  2. note · synth_brain · 图像视频创作分享74

    MiniMax H3's new character-fixing technique RefMod: generate a character file in tens of seconds without training

    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.

Oct 10Sat
  1. note · ryosuke_blog_ · 图像视频创作分享65

    Complete Workflow for Using Claude to Operate Seedance 2.5 for Video Generation

    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.

  2. GitHub:AI视频开源项目发现71

    Sitor AI open-sources Frametell: generate a 1080p explainer video from one sentence or one document

    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.

Oct 9Fri
  1. GitHub:AI视频开源项目发现77

    liyupi open-sources ai-promo-video-maker: generate a 4K product promo video from one sentence

    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.

  2. Kapwing 创作指南62

    How to recreate the Raindance AI two-person music video trend with Seedance 2.5 (with prompts)

    The Raindance AI trend originates from the music video for "Raindance" by Dave and Tems, in which creators use AI to replace the two performers in a dock scene with themselves or any character.

    Why it matters: Using two identity reference images plus one line-art motion reference, the Raindance two-person music video scene can be recreated in Seedance 2.5, and the approach can be transferred to other motion-reference trends.

  3. Comfy 官方博客74

    How I generate real-time video on a single GPU with MiniMax H3

    On a single RTX 5090 (32GB VRAM), the author used FastVideo's FastH3 V2 checkpoint with four-step sampling to produce 15-second 448×256 real-time video, aiming to bring down the cost of continuous generation from 6k per day.

    Why it matters: The author breaks down one by one the six optimizations for running real-time video on a single RTX 5090, which can be transferred to your own production workflow.

Oct 8Thu
  1. Stuubzzz · 视频工作流实测71

    MiniMax H3 Four-Step Output: PDMD LoRA vs. NVIDIA SoL Refiner Benchmarked on RTX 5090

    The author benchmarked MiniMax H3's PDMD 4-step LoRA and NVIDIA SoL Refiner on an RTX 5090: PDMD averaged 101 seconds to output a 6.6-second clip at 2520×1440, while the 20-step Quality path took 753 seconds.

    Why it matters: The author benchmarked PDMD 4-step LoRA and NVIDIA SoL Refiner on an RTX 5090, providing itemized timing and image quality differences that can serve as a selection reference.

  2. Reallusion60

    Reallusion and Escape.AI Reach Strategic Partnership, Connecting 3D-to-AI Video Pipeline with CTV Distribution

    Reallusion and Escape.AI announced a multi-year strategic partnership, integrating Reallusion's 3D software and asset system into the Escape.AI creator ecosystem to form a 3D-to-AI video pipeline, using 3D spatial data as a control layer to address issues such as visual inconsistency, high token costs, and unclear IP ownership.

    Why it matters: The combination of 3D control layers and CTV distribution channels provides AI film and television creators with a reference pipeline from production to monetization.

  3. Luma News60

    Luma and Claude Motion Integration: Animations Can Now Be Output as Video Directly in Luma

    Luma announced that Claude Motion animations can now be opened directly in Luma. Creators can import Claude-generated animations into Luma via MCP connection, then continue expanding, refining, and exporting them as finished videos in any format.

    Why it matters: Claude Motion animations can now enter Luma directly via MCP connection, allowing creators to assess whether a cross-tool video production pipeline is worth integrating.

Oct 7Wed
  1. Kapwing 创作指南60

    How to make Jean Philanthrope-style AI videos

    Kapwing published a tutorial explaining how to replicate Jean Philanthrope-style AI character videos in Seedance 2.5 using character reference sheets and motion reference videos. The process has four steps: first generate a multi-view character reference sheet containing front, three-quarter, and side views with a prompt; then convert an existing video into black-and-white line art as motion reference; then combine the two in Seedance 2.5 to generate new scenes; finally edit in Kapwing.

    Why it matters: Break down the reusable workflow for Jean Philanthrope-style AI characters, using character reference sheets plus motion references to maintain character consistency in Seedance 2.5.

  2. GitHub:AI视频开源项目发现73

    Genjustsu open-source workflow: replace video subjects using depth and SAM 3 masks

    sirioberati released the Genjustsu open-source workflow, which replaces video subjects in a local UI using color depth, SAM 3 masks, and Enhancor's Seedance integration, producing a draft first and generating 1080p after approval.

    Why it matters: The open-source local workflow replaces video subjects using color depth and SAM 3 masks while preserving the original audio track, so its reuse value can be evaluated.

  3. CGPress67

    ComfyUI Launches AI Agent Comfy Agent

    Comfy has launched Comfy Agent, an AI assistant for its open-source node-based generative AI platform that can plan, build, and fix ComfyUI workflows on the canvas. After users describe what they want to make, the agent connects nodes, wires up models, and debugs the parts that go wrong; it currently runs on Comfy Cloud, with a local version coming soon.

    Why it matters: ComfyUI brings an agent onto the canvas that can automatically build nodes, connect models, and troubleshoot, so creators can judge whether the way they build workflows needs to change.