Innovate Futures @ Benji posts
Posts: 709
Files: 875
ComfyUI New Caching For Fast Video Generation - Really Worth To Try?
Tutorial Video : https://youtu.be/FLVcsF2tiXwNew Caching For Fast Video Generation - Really Worth To Use?In this video, we explore the new MagCache (Fast Video Generation with Magnitude-Aware Cache) for fast video generation in ComfyUI, comparing its performance with the older TeaCache method. We test it with models like WAN 2.1, Hunyuan Video, and Flux, examining speed, quality, and practical use cases. Is MagCache truly faster, or does it sacrifice quality for speed? Watch as we dive into real-world tests, workflow setups, and performance ben...
Hunyuan Video Avatar Local Installation Tutorial - Easy Setup!
Video : https://youtu.be/kqRyEnNFqZgExplore Tencent's Hanyuan Video Avatar—a cutting-edge AI model that brings virtual characters to life with realistic facial expressions and multi-speaker detection! Unlike Stable Video Diffusion-based avatars, Hanyuan leverages a diffusion transformer architecture for smoother animations and emotion-rich performances. Learn how to run it locally (even on low-VRAM setups) using the WAN2GP project, with optimizations for as little as 6GB VRAM.Who Is This Content Suitable For?AI developers, content creator...
Wan Vace(Native) V2V With Extend Long Video Workflow - Version 20250616
Finally!from Vace Start End Frames, got inspired with this.Like some recent Vace workflow, it's native. but we need the WanVideoWrapper node pack, because of the WanVideo VACE Start To End FrameThis node is awesome! Or should say Wan VACE architecture is awesome.From now we can stitching long video in ONE generation process.The workflow applied 4x129 frames but you can apply more if you want to.Each sampling pass few frames for next batch of frames to start on another next 129 frames.I apply this Last Overlap frames concept , similar like S...
(For Patreon Supporters) Wan Vace-native-Flux_ref_cn_firstlastframe - Advanced - 20250614
Tutorial Video : https://youtu.be/CmAGOcbU1T4Related Post : https://www.patreon.com/posts/131444835In this workflow, I add a little more advance features. Start frame image creation based on the Pose of the reference video start frame. I use Flux Tool Depth Map this time for Flux image ControlNet Added 2 Vace conditions to handle motion, more accurate motion for the generate videoCheck out the tutorial if you don't know how to use it. Have fun :)
Wan 2.1 Vace Start End Frame - One Of The Most Flexible Frame Interpolation Method
Video : https://youtu.be/CmAGOcbU1T4Discover the power of WAN 2.1 Vace Start/End Frame—one of the most flexible frame interpolation methods for AI video generation! This tutorial dives into creating smooth, dynamic animations using just a start frame (and optional end frame) paired with ControlNet guides like DW Pose and DepthAnything. See how to avoid morphing artifacts and achieve cinematic transitions—perfect for character spins, walking sequences, and complex motions.Who Is This Content Suitable For?AI video creators, animators, and VFX...
Wan 2.1 FusionX Fine Tune Model - They Called It The Master Model!
Video : https://youtu.be/1wD8Iyvv1a4Discover Fusion X—the powerhouse AI model combining AccVideo, Movie Gen 1.1, CausVid, and MPS Reward LoRA into one seamless tool for cinematic, high-quality video generation. In this video, we compare Fusion X with standard WAN 2.1 Vace, showcasing its superior dynamic scenes, human character refinement, and faster processing—all while running efficiently on quantized GGUF models.Who Is This Content Suitable For?AI video creators, filmmakers, and tech enthusiasts seeking next-level video generation. Wheth...
(For Patreon Supporters) Wan Vace - Multi Condition With Face Refiner Ver 20250611
Let's make it short in the post, and I am focus on the additional features.This workflow are based on The tutorial video : https://www.youtube.com/watch?v=XTYwSht6n9YAnd Wan Vace - Multi Condition Basic workflow : https://www.patreon.com/posts/comfyui-wan-2-1-131089902Additional Feature, to add more detail for character facial expression.Added FaceRefine group that is focus on adding facial expression, Ideally for medium or wider shot video. Close up shot, usually the first sampling group are good enough to render detail face. In diffusion ...
ComfyUI Wan 2.1 VACE Multi Conditions Using Native Node - Advanced AI Video Editing
Video : https://youtu.be/XTYwSht6n9YTake a look of ComfyUI Native node full potential of using WAN 2.1 Vace for AI video editing! This tutorial dives into advanced techniques for character swapping, background replacement, and face swapping—all using native nodes. See how to transform landscapes, create apocalyptic scenes, and seamlessly swap faces with precise motion control using DepthAnything and DW Pose.Who Is This Content Suitable For?Video editors, VFX artists, content creators, and AI enthusiasts looking to push the boundaries of AI vi...
ComfyUI Wan 2.1 VACE Multi-Conditions - Better ControlNet And Masking V2V
Tutorial Video : https://youtu.be/wo1Kh5qsUc8Master WAN 2.1 Vace for advanced AI video motion control! This tutorial dives deep into handling complex character animations—like 360-degree turns, dynamic runs, and precise head movements—using multiple ControlNet preprocessors (DepthAnything, DW Pose) and masked style transfers. See how Vace outperforms basic Fun Control for fight scenes, fast motions, and futuristic restyling.Who Is This Content Suitable For?VFX artists, game developers, and AI video creators tackling high-motion sequences or...
ComfyUI Wan 2.1 With Trajectory Motion Control For AI Video
Video: https://youtu.be/UM35z2L1XbIDive into ByteDance's ATI (Any Trajectory Instruction) for WAN 2.1—a new motion control in AI video generation! This video tests its ability to transform static images into dynamic scenes with trajectory tracking, from Spartan combat thrusts to dramatic character movements. See how it outperforms older models like CogVideo X and handles uncensored/local AI workflows.Who Is This Content Suitable For?AI video creators, and tech enthusiasts exploring fine-grained motion control. Whether you're animating...
Flux Kontext From Black Forest Labs - New Level Of AI Image Editing Model That Beats Others
Video : https://youtu.be/E0qmScftDIUDiscover Flux 1 Kontext—Black Forest Lab's breakthrough AI for in-context image editing! This video tests its Pro and Max versions on their Playground, showcasing how it transforms outfits, backgrounds, and text with unmatched consistency. See real-time edits like adding shirts to models, turning lakes to icy winters, or generating cyberpunk styles—all with minimal prompts.Who Is This Content Suitable For?Digital artists, e-commerce designers, marketers, and AI enthusiasts seeking powerful image-editi...
Chatterbox TTS With ComfyUI Locally - You Need This For Your AI Video!
Tutorial Video : https://youtu.be/AquKkveqSvAUnlock next-level voice cloning with Chatterbox—a cutting-edge zero-shot TTS model that rivals ElevenLabs! This video tests its capabilities locally and online, showcasing dynamic voice tones, emotion control via "exaggeration" settings, and seamless ComfyUI integration. Discover how to clone voices in seconds, convert audio with AI, and integrate it into lip-syncing workflows like Fantasy Talk.Who Is This Content Suitable For?Voiceover artists, content creators, AI developers, and tech ent...
(ComfyUI Workflow) Fantasy Talking + Chatterbox TTS - Multimodal AI Talking Avatar - Version 20250531
Tutorial Video : https://youtu.be/AquKkveqSvARelated Post: https://www.patreon.com/posts/130375333This is an update of the Fantasy Talking workflow , and because Chatterbox TTS is better in overall performance than Dia TTS. I decide to apply Chatterbox in this update.I removed Flux Image generation group , because the talking avatar image is better to handle by a dedicate workflow for image generation.
ComfyUI With New Phantom 14B - Another Level Of Reference2Video AI Model!
Video : https://youtu.be/Ti3Z3DSRDw8We dive into the power of ByteDance's Phantom 14B—a cutting-edge fine-tuned model for reference-based AI video generation! This tutorial explores its integration with ComfyUI, showcasing how it revolutionizes subject replication (e.g., clothing, accessories, characters) while slashing render times. Learn to harness the WAN 2.1-based model with CausVid LoRA for faster, higher-quality outputs and real-world applications like e-commerce demos.Who Is This Content Suitable For?AI developers, video creators, ...
(For Patreon Supporters) Update Workflow Phantom S2V Ver. 20250529
Update Workflow based on previous Phantom post: https://www.patreon.com/posts/phantom-run-in-2-127648117If you use CausVid Lora, you don't need TeaCache.For Phaton + CausVid : Floatmatch_causvid CFG = 1, Shift = 8 , Steps = 15-20 get higher quality , CausVid step 8-12 are showing lower quality.Without CausVid :UniPC , Steps = 25+ ,CFG = 7.5, Shift = 5 You get good quality without CausVid Lora, but it take very long time.
n8n with ComfyUI - AI Video Automation Local Setup Walkthrough
Video Tutorial : https://youtu.be/S8Lv-CdGMJUDiscover how to supercharge your AI video automation workflows by integrating ComfyUI with n8n! In this tutorial, we walk through automating image and video generation using AI agents, language models, and seamless tool collaboration. Learn to build end-to-end systems that replace manual tasks, from creating dynamic text prompts to converting images into videos and saving outputs locally or to cloud platforms like Google Drive.Additional : How To Create ComfyUI API Endpoint For AI Image Generation ht...
Float +Vace native - Talking Avatar - Version 20250523
Video Tutorial : https://www.patreon.com/posts/129688882/Step-by-Step Workflow in ComfyUIStep 1: Setting Up ComfyUIBefore diving into the workflow, ensure you have ComfyUI installed and running. If you’re new to ComfyUI, follow these steps:Install Python and set up a virtual environment (e.g., using Conda).Clone the ComfyUI repository from GitHub.Install the required dependencies using pip install -r requirements.txt.Step 2: Installing the Float Custom NodeOpen your terminal or command prompt.Navigate to the custom_nodes folder in your ComfyU...
Float In ComfyUI With Wan VACE - Audio-driven Talking Portrait
Video Tutorial : https://youtu.be/YTZ5J3KcC60For Patreon Supporters: https://www.patreon.com/posts/129689876In this video, we explore Float, a cutting-edge AI talking portrait tool that generates lifelike lip-syncing avatars from just an image and audio input. Learn how to install and run Float in ComfyUI, optimize settings for facial expressions and emotions, and integrate it with WAN 2.1 Vace for advanced outpainting to create dynamic, full-body talking videos. Whether you're a content creator, AI enthusiast, or developer, this tutorial c...
Workflow Update - Wan Vace 5in1 (Ver. 20250522)
Previous version : https://www.patreon.com/posts/comfyui-workflow-129211762Current version update:RemoveBG and White background node changeRenoveBG node new version only have Alpha background, that cause image cannot process further in image resize or pass into Vace Encode.Now Fill background added.Mask Edit Group update:Connect to shared Model loader group, using flowmatch_cuasvid sampling method, it enable to use CausVid Lora + Vace with masking, and quality remain 14B level.81 frames mask edit, I got 2:10 minutes with 8 steps.Video Upscaler ...
(For Patreon Supporters) Natively Running WAN 2.1 VACE + GGUF and Integrating with Flux for First Frame Ver.20250521
Tutorial Video : https://www.youtube.com/watch?v=UUCmCyABmScGGUF Repo https://huggingface.co/QuantStack/Wan2.1-VACE-14B-GGUF ComfyUI Repackaged Repo https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/tree/main ComfyUI Blog Post https://blog.comfy.org/p/wan21-vace-native-support-and-aceStep 1: Setting Up WAN 2.1 Models in ComfyUITo begin, ensure you’ve downloaded the necessary models for WAN 2.1:Text-to-Video Models : These allow you to generate videos from textual prompts.Additional Models for VACE : Specifically, the 14B FP8 and BF1...
ComfyUI Native Using Wan 2.1 VACE & GGUF for Less Memory Processing!
Video : https://youtu.be/UUCmCyABmScAdditional Contents for Patreon Supporters: https://www.patreon.com/posts/129526216In this video, we explore the latest updates for WAN 2.1 Vace in ComfyUI, including native node support and GGUF quantized models for low-VRAM setups. Learn how to run the 14B FP8 and BF16 models efficiently, even on mid-range hardware, and discover optimized workflows for text-to-video generation with ControlNet pose and style transfer. We also cover LoRA integration (COS V14B) for faster sampling and higher-quality outputs, a...
Wan 2.1 VACE 14B Released - Better Quality For AI Video Editing - Tutorial Guide
Video : https://youtu.be/PWVXbeKlXbY(ComfyUI Workflow)Wan VACE 14B 5 In 1 Features Workflow Version 20250518 : https://www.patreon.com/posts/comfyui-workflow-129211762?utm_source=youtube&utm_medium=video&utm_campaign=20250519In this video, we dive into the powerful WAN 2.1 V Ace AI video editing model, now available in 1.3B and 14B parameter sizes on Hugging Face, ModelScope, and official API endpoints. Discover its advanced features like object swapping, video inpainting/outpainting, frame interpolation, and masking—all tested locally in Com...
(ComfyUI Workflow)Wan VACE 14B 5 In 1 Features Workflow Version 20250518
hey!Just tested 14B last 2 days and updated VACE workflows. And I was thinking, how about just put it in one workflow, with all the fundamental features that VACE have, and just share to you all.It's pretty interesting and things come with good timing, we just talk about CausVid and MoviiGen in the last video. Find out that CausVid 14B Lora model do help a lot for loading VACE 14B locally, and within 4-6 steps it does generate something like a normal Wan 2.1 video. Plus, we loaded with Wan 14B Text2Vid model, it do perform even better. I th...
Wan 2.1 Fine Tune Models going wild! CausVid t2v And MoviiGen1.1
Video : https://youtu.be/zO52xUBJlP0In this video, we explore two powerful AI video models, CausVid and MovieGen, both fine-tuned from WAN 2.1. CausVid specializes in fast auto-regressive video generation, ideal for quick, short clips, while MovieGen focuses on cinematic-quality video with enhanced realism and higher resolutions like 720p and 1080p. We test these models in ComfyUI, comparing their performance, speed, and output quality, and provide practical tips for running them on consumer-grade hardware. Whether you're an AI enthusiast, ...
(For Patreon Supporters) Hunyuan Custom subject2vid and LTX Upscaler
Hunyuan Custom Post : https://www.patreon.com/posts/hunyuan-custom-128857913Video about Hunyuan Custom : https://youtu.be/4MwxgtZztOULTX Video 13B Post: https://www.patreon.com/posts/ltx-video-13b-in-128627105Video about LTX Video 13B : Video : https://youtu.be/2asTwiDn0y0------------------------------------------------------------------------Previously, we used the SDXL tile upscaling method for both images and videos. However, when working with AI-generated videos—especially those based on DiT (Diffusion Transformer) models—using SDXL ups...
Hunyuan Custom Subject Video Customization - New AI Model For Referencing To Video Generation
Video : https://youtu.be/4MwxgtZztOUIn this video, we explore Hanyuan Custom, a cutting-edge AI video generation model that allows you to create dynamic videos using reference images, text prompts, and even audio inputs. Discover how to leverage features like single-subject and multi-subject customization, audio-driven video generation, and advanced referencing techniques to produce high-quality, coherent videos. Learn how to optimize performance with FP8 models for local GPUs and integrate this tool into your workflow using ComfyUI.Who is this...
ACE Step AI Music Model Running Locally And ComfyUI - Tutorial Guide + Workflow
Video : https://youtu.be/0TqzphAv1-YIn this video, we dive into Ace Step, an open-source AI music generation model developed by Ace Studio and Step Fun. Discover how to generate, edit, and enhance music using features like text-to-audio, audio-to-audio, lyrics editing, and repainting. Learn how to run this model locally via ComfyUI or the Hugging Face demo, optimize memory usage, and experiment with creative workflows for music production.This time some fun ideas at the end of the video part. :)Who is this for?Musicians and producers exploring ...
LTX Video 13B In ComfyUI - Still The Most Fastest AI VIdeo & Getting Better!
Video : https://youtu.be/2asTwiDn0y0In this video, we explore the latest update of the LTX Video AI model, version 0.9.7 with 13 billion parameters. Discover how to run this powerful model locally using FP8 or GGUF-quantized versions, optimized for consumer-grade GPUs like the Nvidia 4090 and 5090. Learn about the new custom nodes, Spatiotemporal Skip Guidance (STG), and the improved prompt enhancer powered by Llama 3.2.3B. We also cover workflow optimizations, upscaling techniques, and performance comparisons to help you generate high-quality ...
FramePack With LipSync Create AI Talking Avatar - Let's Push the Limits for Local AI!
Video: https://youtu.be/yey8RnwUen4Create AI talking with this step-by-step ComfyUI workflow! In this video, we combine FramePack, Flux LoRA, and Latent Sync 1.5 to generate lifelike avatars—from image generation to lip-syncing and voice cloning. Learn how to optimize text-to-speech (F5TTS), video duration, and upscaling for crisp results. Perfect for YouTube Shorts, virtual influencers, or creative storytelling!Mentioned AI Models In This Video:Flux ACE ++ In ComfyUI https://www.youtube.com/watch?v=2fgT35H_tuELatentSync https://www.youtube.c...
(For Patreon Support) FramePack Create AI Talking Avatar
Related Post : https://www.patreon.com/posts/128455298Video : https://youtu.be/yey8RnwUen4--------------------------------------------------------------------------------------------------------------In this blog post, we'll dive deep into an advanced workflow for creating lifelike AI talking avatars using open-source tools like FramePack, ComfyUI, and Latent Sync 1.5.Why FramePack is Perfect for Talking AvatarsWhen creating talking avatars, video duration matters. Unlike static images, avatars need time to deliver their message - typically...
