Our first live session for our generative AI course is today! As a thank you for being a supporter, we invite you to join for interactive demos, generation credits, and Q&A.
Register here to join
Tuesday, August 12 at 3pm EST
If you were on the fence about the course, feel free to join in for a preview and have your questions answered.
Hope to see you there,
Nate & Chriselle
We’ve spent a lot of time developing this to be the most comprehensive resource for mastering AI for your personal and professional goals.
This isn't just another series of tutorials. It's a structured, 7-module, 38+ lessons, 6-8 weekly live sessions, deep dive into the concepts, workflows, and advanced techniques to help you become a leader in this new era of innovative tech and art.
Thanks for being a part of this journey! If you have any questions feel free to reach out and I'd be happy to help. I'm also making a video explaining the course structure, student perks, partner benefits, guest educators, and more! Here's a link to the course FAQ in the meantime.
Here’s a glimpse of the core modules you'll master:
Go beyond prompting. We'll build your core understanding of how generative models actually work, giving you the foundation to innovate, not just imitate.
Get hands-on immediately. We’ll guide you step-by-step through setting up your creative engine, whether it's a local ComfyUI install or a cloud-based environment on Runpod. You'll go from a blank canvas to a finished image in your very first practical workflow.
Move past basic image creation. We’ll show you how to train your own LoRA models for consistent characters and gain precise compositional control with advanced ControlNet workflows, inpainting, and upscaling. You’ll master the latest AI models to achieve unprecedented results.
Bring your stories to life. We’ll guide you through closed-source online platforms and open-source animation tools within ComfyUI. From concept to finished project, you’ll also learn to generate custom audio, from voice cloning to royalty-free music and lip-sync animation.
Seamlessly integrate AI into your professional workflow. We’ll help you composite AI-generated effects into live-action footage using After Effects and Premiere Pro for a polished, final look. Learn the best techniques for innovative AI-driven VFX and post-production workflows.
Stay ahead of the curve. We'll explore the future of AI, from 3D AI workflows to a full short-film breakdown, combining every technique from the course into a comprehensive case study. We’ll also explore cutting-edge unreleased AI models, and strategies to prepare for the future of AI.
Turn your skills into a career. We’ll show you how to package your AI-powered services, manage client projects, and even deploy your custom workflows for collaborators or commercial use.
We started Black Mixture to empower the next generation of creatives, and this course is the next step in that mission. We can't wait to see what you create.
Cheers,
Nate & Chriselle Dwarika
Black Mixture
Our Black Mixture Generative AI Course is launching in July, and the waitlist is now officially open! This in-depth program is designed to help you achieve ambitious, real-world goals: create your own AI short film, build an AI-powered startup, or land a job in this rapidly growing industry.
Joining the waitlist gives you:
3 FREE hours of RunComfy credits when the course launches.
An EXTRA bonus credit pack if you're one of the first 200 to sign up.
A guaranteed early-bird discount on enrollment.
This is the next step for creatives who want to build the future. Don't miss the bonus for signing up early.
Click Here to Join the Waitlist & Secure Your Bonuses!
Let's build.
Nate & Chriselle
2025-06-06 01:01:14 +0000 UTC View Post
Hey Black Mixture fam! We're excited to release our new exclusive advanced ComfyUI workflow for using Wan 2.1 Phantom. This amazing new model makes it possible to generate videos with multiple subject references.
Perfect for having multiple characters interact, specific character-object interactions, and more!
Required Files & Installation
Place these files in the correct folders inside your ComfyUI directory:
🔹 Phantom Wan2.1_1.3B Diffusion Models
🔗https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Phantom-Wan-1_3B_fp32.safetensors
or
🔗https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Phantom-Wan-1_3B_fp16.safetensors
📂 Place in: ComfyUI/models/diffusion_models
🔹 Text Encoder Model
🔗https://huggingface.co/Kijai/WanVideo_comfy/blob/main/umt5-xxl-enc-bf16.safetensors
📂 Place in: ComfyUI/models/text_encoders
🔹 VAE Model
🔗https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/blob/main/split_files/vae/wan_2.1_vae.safetensors
📂 Place in: ComfyUI/models/vae
Color-coded for clarity:
🟥 Load Models + Pick Your Addons
🟨 Load Subject Reference Images + Prompt
🟦 Generation Settings
🟩 Review Generation Results
🟪 Important Notes
Additionally, we've included a Compare Image Output section to allow you generate awesome documentation videos.
⚙️ Quick Start GuideInstall the latest Kijai WanVideoWrapper custom nodes. *This needs to be installed manually. You can get the latest version by following these instructions:
For new installations:
In "ComfyUI/custom_nodes" folder
open command prompt (CMD) and run this command:
git clone https://github.com/kijai/ComfyUI-WanVideoWrapper.git
for updating previous installation:
In "ComfyUI/custom_nodes/ComfyUI-WanVideoWrapper" folder
open command prompt (CMD) and run this command: git pull
Download and place the Phantom-Wan1_3B safetensors file in the diffusion_models folder in ComfyUI.
Load the Phantom Wan 2.1 workflow
Install missing nodes: https://github.com/kijai/ComfyUI-KJNodes
Set your models, reference image options, and addons
Drag in reference images + enter your prompt
Click generate and review results!
The reference images are used as a strong guidance (try to describe your reference image in your prompt)
Works especially well for characters, fashion, objects, and backgrounds
LoRA implementation does not seem to work with this model, yet we've included it in the workflow as LoRAs may work in a future update.
Different Seed values make a huge difference in generation results. Some characters may be duplicated and changing the seed value will help.
Settings are optimized but feel free to adjust CFG and steps based on speed and results.
📸 We’ve included example generations and their prompts in this post so you can see the magic in action! (GIFS are compressed to fit online)
Check out the video tutorial: https://youtu.be/uBi3uUmJGZI
Created by Nate Dwarika from The Black Mixture Team
💝 THANK YOU to all our amazing Patrons for supporting these releases!
2025-05-29 04:55:00 +0000 UTC View PostMay 29th 2025 Update: Fixed compatability issues with latest updates to ComfyUI and Remove Background nodes
Hey Black Mixture Fam!
Get ready for some amazing new creative possibilities! In this new video we take a deep dive into Google's incredible new video AI release: Veo 3 and the Flow filmmaking tool.
These tools are opening up some really exciting avenues for digital creators, and I can't wait to show you more as the technology develops. For now, I encourage you to check out the links and be sure to join the waitlist for our upcoming Advanced Generative AI Course (backed by NVIDIA): https://blackmixture.mykajabi.com/pl/2148678862
So, what are Veo and Flow?
Veo is Google's state-of-the-art AI model for generating high-quality video from text and image prompts. They've just announced Veo 3, which looks incredibly powerful! It's especially impressive because it can generate realistic audio and video combined. No more silent video generations! The results are impressive but at the crazy price-tag it might be worth waiting a few more months to see how the industry reacts.
Learn more about Veo on the Google DeepMind site: https://deepmind.google/models/veo/
Flow is a brand-new AI filmmaking tool from Google Labs, designed to help you create cinematic scenes and stories using Veo. It’s built with creatives in mind, allowing you to iterate and build out your ideas. The features advertised are impressive, if only they were available to users right away. In the video we showcase what's currently unavailable despite being locked behind a paywall. If only Google would release finished products...
Check out Flow here: https://labs.google/flow/about
Want to try them out and animate your own images?
You can dive into Flow and start creating! Flow includes features like "Frames to Video" (not available) allowing you to upload your images and bring them to life with Veo. Since you can't actually use this feature yet, a workaround is using Veo 2 in Google's AI Studio portal which has less restrictions and more features, despite being completely FREE.
Check out AI Studio here: https://aistudio.google.com/
Access to Flow and the latest Veo models (including for image animation) is available through the Google AI Plans. You can find details about their pricing here:
Google AI Plans: https://one.google.com/about/google-ai-plans/
Stay tuned! Nate
2025-05-23 12:30:38 +0000 UTC View Post
Here's the video guide for our latest Phantom WAN2.1 exclusive advanced workflows. In this video we cover the benefits of the new Phantom WAN2.1 model, the main differences between the supporter exclusive and free versions of the Phantom WAN2.1 workflows, and an in-depth guide to installing and running these workflows in ComfyUI. We also showcase each of the nodes, their settings, and tips to get the most out of your generations! Hope you all enjoy and thanks for supporting!
We're launching a new Diamond supporter tier which includes everything from the previous tiers, plus 1 on 1 monthly consultations. During these 1 on 1 meetings, you will be free to ask any questions about AI workflows, setups, or we can discuss your personal and professional goals. Once subscribed, you'll receive a personal DM with a calendar link to schedule a video call.
Hey Black Mixture fam! We're excited to release our new ComfyUI workflow for using Wan 2.1 Phantom. This amazing new model makes it possible to generate videos with multiple subject references.
Perfect for having multiple characters interact, specific character-object interactions, and more!
Get the exclusive advanced version (more features) of this workflow here: https://www.patreon.com/posts/black-mixtures-2-128029818
Required Files & Installation
Place these files in the correct folders inside your ComfyUI directory:
🔹 Phantom Wan2.1_1.3B Diffusion Models
🔗https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Phantom-Wan-1_3B_fp32.safetensors
or
🔗https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Phantom-Wan-1_3B_fp16.safetensors
📂 Place in: ComfyUI/models/diffusion_models
🔹 Text Encoder Model
🔗https://huggingface.co/Kijai/WanVideo_comfy/blob/main/umt5-xxl-enc-bf16.safetensors
📂 Place in: ComfyUI/models/text_encoders
🔹 VAE Model
🔗https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/blob/main/split_files/vae/wan_2.1_vae.safetensors
📂 Place in: ComfyUI/models/vae
Color-coded for clarity:
🟥 Load Models + Pick Your Addons
🟨 Load Subject Reference Images + Prompt
🟦 Generation Settings
🟩 Review Generation Results
🟪 Important Notes
⚙️ Quick Start GuideInstall the latest Kijai WanVideoWrapper custom nodes. *This needs to be installed manually. You can get the latest version by following these instructions:
For new installations:
In "ComfyUI/custom_nodes" folder
open command prompt (CMD) and run this command:
git clone https://github.com/kijai/ComfyUI-WanVideoWrapper.git
for updating previous installation:
In "ComfyUI/custom_nodes/ComfyUI-WanVideoWrapper" folder
open command prompt (CMD) and run this command: git pull
Download and place the Phantom-Wan1_3B safetensors file in the diffusion_models folder in ComfyUI.
Load the Phantom Wan 2.1 workflow
Install missing nodes: https://github.com/kijai/ComfyUI-KJNodes
Set your models, reference image options, and addons
Drag in reference images + enter your prompt
Click generate and review results!
The reference images are used as a strong guidance (try to describe your reference image in your prompt)
Works especially well for characters, fashion, objects, and backgrounds
LoRA implementation does not seem to work with this model, yet we've included it in the workflow as LoRAs may work in a future update.
Settings are optimized but feel free to adjust CFG and steps based on speed and results.
📸 We’ve included example generations and their prompts in this post so you can see the magic in action! (GIFS are compressed to fit online)
Check out the video tutorial: https://youtu.be/uBi3uUmJGZI
Created by Nate Dwarika from The Black Mixture Team
💝 THANK YOU to all our amazing Patrons for supporting these releases!
In this video, we showcase the power of FramePack for creating stunning image-to-video generations. FramePack runs on Nvidia GPUs from the 30xx series and up, requiring as little as 6GB of VRAM.
You can learn how to install FramePack here:
https://www.patreon.com/posts/framepack-local-127422518Video tutorial: https://youtu.be/ZSe42iB9uRU?si=0KDx4GmLYhqwzAKV
This video compilation features examples generated using an RTX 4090, where each video takes roughly 1-2 minutes per second to render. We've included the prompts used for these generations as well. In our testing, FramePack (powered by Hunyuan 13B) excels with real-world scenarios, 3D and 2D animations, camera prompts, and much more, showcasing its versatility. Although these videos were generated at 30FPS, we sped them up by 20% in Premiere Pro for the compilation (as FramePack often creates a slow-motion effect in its output).
Enjoy the process, and feel free to share your thoughts in the comments!
2025-04-27 09:19:56 +0000 UTC View Post
FramePack is one of the most innovative AI video tools released this year! Here's how you can FramePack running on your computer.
Download the latest version of FramePack by visiting the official GitHub page:
https://github.com/lllyasviel/FramePack
or download the linked file on this Patreon post (Windows Only).
Extract the files from the 7-zip file to a hard drive with atleast 40GB of storage.
Navigate to the FramePack folder and click on "update.bat", then once the update finishes, click on "run.bat" (will download required models ~39GB on first run).
Congrats! FramePack will open in your browser and you will then be able to start generating.
Here's a video tutorial with examples to get started: https://youtu.be/ZSe42iB9uRU
Flux UNO - Generate real world objects in any scene: https://www.patreon.com/posts/black-mixtures-126747125
Advanced Flux V4 Workflow (Our Most Comprehensive AI Image Generation Toolkit): https://www.patreon.com/posts/ultimate-flux-v4-120480039
*** We will release an exclusive updated (modded) version of FramePack which includes prompt saving tools, queuing generations, and more features soon!
2025-04-25 12:15:01 +0000 UTC View PostHey friends! We’ve got some incredible new workflows and tools cooking, and as a top-tier supporter, you get to help decide what we release next.
Your vote not only guides our focus for the week, but you’ll also get early access to the winning option and behind-the-scenes insights into how it was built.
What should we drop next?
2025-04-23 22:14:57 +0000 UTC View Post
Hey Black Mixture Fam! 👋🏾 We’re back with another 🔥 ComfyUI workflow, this time featuring the Flux UNO model, a powerful new method for reference-based image generation!
This new model is perfect for product mockups, consistent character design, outfits, and object-specific scenes! This workflow brings your ideas to life using an image reference as guidance without LoRAs to train or elaborate prompt tweaking. Just clean, guided generations based on your input image.
*** Works well for generating images to animate using Wan 2.1! Get the advanced Wan 2.1 image to video workflow here: https://www.patreon.com/posts/black-mixtures-1-124117186
Required Files & Installation
Place these files in the correct folders inside your ComfyUI directory:
🔹 UNO Custom Node
Clone directly into your custom_nodes folder:
📂 ComfyUI/custom_nodes/ComfyUI-UNO
Or manually download and unzip the ComfyUI-UNO.zip file linked at the bottom of this post.
🔹 UNO Lora File
Download uno_lora.safetensors from this post
or
🔗 https://huggingface.co/bytedance-research/UNO/tree/main
📂 Place in: ComfyUI/models/loras
🔹 Flux1-dev-fp8-e4m3fn.safetensors Diffusion Model
🔗 https://huggingface.co/Kijai/flux-fp8/tree/main
📂 Place in: ComfyUI/models/diffusion_models
🔹 VAE Model
🔗https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/ae.safetensors
📂 Place in: ComfyUI/models/vae
IMPORTANT! Make sure to use the Flux1-dev-fp8-e4m3fn.safetensors model.
Generate highly accurate images using a reference
Perfect for product photography, mockups, clothing consistency
Great for stylized AND realistic image outputs
Side-by-side comparison output option included
Clean, minimal UI flow with optional addons
Color-coded for clarity:
🟥 Load Image + Models
🟨 Prompt + Generation Settings
🟩 Review Generation Results
🟪 Important Notes
Update ComfyUI to the latest version
Install the ComfyUI-UNO node
Download and place the uno_lora.safetensors file
Load our Flux UNO workflow
Drag in a reference image + enter your prompt
Click generate and review results!
The reference image is used as a strong guidance — results are inspired by the image, not copied
Works especially well for fashion, objects, and logos
Pick Your Addons node gives a side-by-side comparison if you need it
Settings are optimized but feel free to adjust CFG and steps based on speed and results.
📸 We’ve included example generations in this post so you can see the magic in action!
Check out the video tutorial: https://youtu.be/eMZp6KVbn-8
Created by Nate Dwarika from The Black Mixture Team
💝 THANK YOU to all our amazing Patrons for supporting these releases!
Hey Black Mixture Fam! 👋🏾 Here's our latest ComfyUI workflow featuring advanced Video-to-Video generation with depth map control for unprecedented motion stability and creative control. This workflow builds on our previous WAN2.1 release but takes video generation to the next level.
👉Get more WAN 2.1 workflows here:
https://www.patreon.com/posts/create-amazing-1-124383647
Required Models & Installation
Place these files in their respective directories:
🔹 WAN2.1 Diffusion Models (1.3B version)
🔗 https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/tree/main/split_files/diffusion_models
📂 ComfyUI/models/diffusion_models
🔹 Text Encoder Model
🔗 https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/tree/main/split_files/text_encoders
📂 ComfyUI/models/text_encoders
🔹 VAE Model
🔗 https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/blob/main/split_files/vae/wan_2.1_vae.safetensors
📂 ComfyUI/models/vae
🔹 Wan2.1-1.3b-control-lora-depth (NEW!)
🔗 https://huggingface.co/spacepxl/Wan2.1-control-loras/tree/main/1.3b/depth
📂 ComfyUI/models/lora
Transform videos with precise control using depth maps
Apply creative styles while maintaining motion coherence
Preserve spatial relationships and scene structure from original videos
Generate motion-consistent results even with dramatic style changes
Multiple quality settings (Low, Great, Max Quality) │Generation times on RTX 4090 GPU⤵
Low (~8 minutes)
Great (~15 minutes)
Max (~30 minutes)
Depth-aware motion control options
Video frame interpolation for smoother results
Support for various input video format
Each workflow is color-coded for easy navigation:
🟥 STEP 1: LOAD MODELS - Set up required model components
🟨 STEP 2: LOAD VIDEO + PROMPT - Load your source video and add prompts
🟦 STEP 3: VIDEO GENERATION SETTINGS - Configure quality, dimensions, and sampling
🟩 STEP 4: SAVE VIDEO - Save and export your results
🟪 HELP & INFO - Support resources and requirements
🛠️ Additional Settings Explained
Quality Settings: Choose between 1-LOW (FAST), 2-GREAT (MID), or 3-MAX (SLOW)
Image Orientation: Select between 1-PORTRAIT or 2-LANDSCAPE
Video Dimensions: Customize width and height (default: 496×768 for portrait, 768×432 for landscape)
Upscale Image: Option to enhance resolution
KSampler Settings: Fine-tune generation parameters with included presets
Depth Control: Adjust strength_model parameter (default: 1.00) to control depth influence
Update ComfyUI to the latest version
Download and place all required models in their respective directories:
Ensure you're using the wan2.1_t2v_1.3B model
Place the Wan2.1-1.3b-control-lora-depth files in the lora folder (NEW!)
Install missing custom nodes nodes
Load your models, video, and prompt
Configure settings and generate!
The workflow includes optional add-ons you can enable:
Save Depth Map Video - Output the generated depth maps
2x Video Frame Interpolation - Double your frames for smoother playback
Patch Super Resolution KL - Enhance resolution quality
Ensure all models are placed in correct directories
Update ComfyUI before using these workflows
Check GPU memory requirements
Join our Discord for support and updates
Use short, simple prompts for best results (example from workflow: "A woman walks on a fashion runway wearing a red hoodie and jeans.")
The "depth_anything" model is included in the workflow (should auto download on first run) for depth map generation
You can enable optional Video Frame Interpolation for smoother results
Compatible with both portrait and landscape orientations with preset dimensions
This workflow is new and highly experimental! Quality improvements will be coming soon so check back here or on our Discord for updates.
Created by Nate Dwarika from The Black Mixture Team
💝 HUGE Thanks to all of our amazing Patreon supporters!
2025-04-04 13:00:15 +0000 UTC View Post
Here's our latest workflow for using Gemini 2.0 Flash in ComfyUI. While you can use Gemini 2.0 Flash in Google's AI Studio, this workflow allows you to harness the image generation and editing capabilities of Gemini natively in your ComfyUI instance. You can combine this with other node setups to take your generations to the next level.
1. Install missing custom nodes and load your images. (You can reference the images in your prompt by describing scenarios like, "Use the logo from the first image, and put it on the hat in the second image". Gemini will understand different images.)
2. Give a prompt to either edit or generate an image.
3. Input your API key into the⭐API Key node. Google Gemini 2.0 Flash is completely free to use. Obtain an API key from here: https://aistudio.google.com/apikey
4. Hit "Run" to send your prompt and receive your generation results.
Thanks for supporting and happy creating!
- Nate from the Black Mixture team
Hey Black Mixture Fam! 👋🏾 We're excited to bring you our exclusive Advanced Wan2.1 Start/End Image to Video workflow for ComfyUI! This powerful workflow allows you to create stunning videos from start and end images with extensive customization options.
*Note: This feature is experimental and some results will vary from generation to generation. We'll post updates here and on the discord.
Required Models & Installation
Place these files in their respective directories:
🔹 WAN2.1 Diffusion Models
🔗 https://huggingface.co/Kijai/WanVideo_comfy/tree/main
📂 ComfyUI/models/diffusion_models
🔹 CLIP Vision Model
🔗 https://huggingface.co/Kijai/WanVideo_comfy/tree/main
📂 ComfyUI/models/clip_vision
🔹 Text Encoder Model
🔗https://huggingface.co/Kijai/WanVideo_comfy/tree/main
📂ComfyUI/models/text_encoders
🔹 VAE Model
🔗https://huggingface.co/Kijai/WanVideo_comfy/tree/main
📂ComfyUI/models/vae
Standard Installation:
1. Clone the Kijai Comfyui-WanVideoWrapper repository to your custom_nodes folder
2. Install dependencies
pip install -r requirements.txt
If you're using the Windows portable version, run this command from the ComfyUI_windows_portable folder:
python_embeded\python.exe -m pip install -r ComfyUI\custom_nodes\ComfyUI-WanVideoWrapper\requirements.txt
Generate videos from text prompts
Transform static images into dynamic videos with start/end image control
Convert and alter existing videos
Multiple quality presets (Low, Great, Max Quality │Generation times on RTX 4090 GPU⤵
Low (~5 minutes)
Great ( ~10 minutes)
Max (~20 minutes)
Customizable frame settings
Advanced motion control options
Video frame interpolation
Lora Support
Naming Logic based on Quality Presets
Longer videos with long context
Each workflow is color-coded for easy navigation:
🟥 Load Models: Set up required model components
🟨 Load Images + Prompt: Input your start/end images and text descriptions
🟦 Video Generation Settings: Configure video parameters and quality options
🟩 Save Video: Export and save your final results
1. Update ComfyUI to the latest version
2. Download and place all required models
3. Install the WAN custom nodes
4. Load the workflow and your start and end frame images
5. Configure settings and generate!
Ensure all models are placed in correct directories
Update ComfyUI before using this workflow
Check GPU memory requirements
Join our Discord for support and updates
For the fastest performance enable torch compile which requires sage attention and triton (separate install)
This is an early release experimental feature and generations may vary a lot based on seed values.
There's a lot of exciting stuff happening with new AI models and the possibilities with ComfyUI. Thanks for supporting and being a part of this journey. We'd love your input to decide what things we should focus on for the next week. Your vote will help determine what gets priority!
2025-03-26 00:01:45 +0000 UTC View Post
Hey Black Mixture Fam! 👋🏾 We're excited to bring you our latest ComfyUI workflows for the incredible WAN2.1 models. This release includes advanced options for Text-to-Video, Image-to-Video, and Video-to-Video capabilities! (Now with LORA Support!)
📥 Model Downloads
Required Models & Installation
Place these files in their respective directories:
🔹 WAN2.1 Diffusion Models
🔗 https://huggingface.co/Kijai/WanVideo_comfy/tree/main
📂 ComfyUI/models/diffusion_models
🔹 CLIP Vision Model
🔗 https://huggingface.co/Kijai/WanVideo_comfy/tree/main
📂 ComfyUI/models/clip_vision
🔹 Text Encoder Model
🔗https://huggingface.co/Kijai/WanVideo_comfy/tree/main
📂ComfyUI/models/text_encoders
🔹 VAE Model
🔗https://huggingface.co/Kijai/WanVideo_comfy/tree/main
📂ComfyUI/models/vae
Standard Installation:
1. Clone the Kijai Comfyui-WanVideoWrapper repository to your custom_nodes folder
2. Install dependencies
pip install -r requirements.txt
If you're using the Windows portable version, run this command from the ComfyUI_windows_portable folder:
python_embeded\python.exe -m pip install -r ComfyUI\custom_nodes\ComfyUI-WanVideoWrapper\requirements.txt
Generate videos from text prompts
Transform static images into dynamic videos
Convert and alter existing videos
Multiple quality presets (Low, Great, Max Quality │Generation times on RTX 4090 GPU⤵
Low (~5 minutes)
Great ( ~10 minutes)
Max (~20 minutes)
Customizable frame settings
Advanced motion control options
Video frame interpolation
Lora Support
Naming Logic based on Quality Presets
Longer videos with long context
🎨 Workflow SectionsEach workflow is color-coded for easy navigation:
🟥 Load Models: Set up required model components
🟨 Input: Load your text, image, or video
🟦 Settings: Configure video generation parameters
🟩 Output: Save and export your results
(short and simple prompts have yielded the best results in our experience)
(Realistic scenes tend to work the best from our testing)
(input video helps reference camera and subject positioning)
Best Loras for Wan:
Hadouken: https://civitai.com/models/1337925?modelVersionId=1510868
Squish Effect: https://civitai.com/models/1340141?modelVersionId=1513385
1. Update ComfyUI to the latest version
2. Download and place all required models
3. Install the WAN custom nodes
4. Load the appropriate workflow for your needs
5. Configure settings and generate!
Video Tutorial: https://youtu.be/dMSz2PSiemk
Ensure all models are placed in correct directories
Update ComfyUI before using these workflows
Check GPU memory requirements
Join our Discord for support and updates
When batching multiple generations, if you experience system hangs, try using the Clear Cache in the addon picker.
For the fastest performance, sage attention and triton is recommended (separate install)
Updates
*March 18th, 2025 Update - V1.1 - Bug Fixes & Quality Improvements
Updated the workflows to work with the newest ComfyUI and WanVideoWrapper nodes. A few new features would provide errors if you were using the old workflows (related to the default sampler behavior and the TeaCache settings). This update fixes those issues.
Propagated the node icons and structure from the Image to Video workflow to all of the exclusive advanced workflows for Text to Video, Text to Video (Long Context), and Video to Video.
*March 21st, 2025 Update - V1.2 - Bug Fixes & Quality Improvements
Updated Image to Video workflow to include SLG (better motion) and changed default text encoder node behavior to "force_offload = true".
*March 23rd, 2025 Update - V1.2 - Bug Fixes & Quality Improvements
Fixed issues with Video to Video workflow and added SLG (better motion).
Created by Nate Dwarika from The Black Mixture Team
💝 HUGE Thanks to all of our amazing Patreon supporters and special thanks to Kijai for the custom nodes!
2025-03-23 16:15:00 +0000 UTC View Post
Gemini 2.0 Flash is Google's latest AI image editor that lets you transform images using natural language. After extensive testing, I've discovered what it's great at, where it struggles, and how to work around its limitations.
Use Gemini 2.0 Flash here: aistudio.google.com
Colorizing Photos: Breathes life into black & white images with remarkable accuracy
2D to 3D Conversions: Transforms flat illustrations into dimensional renders
Adding/Removing Objects: Seamlessly modifies image elements (clothing, backgrounds, text, and more)
Style Transfers: Applies artistic styles across images
Generate Story Scenes: Create a series of images based on story lines
Copyrighted Material Restrictions
Prompt Complexity Challenges
Image Size & Quality Issues
Gemini has built-in safeguards that restrict editing recognizable characters or figures. Here's how to work around it:
Problem: "Make this Naruto character into a 3D render" ❌
Solution: "Make this 2D drawing into a highly detailed 3D render" ✅
Tip: Avoid specific character names or franchises. Use "2D drawing" instead of "character" for better results.
Gemini performs best with focused commands. When you give multiple instructions at once, it typically only follows one well.
Problem: "Colorize this photo, remove the background, and add mountains behind" ❌
Solution: "Colorize this black and white photo while preserving details" ✅
Tip: Break complex edits into sequential steps. Do one clear instruction at a time, then build upon those results.
Large images are automatically compressed by Google, leading to quality loss:
Large file sizes cause Google to compress your image
Multiple iterations on compressed images compound quality problems
Each edit cycle potentially degrades the result
Tip: Start with moderate-sized images (under 2MB). Save a copy after each successful edit to preserve quality along the way.
Advanced Techniques That Actually Work
Start with clear action verbs: "Transform," "Convert," "Enhance"
Specify exactly what to modify: "the background," "the lighting"
Describe the desired outcome: "into a sunset beach scene with warm orange tones"
Include style references: "with photorealistic details," "in watercolor style"
When converting stylized characters from 2D to 3D:
Avoid mentioning specific franchise names
Use descriptive terms instead of proper nouns
Focus on visual elements rather than character identity
Example: "Convert this 2D illustration into a realistic 3D render with photorealistic textures"
For best results when making several modifications:
Start with major structural changes (background elements, composition)
Move to style modifications (artistic style, lighting)
End with fine details and refinements
Save each version separately
Tip: Limit the total number of iterations on a single image to preserve quality.
Sample Prompts That Work Like Magic
"Colorize this black and white photo with natural, historically accurate colors"
"Convert this photograph into a watercolor painting with visible brush strokes"
"Transform this portrait into an oil painting in Renaissance style"
"Transform this photo into a line art"
"Transform this 2D drawing into a detailed 3D render with realistic texturing"
"Convert this simple sketch into a fully rendered illustration with detailed shading"
Final Thoughts
Gemini 2.0 Flash is powerful but has its quirks. Understanding its limitations and using these workarounds will help you achieve impressive results without frustration. The key is keeping your prompts simple, avoiding copyright triggers, and building complex edits step by step.
Happy editing!
- Nate Dwarika from Black Mixture
Hey Black Mixture Fam! 👋🏾 We're excited to bring you our latest ComfyUI workflows featuring the incredible WAN2.1 model. This release includes Text-to-Video, Image-to-Video, and Video-to-Video capabilities!
👉Get the EXCLUSIVE ADVANCED workflows here:
https://www.patreon.com/posts/black-mixtures-2-124117186
Required Models & Installation
Place these files in their respective directories:
🔹 WAN2.1 Diffusion Models
🔗 https://huggingface.co/Kijai/WanVideo_comfy/tree/main
📂 ComfyUI/models/diffusion_models
🔹 CLIP Vision Model
🔗 https://huggingface.co/Kijai/WanVideo_comfy/tree/main
📂 ComfyUI/models/clip_vision
🔹 Text Encoder Model
🔗https://huggingface.co/Kijai/WanVideo_comfy/tree/main
📂ComfyUI/models/text_encoders
🔹 VAE Model
🔗https://huggingface.co/Kijai/WanVideo_comfy/tree/main
📂ComfyUI/models/vae
Standard Installation:
1. Clone the Kijai Comfyui-WanVideoWrapper repository to your custom_nodes folder
2. Install dependencies
pip install -r requirements.txt
If you're using the Windows portable version, run this command from the ComfyUI_windows_portable folder:
python_embeded\python.exe -m pip install -r ComfyUI\custom_nodes\ComfyUI-WanVideoWrapper\requirements.txt
Generate videos from text prompts
Transform static images into dynamic videos
Convert and alter existing videos
Multiple quality settings (Low, Great, Max Quality │Generation times on RTX 4090 GPU⤵
Low (~5 minutes)
Great ( ~10 minutes)
Max (~20 minutes)
Customizable frame settings
Advanced motion control options
Video frame interpolation
🎨 Workflow SectionsEach workflow is color-coded for easy navigation:
🟥 Load Models: Set up required model components
🟨 Input: Load your text, image, or video
🟦 Settings: Configure video generation parameters
🟩 Output: Save and export your results
(short and simple prompts have yielded the best results in our experience)
(Realistic scenes tend to work the best from our testing)
(input video helps reference camera and subject positioning)
1. Update ComfyUI to the latest version
2. Download and place all required models
3. Install the WAN custom nodes
4. Load the appropriate workflow for your needs
5. Configure settings and generate!
Ensure all models are placed in correct directories
Update ComfyUI before using these workflows
Check GPU memory requirements
Join our Discord for support and updates
For the fastest performance, sage attention and triton is recommended (separate install)
Updates
*March 18th, 2025 Update - V1.1 - Bug Fixes & Quality Improvements
Updated the workflows to work with the newest ComfyUI and WanVideoWrapper nodes. A few new features would provide errors if you were using the old workflows (related to the default sampler behavior and the TeaCache settings). This update fixes those issues.
*March 21st, 2025 Update - V1.2 - Bug Fixes & Quality Improvements
Updated Image to Video workflow to include SLG (better motion) and changed default text encoder node behavior to "force_offload = true".
Created by Nate Dwarika from The Black Mixture Team
💝 HUGE Thanks to all of our amazing Patreon supporters and special thanks to Kijai for the custom nodes!
2025-03-18 14:23:00 +0000 UTC View Post
Here's the video guide for our latest WAN2.1 exclusive advanced workflows. In this video we cover the main differences between the supporter exclusive and free versions of the WAN2.1 workflows. We also showcase in depth each of the nodes, their settings, and tips to get the most out of your generations! Hope you all enjoy and thanks for supporting!
GPU Compatibility:
RTX 40xx (Ada) and RTX 30xx (Ampere): Fully supported.
RTX 50xx (Blackwell): Support is very new and requires the latest software.
RTX 20xx (Turing) and older: Limited compatibility – you may experience issues.
Open your ComfyUI directory.
Check if there’s a folder named python_embeded inside it.
(If you previously installed extensions using Triton, you might already see a libs folder inside python_embeded. If not, don’t worry—we’ll add it later.)
Triton is required for SageAttention to work properly. There are two options here:
1️⃣ Standard Installation:
Open a Command Prompt (Win + R → type cmd → hit Enter).
Navigate to your ComfyUI folder using the command:
cd path\to\ComfyUI_windows_portable
(Replace path\to\ComfyUI_windows_portable with your actual folder location.)
Run the following command to install Triton:
python_embeded\python.exe -m pip install -U triton-windows
Note: If the package “triton-windows” isn’t found on PyPI, download the appropriate wheel—e.g., triton-3.2.0-cp312-cp312-win_amd64.whl—from here, place it in your ComfyUI folder, and install it with:
python_embeded\python.exe -m pip install path\to\triton-3.2.0-cp312-cp312-win_amd64.whl
(Replace path\to\... with the actual file path.)
2️⃣ Pre-release Version (if needed):
If you need the pre-release version (Triton 3.3), use:
python_embeded\python.exe -m pip install -U --pre triton-windows
Run this command in the same Command Prompt window:
python_embeded\python.exe -m pip install -U sageattention
Note: If you see a deprecation warning about loading an egg (e.g., sageattention-2.0.1-py3.12-win-amd64.egg), it’s recommended to first upgrade pip:
python_embeded\python.exe -m pip install --upgrade pip
Then force a reinstall if necessary:
python_embeded\python.exe -m pip install --force-reinstall sageattention
Teacache is an optional but useful caching tool that speeds up processing.
Note: You can now install it via the ComfyUI Manager. Open the Manager, go to the "Custom Nodes Manager", search for “comfyui-teacache”, and install it.
or click "Install via Git URL" with https://github.com/welltop-cn/ComfyUI-TeaCache
If Triton doesn’t work right away, you may need to add libs and include manually:
Download the required files:
If you’re using ComfyUI Portable v0.2.4 or later (with Python 3.12.7), get them from here: 👉 Download include & libs folders
If you have a different Python version, check for compatible files:
👉 Alternative versions here
Extract the downloaded ZIP and place the include and libs folders inside:
📁ComfyUI_windows_portable\python_embeded\
🚨 Important: Ensure you place them correctly:
✅ python_embeded\libs (not python_embeded\Lib)
✅ python_embeded\include
Open ComfyUI and try running an Sage Attention workflow: https://www.patreon.com/posts/black-mixtures-2-124117186
If you experience issues, go back and make sure:
Triton installed correctly (triton-windows shows up in pip list)
Your libs and include folders are in the right place.
Command Not Found?
Ensure you’re running commands from inside the ComfyUI directory.
Sage Attention Not Working?
Reinstall it using:
python_embeded\python.exe -m pip install --force-reinstall sageattention
Triton Compilation Errors?
Make sure you have Visual C++ Redistributable installed:
👉 Download VC Redist
Restart your computer and try again.
Still not working?
Send me a DM on Patreon or Discord, or comment below and I'd be happy to help troubleshoot this!
2025-03-13 15:30:03 +0000 UTC View Post
March 24th, 2025 Update - V1.0
We've put together an EXPERIMENTAL sage attention & triton installer which you can download below. (Still working on the additional features)
Place the "BlackMixture-sage-attention-installer.bat" file in your ComfyUI portable root directory.
Click "run anyway" if you receive a pop up from Windows Defender. (There's no viruses in this file. You can verify the code by right-clicking and opening with notepad. As I'm not a programmer, this was written using LLMs based on this text guide article)
I recommend starting with these options (as the others are more experimental):
1: Check system compatibility
3: Install Triton
4: Install Sage Attention
6: Setup include and libs folders
9: Verify installation
Hey Black Mixture Fam! 👋🏾 Here's our latest ComfyUI workflows featuring the amazing WAN2.1 model now with Native support (no sage attention or custom node wrappers needed). This release includes Text-to-Video and Image-to-Video capabilities with less nodes and an even easier setup.
👉Get the EXCLUSIVE ADVANCED workflows here:
https://www.patreon.com/posts/black-mixtures-2-124117186
Required Models & Installation
Place these files in their respective directories:
🔹 WAN2.1 Diffusion Models
🔗 https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/tree/main/split_files/diffusion_models
📂 ComfyUI/models/diffusion_models
🔹 CLIP Vision Model
🔗 https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/blob/main/split_files/clip_vision/clip_vision_h.safetensors
📂 ComfyUI/models/clip_vision
🔹 Text Encoder Model
🔗https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/tree/main/split_files/text_encoders
📂ComfyUI/models/text_encoders
🔹 VAE Model
🔗https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/blob/main/split_files/vae/wan_2.1_vae.safetensors
📂ComfyUI/models/vae
Generate videos from text prompts
Transform static images into dynamic videos
Multiple quality settings (Low, Great, Max Quality │Generation times on RTX 4090 GPU⤵
Low (~5 minutes)
Great ( ~10 minutes)
Max (~20 minutes)
Customizable frame settings
Advanced motion control options
Video frame interpolation
Each workflow is color-coded for easy navigation:
🟥 Load Models: Set up required model components
🟨 Input: Load your text, image, or video
🟦 Settings: Configure video generation parameters
🟩 Output: Save and export your results
(short and simple prompts have yielded the best results in our experience)
(Realistic scenes tend to work the best from our testing)
1. Update ComfyUI to the latest version
2. Download and place all required models
3. Install the WAN custom nodes
4. Load the appropriate workflow for your needs
5. Configure settings and generate!
Ensure all models are placed in correct directories
Update ComfyUI before using these workflows
Check GPU memory requirements
Join our Discord for support and updates
Created by Nate Dwarika from The Black Mixture Team
💝 HUGE Thanks to all of our amazing Patreon supporters and special thanks to Kijai for the custom nodes!
2025-03-06 12:47:01 +0000 UTC View Post
Hey Black Mixture Fam!
We're thrilled to present the next evolution of our most versatile workflow yet - Ultimate Flux V4! This powerhouse workflow combines three essential modes into one seamless experience, making it your go-to solution for all types of image generation and editing. This is the FREE version which does not feature advanced detailing and the face detailer.
👉Get the ADVANCED workflows here:
https://www.patreon.com/posts/ultimate-flux-v4-120480039
🔹 Flux Dev Diffusion Model
Download: https://huggingface.co/black-forest-labs/FLUX.1-dev/
📂 Place in: ComfyUI/models/diffusion_models
🔹 CLIP Model
Download: https://huggingface.co/comfyanonymous/flux_text_encoders
📂 Place in: ComfyUI/models/clip
🔹 Flux.1 Dev Controlnet Inpainting Model
Download: https://huggingface.co/alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Beta
📂 Place in: ComfyUI/models/controlnet
Three powerful modes in one workflow
Advanced upscaling and face detailing options (in Supporter Exclusive ADVANCED version)
Comprehensive inpainting and outpainting capabilities
Color-coded and labeled sections
Intuitive keyboard shortcuts for quick navigation (0,1,2,3)
Optimized settings for each mode
Default generation mode with upscaling and face detailing (in Supporter Exclusive ADVANCED version)
Complete control over prompting and parameters
Note: All modes will pass through the Upscaler and Face Detailer unless turned off in the Addon Picker. (in Supporter Exclusive ADVANCED version)
Reference-based generation
Adjustable denoise settings for varying levels of transformation
Adjust your denoise values to control how much your reference image is preserved.
0.10 for minimal changes │0.50-0.85 for significant transformations
Powerful inpainting capabilities
Background expansion tools
Load your reference image in the bottom Load Image node. This will be used for the image dimensions. Load your image with a mask in the top node to determine what gets filled in.
Install all required models in their respective folders
Install missing nodes using the ComfyUI manager
Use keyboard shortcuts (0,1,2,3) to navigate between sections
Select your desired mode in the Mode Selector node
Generate and refine your images
Video Guide: https://youtu.be/iBzlgWtLlCw
Make sure your ComfyUI is updated
You can download missing nodes and models using the ComfyUI Manager
Created by The Black Mixture Team 💝 HUGE Thanks to all of our amazing Patreon supporters!
2025-02-06 06:30:00 +0000 UTC View PostThe Short Answer:
CLIP_L and T5xxl are tools that help AI "understand" your text prompts. Think of them as two assistants with different specialties:
CLIP_L= Hashtags focuses on keywords: objects, colors, styles.
T5xxl = Storyteller (focuses on narratives: context, moods, relationships).
Tokens are the building blocks of text that AI models like CLIP and T5 use to process language. Think of them as "chunks" of words or parts of words, similar to how hashtags break down ideas into concise phrases. For example:
CLIP_L: Uses Byte-Pair Encoding (BPE) to split text into tokens. A word like "jumps" might become two tokens: (jump) and (s).
T5xxl: Uses SentencePiece, which handles rare words more gracefully and supports longer sequences (up to 512 tokens)
Tokens limits are the maximum amount of tokens that the CLIP_L or T5xxl can process before it starts to paraphrase, which can lead to details being lost.
CLIP works like a keyword detector. It connects words to images but has a short attention span.
Example: “A neon-lit robot, cyberpunk alley, raining, 8k” ✅ 
What to avoid: Long sentences like “In a dystopian future where robots roam the streets, there’s a neon-lit alley…” ❌
*Tip: Use commas, not paragraphs. Prioritize what you see.
T5xxl thrives on details. It’s great for explaining why something looks the way it does.
Example: “A lonely robot walks through cyberpunk alley at night. Neon signs flicker, reflecting on wet pavement. The scene feels dystopian but hopeful.” ✅
*Tip: Add emotions, backstories, or abstract ideas (e.g., “melancholic,” “inspired by 1980s sci-fi”).
Most AI tools (like Flux) combine both. Here’s how to optimize:
1. Start with CLIP_L:
List the essentials (objects, colors, style).
Example: sunset over mountains, oil painting, golden hues, Van Gogh style.
2. Expand with T5xxl:
Add context to refine the mood or logic.
Example: The painting evokes a serene yet dramatic atmosphere, with bold brushstrokes mimicking Van Gogh’s "Starry Night"
You can make CLIP_L and T5xxl prompts with another tool like ChatGPT or Claude 3.5 Sonnet. Here's an example output from our custom agent (linked below):
1. CLIP_L: “A black man Nate and a Filipina woman Chriselle, advanced mech pilots, cyberpunk battle suits, alien planet, fade haircut and beard visible through helmet visor, long black hair flowing in zero gravity, bioluminescent terrain, floating crystals, dual-wielding plasma weapons, holographic HUD displays, mechanical details, carbon fiber armor, neon accents, binary star system, 8k ultra HD, photorealistic, cinematic sci-fi, ray tracing, volumetric lighting, metallic reflections, HDR, hyper-detailed tech, atmospheric fog, energy effects, Blade Runner style, dramatic perspective”
2. T5xxl “A black man Nate and a Filipina woman Chriselle stand in advanced mech suits. The suits are sleek black with neon blue accents. Nate's fade haircut and beard are visible through his helmet visor. His mech has heavy armor plates. Chriselle's long black hair floats in low gravity. Her mech is streamlined with glowing red details. They stand on a purple crystalline battlefield. Two suns set in the alien sky. Floating rocks hover in the background. Their mechs hold glowing plasma weapons. Holographic displays light up their cockpits. Bioluminescent plants pulse around them. Energy shields shimmer around their suits. The scene is rendered in 8K quality. Ray-traced lighting creates realistic reflections. The mood is futuristic and powerful. The composition is epic and cinematic.” 
Final Tip: Treat CLIP_L like a Pinterest board (visual keywords) and T5xxl like a novel (deeper meaning). The better you balance both, the closer the AI gets to your vision!
Happy creating and thank you for supporting!
- Nate from Black Mixture ♥
Here's a video guide for our most comprehensive workflow yet! Ultimate Flux V4 brings the power of text to image, image to image, and inpaint/outpainting all in one unified workflow.
✨ Download Ultimate Flux V4 [Free Version] Text Guide + Workflow⤵ https://www.patreon.com/posts/120952448
🔥 Download Ultimate Flux V4 [Advanced]Text Guide + Workflow ⤵ https://www.patreon.com/posts/ultimate-flux-v4-120480039
🔥 Black Mixture Discord: https://discord.gg/n2N7Hgvn7n
2025-01-28 13:07:44 +0000 UTC View Post
The next evolution of our most versatile workflow yet - Ultimate Flux V4! w/ upscaling, face detailing, image2image, fill/expand and more!
The next evolution of our most versatile workflow yet - Ultimate Flux V4! w/ upscaling, face detailing, image2image, fill/expand and more!
Hey Black Mixture Fam!
We're thrilled to present the next evolution of our most versatile workflow yet - Ultimate Flux V4! This powerhouse workflow combines three essential modes into one seamless experience, making it your go-to solution for all types of image generation and editing.
🔹 Flux Dev Diffusion Model
Download: https://huggingface.co/black-forest-labs/FLUX.1-dev/
📂 Place in: ComfyUI/models/diffusion_models
🔹 PhotoFlux Pro LoRA Model (Optional for Max-Res Outputs)
Download: https://www.patreon.com/posts/119630460
📂 Place in: ComfyUI/models/loras
🔹 CLIP Model
Download: https://huggingface.co/comfyanonymous/flux_text_encoders
📂 Place in: ComfyUI/models/clip
🔹 Flux.1 Dev Controlnet Inpainting Model
Download: https://huggingface.co/alimama-creative/FLUX.1-dev-Controlnet-Inpainting-Beta
📂 Place in: ComfyUI/models/controlnet
Three powerful modes in one workflow
Advanced upscaling and face detailing options
Comprehensive inpainting and outpainting capabilities
Color-coded and labeled sections
Intuitive keyboard shortcuts for quick navigation (0,1,2,3)
Optimized settings for each mode
Default generation mode with upscaling and face detailing
Complete control over prompting and parameters
Note: All modes will pass through the Upscaler and Face Detailer unless turned off in the Addon Picker.
Reference-based generation
Adjustable denoise settings for varying levels of transformation
Adjust your denoise values to control how much your reference image is preserved.
0.10 for minimal changes │0.50-0.85 for significant transformations
Powerful inpainting capabilities
Background expansion tools
Load your reference image in the bottom Load Image node. This will be used for the image dimensions. Load your image with a mask in the top node to determine what gets filled in.
Install all required models in their respective folders
Install missing nodes using the ComfyUI manager
All of the nodes will show up in the missing nodes section except for two. You can install these by using the "Install via Git URL" in the ComfyUI manager
SkipLayerForward: https://github.com/lodestone-rock/ComfyUI_FluxMod
ComfyUI-ppm: https://github.com/pamparamm/ComfyUI-ppm
To get more sampler options (ays+), install the "ComfyUI-ppm" node pack in the ComfyUI manager.
Use keyboard shortcuts (0,1,2,3) to navigate between sections
Select your desired mode in the Mode Selector node
Generate and refine your images
🎥Video Guide: https://youtu.be/iBzlgWtLlCw
Make sure your ComfyUI is running version 0.3.27
(version 0.3.30 has bugs as of 4/30/25)
To downgrade your ComfyUI version, open a command prompt window in your ComfyUI_portable folder and run this command: git checkout v0.3.27
Use the linked ComfyUI-Inpaint-CropAndStitch node if you receive an OLD warning.
You can download missing nodes and models using the ComfyUI Manager
Make sure to download the ComfyUI-42lux.zip file attached to this post and extract it to your ComfyUI/custom_nodes folder. The node developer removed a set of nodes from Github and if you're still missing nodes, this pack should resolve this issue.
Created by The Black Mixture Team 💝 HUGE Thanks to all of our amazing Patreon supporters!
2025-01-25 03:39:00 +0000 UTC View Post
Hey Black Mixture Fam! Get ready to elevate your AI photography game with our latest professional-grade workflow designed specifically for creating stunning, high-quality photorealistic images with built in large image inferencing for Flux.
📦 Required Models:
🔹 Flux Dev Diffusion Model
Download: https://huggingface.co/black-forest-labs/FLUX.1-dev
📂 Place in: ComfyUI/models/diffusion_models
🔹 PhotoFlux Pro LoRA Model
Download: linked below: BlackMixture_PhotoFlux_Pro
📂 Place in: ComfyUI/models/loras
🔹 CLIP Model
Download: https://huggingface.co/comfyanonymous/flux_text_encoders
📂 Place in: ComfyUI/models/clip
Text to Image & Image to Image generation
Professional-grade photo generation with 3MP & 6MP Inferencing
Built-in Upscaling Options (Uses Image-to-Image)
Optimized Generation Settings
Color-coded and Labeled Sections
Text to Image Workflow:
🟥 Step 1: Load Models & Setup
🟨 Step 2: Prompt
🟦 Step 3: Generation Settings (already optimized)
🟩 Step 4-6: Save Images & Upscale
🟧 Step 7: Compare Images
🟪 Important Notes
Image to Image Workflow:
🟥 Step 1: Load Image
🟦 Step 2: Select Mode & Denoise
🟩 Step 3: Compare Images
Update ComfyUI to latest version
Install all required models in specified directories
Load the workflow and install missing custom nodes
Adjust prompts and settings as needed
Generate your masterpiece!
For Image to Image, make sure to change the mode to "2" which is located in the "Mode Select" node. Also lower your denoise value to ~0.85 for best results.
Keep Clip_L below 77 tokens
Keep T5xxl below 512 tokens
Start with default generation settings and experiment if you know what you're doing.
Created by Nate Dwarika from The Black Mixture Team
💝 HUGE Thanks to all of our amazing Patreon supporters!
2025-01-11 13:15:01 +0000 UTC View Post
Ever wanted to create epic movies with your own characters? Or maybe you'd like to star in your own film? All you need is a single image, an idea, and the power of this new AI model called Minimax Subject Reference.
THIS DOES NOT REQUIRE COMFYUI. Hopefully in the near future we will be able to generate videos like these locally, but in the meantime, this new AI model from Hailuo's Minimax can help us create impressive cinematic visuals and create a character library from a single refence image.
Link to Minimax: https://hailuoai.video/create
You'll need Minimax account to create with this model. After signing up, head over to the "Create" tab or click the link above.
Then select "Subject Reference".
Upload a reference character image. Preferably front facing for the best results.
Here's the image used for Nate from Black Mixture.
Afterwards, type in a prompt and hit generate.
Prompt: A man runs through a wildfire. He aims his hand at the wildfire. His hand begins to glow blue and icy frost particles and blue magic light emits from his hand and extinguishes the fire.
Prompt: A man smiles then starts to cry.
Minimax is great at facial expressions and emotions! It is also great at keeping similar clothing styles to what was used in the reference image.
Hope you have fun creating and be sure to send us your creations! @blackmixture on Instagram.
(If you'd like to test out Minimax and are a supporter, send me an image + prompt, and I'll generate a video for you. Limited availability)
Hey Black Mixture Fam! Transform your text descriptions into stunning videos with our latest Hunyuan-powered workflow! This implementation brings professional-quality video generation capabilities right to your ComfyUI setup - no Sage Attention or Triton required for Windows users. Using native ComfyUI nodes, we've simplified the setup process so it's easier to use!
Diffusion Model
• File: hunyuan_video_t2v_720p_bf16.safetensors
📁Place in: ComfyUI/models/diffusion_models
🔗 Download: https://huggingface.co/Comfy-Org/HunyuanVideo_repackaged/tree/main/split_files/diffusion_models
Text Encoders
• Files:
llava_llama3_fp8_scaled.safetensors
📁 Place in: ComfyUI/models/text_encoders
🔗 Download: https://huggingface.co/Comfy-Org/HunyuanVideo_repackaged/tree/main/split_files/text_encoders
VAE
• File: hunyuan_video_vae_bf16.safetensors
📁Place in: ComfyUI/models/vae
🔗Download: https://huggingface.co/Comfy-Org/HunyuanVideo_repackaged/tree/main/split_files/vae
Native Windows compatibility - no extra dependencies
Professional-quality video generation
Advanced text-to-video synthesis
Streamlined setup process
Customizable video parameters
🟥 Load Models: Configure Hunyuan Video and text encoding components
🟨 Prompt: Input your video description text
🟦 Generation Settings: Adjust video parameters and quality
🟩 Save Video: Export your generated video
🟪 Important Notes: Model downloads and setup instructions
Worfklow/Model Setup
1. Update ComfyUI to latest version
2. Download and place all required model files
3. Load workflow into ComfyUI
Generation Process
1. Input your creative text prompt
2. Adjust video parameters (length, FPS, quality)
3. Generate and export your video
Ensure all models are in correct directories
ComfyUI must be updated to latest version
Follow exact model placement instructions
Manual model downloads required (no auto-download)
Image→Video and Video→Video Coming Soon!
Created by Nate Dwarika from The Black Mixture Team 💝 HUGE Thanks to all of our amazing Patreon supporters and Happy Holidays!
2024-12-24 21:00:07 +0000 UTC View Post
This tutorial will show you how to create consistent character turnarounds with our latest instant 360 turnaround comfyUI workflows!
Download the workflow here:
https://www.patreon.com/posts/ultimate-sv3d-2d-118064425
