Black Mixture
patreon
Black Mixture V3 Ultimate Flux ComfyUI Workflow + Guide
🕑 Added 2024-11-10 13:00:07 +0000 UTCHello Black Mixture Fam! 👋
We're excited to unveil our latest custom ComfyUI workflow: V3! We know it might look a little overwhelming at first glance, but don’t worry – we’re here to make it simple, step-by-step. This guide will give you an overview and some helpful hints so you can jump right in.
Step 1. "Pick Your Add-Ons" = You’ll find add-ons with emojis that highlight the key ones to keep enabled. (Pro-tip: The “face detailer” and “detail daemon” add-ons are great for enhancing image quality!) Skip options you’re unfamiliar with; the defaults work great for most needs.
Understanding the Add-Ons:
The add-ons are powerful features that can enhance your workflow results. Let’s go through each one to help you decide which add-ons to enable and why.
Enable Upscale 🔍
This add-on allows you to improve the resolution of your final image, making it sharper and more detailed. Upscaling is especially useful if you want high-quality output suitable for larger displays or print. For most users, it’s a good idea to keep this option enabled to get the best visual fidelity.Enable Face Detailer 😊
If your image contains human faces, this add-on helps refine facial features, making them more realistic and polished. This can be a lifesaver for portrait-focused images, enhancing clarity and subtle facial details. We recommend enabling Face Detailer if you’re working on any image where the face is an essential element.Enable Canny ControlNet 🚫
Canny ControlNet helps add fine line details by using edge detection, but it’s only necessary if you specifically need detailed outlines in your images. If you’re not familiar with ControlNet or if the lines are too prominent for your liking, you can leave this turned off for now. This is more for advanced users who want to play with line-based adjustments.Enable Depth ControlNet 🚫
This add-on adds depth to your image by simulating a 3D effect, making objects in the background appear further away. If you want to create more dimensionality in your compositions, try turning this on. Similar to Canny ControlNet, this is more beneficial for advanced users experimenting with spatial depth.
Make sure you have the Flux Depth ControlNet and Flux Canny ControlNet placed in your "ComfyUI/models/xlabs/controlnets" folder.
If the "xlabs/controlnets" folders do not exist you can create them and place the files there.
Here are links to the ControlNet files.
Flux Depth ControlNet: https://huggingface.co/XLabs-AI/flux-controlnet-collections/blob/main/flux-depth-controlnet-v3.safetensors
Flux Canny ControlNet: https://huggingface.co/XLabs-AI/flux-controlnet-collections/blob/main/flux-canny-controlnet_v2.safetensorsEnable LoRAs 📚
LoRAs (Low-Rank Adaptation models) are custom-trained models that can change the look and style of your images. For example, if you have a specific style LoRA loaded (like Studio Ghibli), it will influence the overall style of your output. Keep this enabled if you want to apply a specific stylistic model to your work. For those without LoRAs, it’s safe to leave it disabled.Enable Prompt Replace 🔄
This is a handy feature for customizing your prompt inputs. It lets you quickly swap certain words or phrases in your prompt based on pre-set keywords. For instance, if you use “color1” in the prompt, this add-on can replace it with any color you’ve assigned. This is great for testing different color themes or keywords without changing the entire prompt manually each time.Enable Detail Daemon 🔥
The Detail Daemon enhances finer details across your image, making it richer and more dynamic. It’s one of our favorite add-ons for giving that final touch of quality to images, so we recommend keeping this on by default. In our experience, it improves output quality in about 99% of cases.Enable IMAGE DIMENSIONS 📏
This add-on lets you specify custom dimensions for your images. It’s especially useful if you need to create images in a specific format or size for a project. By enabling it, you’ll have more control over your final image’s resolution and proportions.Enable MASK OPTIONS 🎭
Mask Options give you the flexibility to apply masks to certain areas of your image. For example, you could use a mask to selectively enhance one part of the image without affecting the rest which is useful for face swaps and img to img generations. If you’re working with complex compositions and want to focus edits on specific regions, turn this on.
Each of these add-ons can be enabled or disabled depending on your needs. Start with the ones we’ve highlighted as beneficial for general use, and feel free to explore the others as you grow more comfortable with the workflow!
Breakdown of Key Sections
LOADING YOUR IMAGE & PROMPT SETUP
Once you have your base image and checkpoint, load your LoRA models if desired. Don’t worry if you don’t have any; just turn them off to use the base model. By default, there are 3 that operate on the toggle switch from the add-on picker, but you can also disable individual LoRA loaders by "left-clicking" on them and then pressing "ctrl + b". We’ve even included a prompt replace feature where you can swap in custom colors or titles, making your prompts adaptable on the fly!PROMPTING, CONTROL NETS & ENCODING
In Step 2, set your main prompts, like lighting or tags, for extra flair. Turn on ControlNets to apply effects like canny or depth if you have these models installed. Play around with encoder options for unique results – we’ve found that adjusting these can really change the look and feel. Our preferred values are between 0.30 - 0.80 before the image becomes unusable.LATENTS & SAMPLING
The “latents” section gives you options for how you want the image to evolve. For example, you can load a mask for specific character detection. This section also has some sampling settings to help fine-tune the final look, particularly with denoising, which is essential if you’re working image-to-image.SAVING & UPSCALING
Compare your output image with the original using the image compare node. From there, choose your preferred upscale model and enhance details, like facial features. At the bottom of each panel, we’ve included all the nerdy stuff for fine-tuning if you’re into optimizing every setting.ORGANIZATION & COLOR CODING
Keep things tidy with our color-coded panels, and download our exclusive Black Mixture spectrum for seamless organization. Your workflow will look as polished as your projects!
Thank You for Your Support! 🌟
Your support on Patreon is what lets us build and share resources like this. We hope you love this workflow, and we’re excited to see what you create with it. Let us know if you have questions, and happy creating!
Stay Inspired,
Nate Dwarika
The Black Mixture Team
Comments
Black Mixture
Hi Zoltán, happy to help! Send me a DM and we can troubleshoot what's going on with your setup.
Zoltán István Bíró
Hi, I need your help. I downloaded your free workflow https://www.youtube.com/watch?v=ITkvdxIfvj4 and I liked the Upscale solution so much that I subscribed on Patreon. But in Black Mixture V3, if I set the Denoise to 0.10 under the Upscale option, there is still a huge change in my image. But I would like to have as little change as possible, so that the picture is bigger. As for my settings, I have everything turned off except Upscale Enabled. What am I messing up?
Relatos para escuchar
Thank you so much. That explains a lot of something I was completely unaware of. I appreciate the thorough explanation. Very smart the idea for the colours. I noticed too that there are very few workable standard colours for the nodes and the custom colours are hard to keep consistent. Your solution is elegant, easy and quick. I look forward to keeping learning with your detailed explanation of the whole workflow.
Black Mixture
No worries, To answer your question about the "Black Mixture Spectrum.png" file: This file is our color guide for color coding the workflows. Since ComfyUI by default has a not so elegant color preset system for the nodes, we used our spectrum graphic with the eyedropper custom color selection to color the node steps. We included it here in case you wanted to expand on our workflows, while keeping the same visual color harmony. You can turn off this node or delete it as it's not needed for the functional operation of the workflow.
Black Mixture
Awesome to hear and thanks for letting me know! I will most definitely explain more in detail. To answer your questions about the checkpoint loader vs Unet loader: If you download the original Flux Dev models at fp8 or fp16, you can actually load them as checkpoints without a separate vae and clip loader. The FP8 version is ~17gb and the FP16 version is ~24gb. These were released as checkpoints by Black Forest Labs (original Flux developers). There are other versions of the same name floating around online that separate out the Unet model from the VAE and Clip. This is done for graphics cards with less VRAM and for people that have already downloaded large clip and vae files from other models. For example, the other most popular one is by Kijai which has the same name but much smaller file sizes Kijai's FP8 version is ~12gb vs the original 17gb. These smaller versions work on less powerful graphics cards but need the separate clip and vae files to work properly. If the model you downloaded is a smaller sized separated version, you would need a UNET loader for the model, and the VAE and DualClip loaders as well. Original Flux Checkpoints: https://huggingface.co/black-forest-labs/FLUX.1-dev Kijai Flux Unet: https://huggingface.co/Kijai/flux-fp8/tree/main
Relatos para escuchar
One small addition: What's the purpose of the colour spectrum image? Maybe you explained that in previous videos or posts, but I just got here and it left me quite confused about its usage.
Relatos para escuchar
Hi, Nate. I just discovered your channel and joined immediately your Patreon. It's not only the workflows that interest me, but to learn how you do it to improve my own knowledge so that I can adapt it to my own needs. Therefore, I would please beg you to explain more in detail the complex structure in general, such as the reprompting area, for example, or the advanced groups. I will be analysing them myself, but I am sure I will learn quicker with your explanations and will discover new opportunities if you lead us in that discovery and learning path. I have noticed that you are more creative than most other ComfyUI YouTubers, and that convinced me to join you here without any delay. Besides, I have a question regarding the checkpoint loading node used in your workflow. If it's supposed to be focused in flux, why use a checkpoint loader, instead of an unet loader? The checkpoint loader can only load adaptations of flux models that integrate the clips and vae, if I'm not wrong. But you show the original flux dev fp16 as the model being used, which I don't get. It won't load in my case. Or is it a compact version with the same name? I am yet quite new to this world and I am probably missing the reason. I love what you are doing and will expect with haste new ways to develop the creativity à la Black Mixture.
