β‘ What Mega Flux Can Do β ControlNets, Pulid, Redux, and More
Added 2025-06-21 19:51:14 +0000 UTCLet me drop some examples from using the Mega Flux exclusives.
My prompts were very simple, so I didnβt include them to avoid clutter.
π§ Depth:
Flux Depth uses the depth map of an image, letting you prompt for what to fill it with. It's great for keeping the spatial layout while restyling the image however you want.
βοΈ Canny:
Flux Canny uses an edge map and also lets you prompt freely. Itβs especially good for repurposing parts of an image like the planet in the first example.
πΊ OpenPose:
Self-explanatory. Itβs great when you donβt want Depth or Canny getting distracted by other elements in the original image.
π« Pulid Face Transfer
Pulid lets you use a face as a reference and prompt as usual. The result will keep a rough match of the reference face. Itβs pretty solid without needing any training, though not perfect. As you can see, it struggled with Aloyβs hair.
π Redux
Use the style of one image to influence another.
π€ What Else?
Thereβs also Blend Images, which combines two images into a new one (the dog reference image for Redux was made with this). Then thereβs Face Swap and Face Restore.
If you want me to show those too, drop a comment. But for now, Iβll leave it at that and let the images speak for themselves. π«‘
β‘ Want the Workflow?
Everything you just saw was done using the Mega Flux workflow.
It stacks Level 1 and 2, adds powerful tools, and comes with a full one-click installer.
If you want it, itβs available to Power Users who directly help me keep building all of this by supporting me. π
Comments
There are a lot of questions here, so Iβll keep this brief to avoid essay length. π Face swap with reactor nodes mostly ignores hairstyles and just focuses on the face, though results can be hit or miss. Adetailer can also be used for face swaps by loading a character LoRA and trigger words. I havenβt tested it, but youβd likely need a high denoise so it fully changes the face. For training, I recommend AI Toolkit since the creator is great at keeping up with new models π https://github.com/ostris/ai-toolkit. Images get scaled and bucketed, so you donβt need to stress about resolution. Just avoid very small ones. For Flux, everything scales up to around 1024, so a 200Γ200 image could teach pixelation. For poses, keep the subject consistent but vary backgrounds, clothes, and objects. That way the model learns the actual pattern you want. LoRAs are basically small tuned weight files. In Comfy, adding them through a LoRA node just patches those weights onto the base model. The goal is simply teaching the model something it doesn't already know. Itβs a lot, and I didn't cover everything, but hopefully itβs a solid start. π Happy to clarify or dive deeper.
Max
2025-08-21 20:51:55 +0000 UTCFaceswap would be very interesting. How can i get the same "face" while using different hair styles and poses? And another interesting topic: LORA. How to train a Lora. What Input Images do i need, what resolution, what poses / forms / objectives do i need? What does a Lora realy train and how do i get what i want? Do i need an detailer befor or after the Lora trainer?
NadiaNovak
2025-08-21 09:23:46 +0000 UTCI want the workflow bro I have subscribed to your channel I was waiting for this workflow
Kishore Shopify
2025-07-27 19:36:34 +0000 UTC
