Video - https://youtu.be/Wzg0tv8oxig

Overview

The latest updates to ComfyUI include a range of templates for INT8 models, so I decided it was time to dive in and see what all the fuss was about. Glad I did, as I was very impressed by both the speed and accuracy!

Integer maths has to be less accurate than floating point... right? Well, thanks to the convolutional rotations, it seems then when it comes to models you can use in ComfyUI, FP8 is dead. Long live INT8 Conv-Rot! (ref: Wikipedia for those who haven't heard the phrase before). For those of us with 3000 series cards (like me!), INT8 is extra great - but all 3000, 4000 and 5000 series card owners should benefit.

If faster and better is what you're into, then it could be time to delete all those FP8, FP8 Scaled and MXFP8 models you have and replace them with their INT8ConvRot alternatives where available.

Workflow Information

Being a workflow rather than any special coding, it will work with both the new INT8 and older FP8 models as ComfyUI's nodes now load all the various versions happily - be it BF16, MXFP8 or INT8-ConvRot.

To load the workflow, drag the .json file into Comfy, or use File -> Open.

Model Files

Comfy has a whole bunch on INT8 models available now, with download links available in-workflow:

I've been using the workflow to test the new versions of the KREA-2, Qwen Image Edit 2511, Z-Image-Turbo and Boogu Edit Turbo models.

Custom Nodes

No new customs compared to last week. As usual custom nodes for ComfyUI can be installed using Manager or the normal git clone commands:

A standard local install of ComfyUI is suggested, and you'll need the very latest version for the current level of node support:

Do NOT enable the Nodes 2 Beta as ComfyUI doesn't work well in that mode... yet!

Model Loaders

The model loaders have various model-specific nodes and options such as LoRA loaders:

Remember, of course, that any model supported by Comfy can be used!

Unlike my previous workflows which just had options at the refiners stage, any of the various models can be used at any stage in image generation - start, middle or end.

As with the refiner model selector in previous workflows, you can select which model you want to start with:

You can, of course, change the model loads to whatever your favourites are, but the default options can be seen both in the note and via the Lenient Switch labels:

This was lots of fun to make as I needed to turn my brain on for a moment in order to use the lenient switch such that it could handle all the various modes (in the "outside the map" engine area):

Qwen Image Edit & Krea-2 work well with the Qwen Image Edit Plus text encoder, but Boogu Edit has a different one, and Z-Image only really works well with the standard CLIP Text Encode!

Luckily, Comfy comes with a nice Compare Text node which outputs boolean values:

Unlike the typical Rgthree Any Switch, the Lenient Switch can evaluate boolean inputs, making it a piece of cake to correctly switch tracks based on your model selection.

Does the label say Boogu Edit? Then use the first track, source_a. Does it say Z-Image? If so, source_b is the way to go. Neither of those? Fine. The Qwen Image Edit track via source_c is what we need.

Now all you need to do is select A, B, C or D as usual and the switch does the logical choices for you. Handy!

Control Panel

As usual, the control Panel is where you can easily enable or disable the various groups:

Note that some options only work with some models - such as the Krea-2 Conditioning Rebalancing.

If you are loading images, don't forget to enable "Encode Images?" if using Qwen Edit, Krea-2 or Boogu Edit if you want the model to "see" the images.

Z-Image can only do IMG2IMG and In-Painting, so using the Latent Option of "Image 1" and lowering the Ksampler denoise value, or "Masked Image 1" are the only options when using images that model.

Check the in-workflow notes for more information.

Image Loading

You can load up to 3 images:

Boogu Edit has all 3 images attached, but the ComfyUI backend code apparently doesn't quite work with more than 1 image - yet. Once it does, you'll be able to use multiple images without changing the workflow.

Resize your images as you'd like using the Resize Image V2 node - it has a lot of options!

Qwen Image Edit can use things such as pose, depth and canny images which can be created via the All-In-One Aux Preprocessor options.

If you want to use Automated rather than Manual masks, flip the switch and enter your auto-mask prompt. The threshold and iterations values can be adjusted too in the SAM3 Detect node.

Prompting

Positive and negative prompt options are available:

I mostly use the turbo models at CFG1, so the -ve prompt isn't used. You can load non-turbo models and/or use a different CFG to use this if you like.

KSampler

The KSampler features Detail Daemon for some fine-tuning goodness:

With the detail_amount at 0 it is basically disabled, so increase or decrease that value as you desire.

Output

The output section features 3 options, the main one being the standard, default VAE output:

If using Qwen or Krea, then you can get a 2x upscale via the 2x VAE output:

Or, for 1-step 4x output, the INT8 version of SeedVR2 is the 3rd option:

Refiners

Refiners 1 & 2 give you your choice of model for the upscale / refiner passes:

Change the Ksampler settings for either refiner to your favourite settings:

The default values are very light, keeping images similar to the original result. Different models also work differently given the step count, sampler, scheduler and denoise values.

Refiners also feature automated detection for Edit models:

This means if you use Qwen Image Edit to start with, and the also choose Qwen Image Edit again in either refiner, the original conditioning will be used in order to maintain likenesses, poses etc from your initial result.

Re-Prompter

If you started with an Edit model, such as Qwen or Boogu, then you may wish to use the Re-Prompter:

For the most part, even if you have an editing prompt such as "Remove the cat on the left" then low denoise and low step refiner passes will still work OK. If you find your editing prompt causes problems with your chosen refiner settings, the Re-Prompter can be enabled and hopefully that will work better!

Ultimate Upscale

If you're looking to do some minimal refining, then the Ultimate Upscale group is great:

Image Compare

Compare images to more easily see any differences. As with any group, you can change the inputs in the GET nodes.

Enjoy! :)

Update: 19/7/2026 - Uploaded v5a without the Krea-2 Text Enhancer node as recent ComfyUI updates have rendered it non-operational. The change also impacts NegPIP, but that is bypassed by default anyway.

22/7/2026 - The Krea-2 Text Enhancer node has been fixed, though NegPIP remains non-operational.