Hey guys. Like I mentioned in the last article, I’m back with the upgraded version of my workflow.

If you read my previous article “Basic Comfy Workflow – What I Use”, you already know what drives my workflow design:

Keep it simple. Get great quality. Don’t wait forever.

I know there are tons of ways to push renders further. Things like using multiple LoRAs, tiled upscaling, checkpoint swapping, Adetailer chains, image to image passes, and similar tools. They absolutely work and can improve results a lot. The issue is that most of them also increase generation time and complexity.

I was looking for a way to reach a similar improvement without turning every generation into a 15 minute experiment session.

Here is the TLDR of this workflow upgrade:

It is still the same simple structure from last time. I only added a few lightweight nodes that improve quality noticeably while barely affecting generation time.

---> Model Page <---

The only additions to the original “Basic Workflow” are

• Multiply Sigmas (Stateless)

• Enhance Detail

• Upscale Image (using Model)

• A set of optional resizing nodes

On my setup, even when using the upscale model, this adds at most 30 seconds to the total render time. The quality boost is much stronger than I expected.

⚠ This is not a magic setup or hidden trick.

It is a low overhead workflow upgrade for users, especially for lower end rigs, who want sharper and cleaner results without heavy post processing.

New Nodes

Multiply Sigmas (Stateless)

Adjusts the sigma curve during generation, which helps the model refine details more cleanly without increasing render time. It slightly improves structure and clarity in the final image while keeping the process just as fast.

Enhance Detail

Used after the upscale and after Multiply Sigmas. This node adds a subtle amount of noise and sharpening, which introduces a light grain effect. It helps recover clarity that can get slightly lost during the quick upscale. Since model upscaling tends to soften or smudge fine elements, Enhance Detail counteracts that and brings back definition and texture.

Adds only a few seconds to the render and gives a noticeable bump in detail.

Upscale Image (using Model)

This upscaling method is fast and the results depend on the model used. I personally use a 4x upscale model, which multiplies the current image size by four. It is more than enough to boost detail, texture and overall visual quality.

Sometimes it can be a little too strong, which is why I do not use it on every render. When I do apply it, I almost always resize afterward. In my experience, an image that is upscaled and then reduced looks better than a render generated directly at the same final resolution, but that is a personal preference and you should judge for yourself.

Most of the time I generate using one of my usual formats, such as 1024×1536 or 896×1600, then upscale, and finally resize to a slightly larger size than the original but still lower than the full 4x result.

Resizing Nodes (Lanczos)

These are simply optional finishing tools to get the exact output size you want after upscaling. I use Lanczos because it keeps the best sharpness and visual integrity during downscaling.

There are three types in the workflow:

• Resize Image With Resolution
  Forces a specific resolution.

• Resize Shortest To
  Sets the shortest side to the chosen value and keeps aspect ratio.

• Resize Longest To
  Sets the longest side and keeps aspect ratio.

I use them depending on the source image and what final format I want. Most of the time, after upscaling, I downsize to something slightly larger than my original generation size for a cleaner final result.

These nodes add no noticeable time to the render and simply help finalize the image.

Here are comparison examples using my previous basic workflow (Vanilla Flux) versus the upgraded version (Raz’s Workflow). I will also include the images in the attachments section so you can see the comparison without any video compression.

The workflow contains notes and my own preferred default settings for the nodes. These are simply what work best for me based on my testing and rig, but they are not meant to be followed strictly. I strongly recommend trying different values, pushing them further or dialing them back, and seeing what fits your style. If you find settings that perform better for your own workflow or hardware, feel free to share them so others can benefit too.

That’s it for this update. Thanks to everyone who followed the first article and especially to those who suggested improvements and pushed me to tweak the workflow further. Your feedback helped shape this upgrade and I hope it gives you better results without slowing you down. Feel free to share your own adjustments and ideas if you try it out.