10 Days with Z Image Turbo: My First Impressions

 It's been a while since I last posted here.

I had a bit of a scare over the past week because my Google account login got messed up, and for a moment I thought my whole blog had been deleted.

But I'm back.

I've been learning ComfyUI, but in the end I couldn't get FLUX running the way I wanted because my GPU just didn't have enough VRAM.

Then I recently discovered a workspace called WebUI Forge Neo, and that pretty much meant I didn't have to keep digging into ComfyUI anymore.

(Yippee!)

A lot of the old WebUI extensions aren't compatible, which is a little inconvenient. But after setting it up with only the extensions I actually need, I was finally able to use the Z Image Turbo version.

As of the day I'm writing this post, I've been using Z Image Turbo for about 10 days.

One thing I've realized so far is that Z Image Turbo isn't always better than Illustrious or NoobAI.

(That said, I'm just happy I can finally run it locally.)

So the reason I'm making this post today is to share a few of my results.

The first set of images compares artwork I originally generated with SDXL back around 2024 against images generated with Z Image Turbo using the exact same prompts.

As you can see, the SDXL images above couldn't produce the results I was aiming for because of poor character separation, color bleeding, and incorrect scene composition.

The images generated with Z Image Turbo aren't perfect either, but you can see that they capture my intended results much better than the SDXL versions

The next set shows images that were originally created with SDXL and then run through Z Image Turbo using img2img.

It looks stylish, and the results are pretty solid.

Much better.

You could say I caught a glimpse of how some of the limitations of local image generation models can be worked around.

P.S

A few images worth sharing