z-Image Turbo directe support NSFW
Hi Phr00t ，Pls give up Qwen Image and switch to z-Image. We can waiting for the z-Image Edit version.
Z-Image is useless to me while it does not provide edit capabilities. Being able to use images alongside a prompt is just too useful and I'm not going back.
Qwen Image Edit is an amazing Text to Image and Edit model wrapped in one with good LORAs already available. Z-Image is pretty neat, but I still far prefer Qwen Image Edit. We'll see if Z-Image Edit takes the crown, but Qwen Image Edit 2511 is right around the corner.
Also, Z-Image NSFW capabilities are quite poor compared to Qwen Image Edit with LORAs. The "base" Z-Image NSFW capabilities are better than expected, though.
Haha I was about to ask you the same request
Z image is releasing there image edit model soon
I'll be definitely evaluating z-image-edit. However, z-image-edit is a much smaller model. Qwen Image Edit may still provide more quality and versatility for just a few more seconds per image. I don't really want to maintain two edit models and I'd prefer to pick the overall "best" one. We'll see, but my money is on Qwen Image Edit just due to its size (although z-image has Qwen 3).
Yeah can't wait to see what Will come about with z-image edit. Qwen is amazing, but the speed is yuck
I want to generate Audio for the contents I generate.. Which model would you suggest Phr00t?
I want to generate Audio for the contents I generate.. Which model would you suggest Phr00t?
Hello, I am not Phr00t, would you mean cloud19's MMaudio fine-tune?
Yeah can't wait to see what Will come about with z-image edit. Qwen is amazing, but the speed is yuck
I can create images with Qwen Image Edit in about 20 seconds. Z-image is probably in the 5-10 second range. Z-Image is definitely faster, but both are still "within seconds" for it to matter much relative to quality and versatility, imho.
Yeah can't wait to see what Will come about with z-image edit. Qwen is amazing, but the speed is yuck
I can create images with Qwen Image Edit in about 20 seconds. Z-image is probably in the 5-10 second range. Z-Image is definitely faster, but both are still "within seconds" for it to matter much relative to quality and versatility, imho.
On my hardware qwen image edit is 130s per step. Z-image is 10 seconds per step at the same resolution.
Regular qwen non edit 44s per step.
Yo tell me the best audio generator.. For Vo and SFX, music
Edit: ignore what I said below. Upon reading the actual model card again I can see that the lightning lora in use here is the 4-step v2.0. I just tested this against the actual edit-1809 4-step and this seems to be a big improvement. It will be interesting to see how z-image and upcoming z-image-edit compare once it's hit with something like the snofs dataset.
The problem to me is in photo-realism and I think once you introduce any of the lightning loras, either 4-step or 8-step, the realism turns to plastic. Almost all outputs of qwen-image-edit-2509 + lightning look like illustrations to me. I don't think this is being introduced by the SNOFS lora. @Phr00t I'm curious how far you've dug into this. I seem to remember someone asking about the list of loras you baked in; I need to find that again. I have found that introducing the SamsungCam UltraReal lora on civit can restore some realism. What I am seeing from z-image turbo is worlds beyond qwen-image-edit-2509 for gen. Also, I'm using edit-2509 for gen just without an input image.
(Please forgive this partially baked comment as I really want to look at your lora list now.)
Yo tell me the best audio generator.. For Vo and SFX, music
It seems I didn't get it, I thought you wanted the MMAudio fine-tune.
Ace Step is the first thing coming to my mind
Were to find the workflow?
I better go with Kijai/MMAudio_safetensors what say?
experimentation is healthy is this context. experimentation is healthy in this context. Phr00t
experimentation is healthy is this context. experimentation is healthy in this context. Phr00t
Indubitably.
I tried to improve realism in the v12 merges. Check 'em out.
experimentation is healthy is this context. experimentation is healthy in this context. Phr00t
Indubitably.
I tried to improve realism in the v12 merges. Check 'em out.
V12 is shifting identity again.
It also brought back the grid lines, and changes poses even when strictly told not to. 11.4 has none of these issues apart from increased saturation and contrast.