Skip to main content ↑ ↓
Topic: Welcome to AI Image Central! (Read 11864 times) previous topic - next topic

Re: Welcome to AI Image Central!

Reply #135 –
New update:

- Added a new model called Z-Image to the Flux page.  The model uses Qwen 3.4b for understanding and is quite good.
- Modified the Flux page so that the sampler, scheduler, and max resolution are now changed when you select a model
   allowing me to customize these settings per model.

Recommend using 9 steps as loss of detail is noticeable at 4 (Big loss) and 6 steps. Best open-source model to date for most tasks. Recommend Flux Dev Turbo as second-best open-source model. Third choice is SD3.5 Turbo. If one model doesn't work, try another.

Re: Welcome to AI Image Central!

Reply #136 –
New update:

I have added another option to the Z-Image model allowing you to do full 1080p images.  This takes 2-3 minutes due to the high resolution.  You can also click the Upscale X2 option to increase to 3840x2160.

Re: Welcome to AI Image Central!

Reply #137 –
New update:

I have now updated all the other languages to the latest format with the Z-Image model.
I also updated my games so they also display in the user's language now.
Just a few small fixes left for the Arabic versions of the rules, about, and searches.

Re: Welcome to AI Image Central!

Reply #138 –
New update:

Added a link in the menu bar for the Reddit sister site.
I have modified the Z-Image model to use a better scheduler and increased the 8 step option to 9 steps.
I have also added a 6 step option to the Z-Image 1080p section.
 

Re: Welcome to AI Image Central!

Reply #139 –
New update:

- Added auto language translation to English for the Flux Fill model for all languages.
  Remember for Flux Fill: Mask the area of the image you want to change only. Prompt what you want changed in the masked area.

- Modified the Z-Image dropdown to reverse the order of steps.  The default is now 9 steps as 4 and 6 steps lose a lot of detail.
  It also works much better when adding text with the higher step count.  Unlike other models, the loss of detail is noticeable.
  In most use cases, this is the best open-source model available.

- Small fixes to slider controls
 

Re: Welcome to AI Image Central!

Reply #140 –
New update:

Some model's dimensions are hard coded to use values in multiples causing undesired resolutions.
For Flux models, trying to get 960x540 does not work as it has to be 960x544 as well as 1920x1080 has to be 1920x1088.
I have modified the code so this is now hidden behind the scenes and the images will be trimmed to the correct size.
960x540 is important because upscaling it x2 gives you 1920x1080 (1080P).

For Z-Image, you can also upscale x2 from 1920x1080 giving you the 3840x2160 (4K) resolution.

Re: Welcome to AI Image Central!

Reply #141 –
Looks like my 24GB GPU is fried. Since no one is donating I doubt I will run the site much longer as I don't have the money to replace the GPU.  I moved all the SDXL models to the working GPU so all the models are back to working.

Will leave up for now.  If you want the site to stay, consider donating. Link on rules page.

Re: Welcome to AI Image Central!

Reply #142 –
Site was down last week due to issues with ISP. Issue finally resolved.

 
AI Image Central - Welcome to our forum!