@Whistler_ai Your an amazing lora creator.

@Whistler_ai As will you make an amazing models.

Can't wait to see your loras once you finally make one. For now, I will have to be content with your amazing checkpoints and wildcards. You're (you are) very good at those for sure! Jury is still out on your loras though!

@Whistler_ai your welcome bro, about the v2 model I still didn’t test it a lot yet, but I already did a simple comparison between them. I used 10 different prompts, each prompt focuses on something like small and tiny details, skin, poses, colors and others, and I found them almost close in the results.

the points where V2 is better than V1 are the lighting, dynamic poses, and small fabric details, and the colors are more vibrant.
V1 was better in body details especially the fingers, V1 didn’t have problems with fingers unlike V2, and depth of field shots are better on V1 too.

for your information I used both models without any loras, I only used lazyneg in every prompt, I didn’t add anything else because I wanted to see the raw difference between them. maybe V2 becomes much better when using embeddings and loras, so this is just my view after a simple test between them, and I only tested V2 in sfw, so I will see more in the future after working with it for a while just like V1.

@Whistler_ai Sorry, my English isn't the best, but this is great. Very good work. I'll be excited for a new checkpoint.

@Whistler_ai on site, yes, I tried many different setting , but just with a simple k sampler and a lora character , sometime its not working and sometime yes ^^

I was about to say the same thing. I am using comfyui with mine. Run the other one and it has issues run this one and its black screen in the ksampler. I am still seeing if I have a weird setting somewhere. Going to run more tests with it because it is weird.

@Whistler_ai  I have been cleaning up the workflow as it is very long and chaotic. I found a Lora that was causing some issues with the generation but not the Black Preview. I cannot replicate now on my image generation at all. But the up-scaling workflow I can if I change the clip from -2 to -1 bingo it always fails to render image just a black screen.

Also my CFG was way to high. It was a 7 and I normally run it at 1.6.

Also even though my Control_Nets I was using worked they are not for SDXL/Illustrious based Checkpoints. I swapped the Canny and Open_Pose to illustrious ones and it is a cleaner image now on that.

My suggestion for Tlood is to check his CFG, clip, Loras and VAE to ensure they all play nicely with Illustrious.

Check the Logs from Console to see if you are getting a Lora Diffusion model error. (How I found out my Lora was a pain and not doing anything)

Also needing to remove some Custom Nodes I was using that are outdated. Comfy has some now that replace it built in. So Bonus.

I can still send you @Whistler_ai The Upscaler IMG2IMG Workflow if you want.

I am still learning Comfy and all the things it can do.