Discord to try the model with my bot and my settings. LCM & Lightning LoRas available.
For Creative and Colorful content. Unleash your imagination !
----------------------------------------------------------
--This model is a Fork of ZavyChromaXL at v1.0 --
----------------------------------------------------------
Realism is not the purpose of this model.
Still the very first versions with many errors, please consider this as a fail until V1.0.
Very sensitive to the prompt tokens. Result may vary. 🙃
To use Highres_Fix is the way to use this model, and more generally SDXL.
Eventually Adetailer.
I don't recommend TensorRT, nor any "optimization", to maintain the maximum level of quality.
LCM-Lora is pretty funny with this model (closeup, portrait..). With Euler a or the LCM sampler.
Settings:
As usual, it's subjective. SDXL works the same for everyone. It's a story of expectations.
My settings :
Negative prompt: (NSFW), (nude), (naked), worst quality, blurry, low quality, lowres, watermark, out of frame, overly saturated, disfigured, ugly, bad, overexposed, underexposed
Steps: 30
Size: 768x1280 + Hires x1.4
Samplers: DPM++ 3M SDE Exponential, DPM++ 2M Karras, Euler a
- Highres_Fix:
Hires steps: 15 (Half the Steps)
Hires upscale: ~1.5
Hires upscaler: 4x-UltraMix_Balanced (You need any custom upscaler when using DPM++)
Denoising strength: 0.45
Come on my Discord or Zavy's Discord for more informations.
Cheers ! 🥂
Description
UltimateDiffusion v0.1
FAQ
Comments (4)
hey freak
i see, you love your self and put 20 images as own post , my god you you must be bored for a long time :D
Hey peeps !
Nothing tremendous. Just an update as new basis for further training.
I disabled the on-site generation as the model is not good enough yet, nor the generator.
Hires is the minimum requirement for a decent result.
Some prompts are better using DPM++ 3M SDE Exponential, some others using DPM++ 2M Karras...
Join the Discord if you wanna generate.
Cheers ! 🥂
Love your model I think its Unique in its own way.
looking forward to your other models
I ran some tests comparing v0.2 to v0.5 and there's almost no difference in the results.
Can you detail the difference between 0.5 and 0.2?
I'm curious to how it's so similar if you rebased it.
I don't mean to offend, I'm genuinely curious to learn more about the process of finetuning base models.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.