Pinned Tweet
Keros is exploring the parts of the models that no one else is. Our math-based techniques provide endless control and variation without changing the creativity of the model like fine-tuning.
We are in alpha and if you'd like to contact us, DM us or visit keros.ai
Keros.ai has developed semantic sliders for diffuser models! No fine-tuning required. Just type in a concept and use instant sliders to guide the model towards or away from an idea! Created with @StabilityAI under the hood. #AIart #SDXL
Use our sliders to create a style and prompts are highly consistent with no fine-tuning required! #AIart
Tired of plastic-y airbrushed "Stable Diffusion face"? The model can do better! It just needs a little bit of help from Keros. #stablediffusion #AIart
1
71
We begin where Stable Diffusion ends. We applied new math to a single text to image generation with sliders.
Here are just two of the vectors we've discovered hidden in the Stable Diffusion model. Richness and contrast (with high texture). #AIart
Keros AI accepted into the NVIDIA Inception program for startups. Yeah! Ok. Good start. Let's go! 🤓
nvidia.com/en-us/startups/
Keros has a new site up! A brand new way to play with AI models. Come explore latent space with us. 👾
keros.ai/
Keros retweeted
Thanks to @prof_sinister I have been testing out a new product (Keros) that allows some very fine-tuned control over SDXL output. Same prompt, seed, strength, and number of steps for each of these.
Keros retweeted
That's what I call exploration :D
the same prompt and seed,
different settings
and that's <0.1% of sensible settings space
p: "soul within the code, glitchcore, AI singularity by Emil Alzamora by Hirst"
n: "sculpture, monochrome"
Keros retweeted
Has anyone seen a tool for Stable Diffusion that can do what variations (subtle and large) do for Midjourney?
When you've got a good SDXL image and you want subtle variations around it.
Keros retweeted
sometimes small differences in model capacity give a great time boost
e.g. fact that in most situations fingers and hands are made properly ;)
#SDXL by @KerosGaming
#gamedev needs control over its outputs
We are working hard on adding additional control over IMG generation
other than words or controlnet
One that will keep full knowledge of the model
But ads new knobs in artists hands
Thats a consistent result
without any post-processing or dreambooth
on the guidance scale 19
on vanilla SD1.5
Who has a prompt they would like to check?
3
147

























