Life Like Diffusion V3 is live. I have been working on this update for few months. Now I feel like it is ready so publishing it.
Comment, explore and give feedback. This version adds better faces, more details without face restoration. Face restoration is still recommended. This model as before, shows more realistic body types and faces.
My SDXL model will also be coming out soon, Check my profile this weekend to try them.
If you don't have a powerful enough GPU, or Not interested in installing automatic1111 locally then checkout mage.space
Life like diffusion is also now on Mage along with other top 70+ AI models as well as 4,000+ LoRAs and Textual Inversions. It is only $15 a month for Fast & Unlimited generations. Check my model and others here - https://www.mage.space
Supported ethnicities - Native American, Indian, Arab, Hispanic, Latino, South Asian, Turkish, Black / African, Korean & Chinese
This model performs exceptionally well at creating images of people. This model is focused on depicting humans as they are, with all their imperfections. It also performs well at producing images of people of all ethnicities, unlike other models. All Ethnicities, I believe should be represented in Al world also.
Check out these articles from Vice and NBC. We need more models that supports people of all genders, ethnicities etc in our models to offset this issue.
READ MORE HERE BELOW
Vice - Click Here
NBC - Click Here
If you are having any issues with this model, Please try the smaller Pruned Version.
This model generally performs much better at higher resolutions. Unconventional resolutions can also lead to better results.
Here is an example Prompt but you don't have to use this, any prompt that emphasis photorealism should perform well with this model
((instagram photo)), ((Full Shot)) full body photo of 28 year old beautiful ((woman superhero costume)) strong girlboss in (( avatar the last airbender movie)) (high detailed skin:1.2) 8k uhd, dslr, soft lighting, high quality, film grain ((amateur photo))
small-pupils, small-eyes, closed-eyes, ugly-eyes, multiple-navel, multiple belly-button, ((lazy-eyes)) (deformed iris, deformed pupils, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime:1.4), text, close up, cropped, out of frame, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck, closed-eyes, ugly-eyes, ((lazy-eyes))
Tip :- To improve realism and add more skin texture, try lowering denoising in high res fix between 0.25 - 0.45 instead of 0.7, select upscaler to None or R-ESRGAN 4x+ and upscale by between 1 and 2. This should help you increase the resolutions, reduce the number of artifacts and give you high degree of details and skin texture. Use this calculator to figure out correct resolutions to use https://preyx.github.io/sd-scale-calc/
Sampler - Euler A
CFG Scale 7-9
Hi res fix - ON (https://preyx.github.io/sd-scale-calc)
0 Hires steps ( same as sampling step)
Denoising Strength between 0.3 - 0.4 (very important to keep this under 0.48 in automatic1111, especially when output resolution is higher)
Creative ML license prohibits Producing illegal and harmful images. This includes but not limited to underage content and more. Please checkout the original SD license to learn more. Do Not use this model or any SD models to create illegal underage content !!!!
I also want to make it clear that this particular merged model should not be used for training or creating images that are based on portraits of famous people such as celebrities and public figures .The reason for this is that such usage has the potential to cause controversy and negative impacts within the AI community. Therefore, I kindly request that you refrain from using this model for such purposes.
No additional training material was introduced to create this model. Since it is a merged model, it utilized some openly available models created by others from civitAi community to create this model. Again, No additional training data has been given to make this model.