Version 2 (SHOW MORE)
Checkout: ClearRealismMix-European
Changelog
Version 2 incorporated LEOSAM's FilmGirl Ultra model to add a refined Spin Diffusion model. On top of this multiple trained LoRAs were included.
INFO
The goal of this checkpoint is to create high quality realistic images for use with & without my LoRAs, existing and future.
Suggestions
NO BAKED VAE - Please use vae-ft-mse for base checkpoint and vae-ft-ema for inpainting
Use the above mentioned embedding to reduce asian facial features
Use to create realistic person images
CFG Scale between 5 and 10
DPM++ SDE Karras / DPM++ 2M Karras / DPM++ 3M SDE Karras Samplers have been tested - other should work but have not been tested by myself.
Recommend using 25 - 40+ sampling steps
Use ADetailer to enhance facial feature quality - Not need so much on close-up images but mid and wide shots it will improve quality
Please share any images you're happy with while using this checkpoint and review
Check the Suggest Resources!
V2-Inpainting Model Preview Differences:
preview 1: Face replace
preview 2: Changed hair, changed dress, changed earring, changed eyes, changed lipstick
V1-Inpainting Model Preview Differences:
preview 1: Face replace
preview 2: Added belly chain, removed palm tree,improved right hand
preview 3: Changed dress, added choker,removed hair on middle of forehead
preview 4: Changed hair colour, changed eye colour
Description
Version 1
FAQ
Comments (5)
I try to recreate the red girl standing but i get 1 girl standing and other girl with face and upper body in the same picture. Do you know why?
Hi, did you copy the generation data from the image? If so make sure you have UnrealisticDream, badhand, & deepnegative embeddings.
If you're using a different seed you will get differences in output. Also can share the image, with the generation data, and can have a look at any differences.
maybe the resolution? happens in most 1.5 checkpoints at 1024 resolutions, try 512 or 768 instead
Where can I find the ClearNegative embedding?
I haven't released the embedding yet, but as it currently exists it makes very little difference having it included in the negative if you have the rest.
I want to try create a more custom embedding togo along with the model which will require generating some mangled stuff & training. Soon! ™
Edit: Quick example: https://imgur.com/a/zrOfZ4Q - same settings and wonky seed generating two handbags.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.