omni-diffuse
SD checkpoint
·
Uploaded Feb 19, 2024
·
Used 485 times
About this version
For best outputs:
Sampler: DPM++ 2M Karras, Euler a
Steps: 30 ~ 40, cfg around 8, CLIP skip 2.
VAE: vae-ft-mse-840000-ema-pruned_fp16.safetensors
Common pos starting: "((Masterpiece)), ((best quality)), ((professional photo)), ((perfect anatomy)), ((best light))", adding (optional but recommended) the LoRa's "FaceBeauty_qinglong_V3" (weight 1.0) and "add_detail" (weight 0.7), both included in the example runs.
Common neg: "low resolution, ((mannequin, silicon doll, render)), childish, explicit nubile, painting, digitalart, poor collage, photoshopped, distortion, dysfunctional, augmented, detached, ripped, minor background, easynegative, boring_e621_v4, negative_hand-neg", being the easynegative, boring_e621_v4, and negative_hand-neg embeddings included in the example runs.
Note: this model tends to create nudes if u "forget" to cloth the people. Make sure to add "nsfw, nipples" ect if u want to create sfw outputs.
The model is actually quite obedient and listens perhaps too well, so describe and add weights wisely.
About this model
omni-diffuse
Related Posts