For best outputs:
Sampler: DPM++ 2M SDE Karras / DPM++ SDE Karras.
Steps: 40, cfg 8, CLIP skip 2.
VAE: for v3.0 -> blessed2_fp16.safetensors / for v3.5 -> none required.
Pos prompt: for v3.0 -> add the LoRa's "more_details" (weight 0.8) and "FaceBeauty_qinglong_V3" (weight 0.9) at the end of your pos. For v3.5: nothing in particular, just what you want to see. Quality words like "Masterpiece, best quality..." ect do enhance overall quality and detail.
Neg prompt: too long to be pasted in here, but feel free to use my example runs'. The embeddings "easynegative", "bad-picture-chill-75v" and "bad_prompt_version2" are included in my example runs for both versions.
What changes from v1.5 to v1.8: input_image feature.
Checkpoint Merge.
For best outputs:
Sampler: DPM++ 2M SDE Karras / DPM++ SDE Karras.
Steps: 40, cfg 8, CLIP skip 2.
VAE: for v3.0 -> blessed2_fp16.safetensors / for v3.5 -> none required.
Pos prompt: for v3.0 -> add the LoRa's "more_details" (weight 0.8) and "FaceBeauty_qinglong_V3" (weight 0.9) at the end of your pos. For v3.5: nothing in particular, just what you want to see. Quality words like "Masterpiece, best quality..." ect do enhance overall quality and detail.
Neg prompt: too long to be pasted in here, but feel free to use my example runs'. The embeddings "easynegative", "bad-picture-chill-75v" and "bad_prompt_version2" are included in my example runs for both versions.
What changes from v1.5 to v1.8: input_image feature.
Checkpoint Merge.
