Opacity issue with generations using trained LoRA

[Saquib764]

I trained two LoRA, and used the output model to inpaint. The trained LoRA is a character LoRA - girl and cat in the image.

I am trying to inpaint each character by masking them one by one. However, overall image seems to get dull(or faded) after each generation.

In the image, first image is the latest generation.

Input looks similar to this-

{
  "image": "https://artistic-qr.s3.us-west-2.amazonaws.com/thefluxtrain/6a9ac40e-67a5-4ec8-a4ab-075ac13fcac2/f3cuwv.png",
  "model": "dev",
  "width": 512,
  "height": 512,
  "prompt": "The cartoon girl balikavadhu is sitting on a park bench, looking lovingly at her cat catvadhu. The cat catvadhu is curled up next to her, nuzzling against her in an affectionate manner. The girl balikavadhu has a warm smile on her face as she reaches out to stroke the catvadhu's fur. catvadhu's fur is fluffy and has a curly texture around the face, with a prominent tuft of fur at the top of the head. The background shows a peaceful park setting with trees, flowers, and a clear blue sky. perfect fingers.",
  "extra_lora": "https://replicate.delivery/xezq/0JdQvdL98LIXKRqof9qZOf6705eGaaaS4NrSzorlA2w5UTgnA/trained_model.tar",
  "lora_scale": 1,
  "aspect_ratio": "16:9",
  "output_format": "png",
  "guidance_scale": 4.12,
  "prompt_strength": 0.7,
  "extra_lora_scale": 1,
  "disable_safety_checker": true
}

[Saquib764]

Probably the issue might be related to diffuser's inpaint pipeline.

huggingface/diffusers#9867