USS Anchorage (Azur Lane)版本Release V2.0 (ID: 21329)

Updated Dec/11/2023:

The 3rd release version is here, trained on animefull-pruned with Prodigy optimizer and updated & refined dataset.

Her long deserved new costume is hereby included!

The inflated orca has earned my personal distaste though, that thing alone almost ruined the entire training

---

Features USS Anchorage from Azur Lane in her original costume & Dolphins and Swim Lessons swimsuit costume (legacy v1 & 2 has only the original costume and was trained on ACertainModel).

Usage:

Works fine on full 1.0 weight, albeit stabilizing her bandeau-shirt inner composition could take a bit prompt engineering.

For each costume's prompts, see Trigger Words section.

Known Issue:

Swimsuit costume slightly biased towards a blue coat or dark background, due to the existence of an inflated orca.

---

Previous Update Mar/11/2023:

Tried out the LoHA (LoCon, aka LyCORIS as per its creator's preferred but eccentric name) format for Anchorage's update. Quality is indeed impressive and switching to this training method takes surprisingly little time to adapt to. I strongly recommend other trainers to give it a go.

An obvious defect however, is that due to the very high level of details it acquired through training, some features I thought to be unique to her original costume (the chest harness as I noticed) would pop up somewhat more often when not prompted to do so. Not sure if negative prompting could fix this or I need to further refine captioning.

---

Release V2.0 works well on full weight (1.0) and here are prompts for replicating her original costume. Copy & paste it for ease of use, as skimming chara/outfits into a single keyword is NOT remotely as flexible despite it surely is convenient:

• acrgazln, very long hair, headgear, blue coat, cleavage, midriff, off shoulder, very long sleeves, sleeves past fingers, short shorts, thigh strap, black footwear,

Release V1 has been tested and works with high level of flexibility on high weight range (0.7-0.9), for lower weight details such as her eye color or harness may become inconsistent.

Suggested Prompts:

• Base: acrgazln, headgear, blue coat, sleeves past fingers, short black shorts, black footwear

描述:

• IMPORTANT: This is a LoHA model. Install this extension for webui to enable using it like other LoRAs: https://github.com/KohakuBlueleaf/a1111-sd-webui-locon.

  FP16, trained on 66 pic dataset.

训练词语: acrgazln