Evangelion "Post-impact" Apocalyptic Style [Flux.1 D] by AutoPastel

Please Read Before Use

Thank you for your interest in the "Evangelion 'Post-Impact' Apocalyptic Style" LoRA for the FLUX model. This LoRA is designed to capture the apocalyptic landscapes depicted in the "Rebuild of Evangelion" series, particularly from "Evangelion: 3.0+1.0 Thrice Upon a Time".

Important Notes:

1. No Character or Mecha Images: The training dataset for this LoRA DOES NOT include images of any characters (e.g., Shinji Ikari, Rei Ayanami) or Evangelion units (e.g., EVA-01, EVA-00, EVA-02) from the Evangelion series.

2. Focus on Apocalyptic Landscapes: The primary aim of this LoRA is to replicate the stunning apocalyptic scenery from "Evangelion: 3.0+1.0 Thrice Upon a Time." In this film, post-Fourth Impact Earth is depicted with vast areas tainted by the L Barrier, resulting in a rust-like reddish hue. Due to this focus, generated images may exhibit a strong red bias.

3. Potential Generation Issues: When using this LoRA, you might encounter issues such as distorted human figures or compositional anomalies. To mitigate these problems, consider the following adjustments:

   • Lower the FLUX Guidance Value: High guidance values can lead to undesirable artifacts. Reducing this value may help in achieving more coherent outputs.

   • Adjust the LoRA Weight: If the generated images display significant issues, decreasing the LoRA weight can help. A weight range between 0.2 and 1 is recommended, depending on your prompt.

   • Experiment with Different Random Seeds: Trying various random seeds can lead to diverse results and may help in overcoming specific generation challenges.

About This LoRA

This is my second experimental LoRA, crafted to recreate the breathtaking apocalyptic landscapes from "Evangelion: 3.0+1.0 Thrice Upon a Time". In this film, following the Fourth Impact, many regions on Earth are depicted as being contaminated by the L Barrier, giving them a rust-like reddish appearance. The training dataset comprises high-resolution (5070x2160) stills focusing on these landscapes. As a result, a side effect is that images generated using this LoRA may exhibit a strong red bias.

Sample Training Data

Here are some examples of the training data used:

1. Contaminated Moon Surface:

   

2. Massive Halo During Impact:

   

3. Power Lines:

   

4. Red Waves:

   

Recommended Parameters

• Image Aspect Ratio: 21:9 (closely matches the training data's aspect ratio).

• FLUX Guidance Value: 2.5 (higher values may increase the likelihood of generation issues).

• LoRA Weight: 0.2 - 1 (adjust based on your prompt; if significant issues arise, consider lowering the weight).

• Steps: 30 - 40.

• Trigger Word: p0st_1mp4ct style.

Exploring Alternative Styles

During testing, I discovered that this LoRA is also adept at generating images in a watercolor ink painting style. To achieve this effect, prepend your prompt with:

p0st_1mp4ct style, watercolor, pencils, ink, clear contours, airy, transparent, haze, abstract, pastel colors, tansu ozmen,

Credit to a Civitai user @mailmonroe for this combination (original post).

By experimenting with these settings and prompts, you can explore a range of creative outputs inspired by the evocative landscapes of the Evangelion universe.