POSTAL 2 (2003) - Videogame Style Lora [ILXL/NoobAI] - ILXL [V1.2] E10 - EPS 1.1

NSFW

22.08.2025: Released E06 Version. Give it a try as it produces some really cool results but use ``` 3d, realistic ``` tags in your positive prompts! This version was released for those who may or may not have had issues with E10 version.

12.08.2025: Version 1.2 is out now! Give it a try

1.2 means new database, no more smudges and overall better consistency and mostly importantly, no more blurriness! Play with your favourite CFG scale, it should no longer cause any trouble from my testing on both WAI-NSFW-ILLUSTRIOUS-SDXL / Hassaku

And since today is the start of the kickstarter for the POSTAL 2 Redux, the release of this version couldn't be timed any better than today, I would say it's almost a coincidence.

All images shown on the gallery are generated locally with Forge.

Hello. This is my third style Lora that is based on the style of a videogame called ''Postal 2'' made by legendary Running With Scissors that I adore and I'd encourage you to support them, as for this lora it should give you cool early 2000s look and vibe in your images!

28.08.2025 - 1.1 version has been released, give it a try!

Surprisingly, this lora was released before POSTAL 2 Redux was announced, so please give that one a look if you can. I really am looking forward to POSTAL 2 Redux and hope all the best for RWS / Flat2VR studio to make it as best as they possibly can.

As of today (5th of June, 2025), I have re-trained the lora model under Illustrious 0.1 for the purpose and did reformat the description to match it, if you see the loras that have ''WAI'' on them, I'd kindly ask you not to use them unless you use EPS/V-Pred 2,0 based models.

From my testing, Hassaku XL (Illustrious) and WAI-NSFW-illustrious-SDXL checkpoints give great results as long as the lora is marked with ILXL_01 in it's name.

This one will be the main release for the time being now.

There are no activation keys/tags but I must insist that you include ``realistic, 3d, detailed background, scenery`` in your positive prompting, it will make all the image compositions look proper. Add ''3d'' if necessary.

Technical Info on how to use it:

Works best around with these base resolutions:

1024x1024 / 826x1216 / 1216x826 / 1536x1024 / 1024x1536

- Lora Weights: 0.7 to 1. 0.8 Is the most optimal value from my tests, inpainting boosts the quality even more.

- Sampling Method: 31 steps, DPM++ SDE / Euler A

- Scheduler: Karras / DIMM

- CFG Scale: Between 1.5 to 5.

(With release of 1.1, you can crank it up all the way to 7 and above)

Works even better with CFG+ Samplers if you're using Forge like me.

HiRes Fix Upscaler: Latent (Antialiased), but you can use one that you prefer by your own choice. 4x-Ultrasharp hasn't been tested because I took the liking to Latents a lot more and tiling takes time on my GPU.

For even better image quality, I highly recommend inpainting with Euler A Sampler and DIMM scheduler.

Thank you for reading, have fun generating your images, I'd would love to see your generations that have been made with this Lora! All the support is greatly appreciated. Your image contributions to this model page help shape future versions.

Description

1.2 Dataset! Expanded with even more data that should be closer to the visuals of the game.
Improved prompt adherence from my various tests and it seems pretty alright however there may be still issues.
Fixed one of the biggest issues with improper color balancing and overall blurryness
Trained on NoobAI Epsilon-pred 1.1-Version.
1,397 Buzz gone, just like that! But it was worth it.

Leaving my Lora settings for the curious:

{
  "engine": "kohya",
  "unetLR": 0.0005,
  "clipSkip": 2,
  "loraType": "lora",
  "keepTokens": 0,
  "networkDim": 32,
  "numRepeats": 3,
  "resolution": 1024,
  "lrScheduler": "cosine_with_restarts",
  "minSnrGamma": 5,
  "noiseOffset": 0.1,
  "targetSteps": 2867,
  "enableBucket": true,
  "networkAlpha": 16,
  "optimizerType": "AdamW8Bit",
  "textEncoderLR": 0.00005,
  "maxTrainEpochs": 10,
  "shuffleCaption": true,
  "trainBatchSize": 9,
  "flipAugmentation": false,
  "lrSchedulerNumCycles": 3
}

FAQ

LORA

Illustrious

by ai_appreciator

Looks like we don't have an active mirror for this file right now.

CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.

Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.