CivArchive
    BALAN Ultima (R3) - R3U
    Preview 98958478Preview 98958476Preview 98958470Preview 98958482Preview 98958481Preview 98958468Preview 98958465Preview 98958469Preview 98958477Preview 98958472Preview 98958474Preview 98958471Preview 98958475Preview 98958473Preview 98958479

    * Always read carefully attached license and subsequent model permissions before using any model, including this one (usually it sits to the right side of the screen, below creator's avatar)

    * Provided examples were generated in ComfyUI, different web-ui might produce different result.

    The R3U Version

    Version R3U is a beta checkpoint and not as stable as any previous version. The objective in version R3 is to re-capture some elements from version R1.

    BALAN Ultima

    This model is the result of merging more than a dozen checkpoints varying from 2D models, fantasy models, landscape models, architectural models, doll models, to photorealistic models with techniques such as block merging, LoRA fine-tuning, and DARE merging.

    Most notable advantage of this model is its ability to produce image with relatively less noise yet still retaining most of its important details such as skin and textures. Additionally, through exhaustive and complex merging process, the model also capable of delivering high quality unique output using relatively simple prompt.

    Due to how this model was created, this model can produce high quality images with minimal prompting, but on the opposite side (not always the case) forced prompt or prompt that just too complex may ruin its generations, so be more gentle with the prompt (or maybe try to increase the steps on longer prompt).

    Tips

    • 512p :: For fast generation, use 512p resolutions (512 x 768, 768 x 512, etc.) , then use around 3 - 7 CFG Scale with minimum steps of 14.

    • 640p :: Try resolution such as 640 x 960 (portrait), 960 x 640 (landscape), or any resolution in between for starter. Then, pick any CFG Scale from 5 - 8 and at least 18 for the steps (depending on the prompt, it may need higher CFG Scale and more steps to get desirable result). For most stable output use 8 CFG Scale with at least 28 steps (note that CFG as high will make the output less creative). [ Note: R3U is not a stable checkpoint. Generating base image at higher resolution will take trial and error in version R3U ]

    |!| This model can produce content that may be considered as sensitive and I bear no responsibility for the outcomes of any content you generate!

    Credits

    The following are utilized resources that require credit mention:

    * Checkpoints

    • Grounded - Realistic Mix [v1.6] (by farcycle)

    • GODDESS of Realism [v4] (by Oppkllll)

    • epiCRealism [Natural Sin] (by epinikion)

    • MFantasy奇梦女孩 [v1.0] (by AI_Heart)

    • A.Ikari - SFW&NSFW [v1.0 Lumina] (by 7ELIX)

    * LoRA

    • Real Skin Conditioner [v1.0] (by NextMeal)

    Description

    R3U is a beta checkpoint

    Checkpoint
    SD 1.5

    Details

    Downloads
    180
    Platform
    CivitAI
    Platform Status
    Available
    Created
    9/7/2025
    Updated
    9/28/2025
    Deleted
    -

    Files

    balanUltimaR3_r3u.safetensors

    Mirrors