CivArchive
    REYAMix - 1.00
    Preview 1
    Preview 2
    Preview 3
    Preview 4
    Preview 5
    Preview 6
    Preview 7

    Description

    Version 1.00. There is no plan to update this in the future unless there
    is update to the base components or if there's a need to update this
    particular model (nah, real reason is I don't have good enough equipment
    to do a thorough test before release and not enough knowledge to iron
    out issues). All the Beta version will still be made available for those
    who prefer them than v1.00.

    Thank you for choosing REYAMix to generate your images.

    Changelog:
    -Updated base component's block weights.
    -VAE changed from MSE (a.k.a SD-VAE) into WD v2 VAE.
    -No initial noise reduction.
    -Slight increase in contrast.

    Known issue:
    -This model can't do img to img inpainting.
    -CAN'T GENERATE A PROPER LIGHTSABER, WHYYYYY
    -Better style at the cost of hand quality.
    -I don't know if this is an issue or not but most generation will show
    ears.
    -Trying to generate simple stuff (example: background) using the
    masterpiece tag somehow results in a person being added to the
    generation.

    Prompt such as masterpiece, best quality, etc. isn't necessary but
    always welcome to be used (especially if you want to generate godly art,
    though I don't know if this particular model can handle god prompt XD).

    Negative would be: "(worst quality:1.4), (low quality:1.4), (normal
    quality:1.4) and (interlocked fingers)". Also use "(text), (watermark),
    (logo), jpeg artifact" for avoiding things like watermark or text being
    generated. You can also try the available negative embeddings and
    compare which has the best result! (Suggested to try using negative
    embeddings to generate more things in the background).

    > Sampling: Euler A, DPM++2M SDE Karras, DPM++2M/Karras (prefered),
    restart, etc. (basically all works, but this model is just not suited to
    generate using LCM).
    > Upscaler is either R-ESRGAN4x+Anime6B, 4x-AnimeSharp with 0.4
    denoising strength (can be lower) or Latent Nearest Exact with 0.5
    denoising strength (can be higher). AnimeSharp is recommended if you
    have text in the image. Also NMKD's Superscale works really well,
    especially in creating cinematics.
    > No VAE is prefered (already baked in) bu* ***ng others like ft-mse
    840000 is also fine.

    FAQ

    Details

    Downloads
    3,328
    Platform
    ShakkerAI
    Platform Status
    Available
    Created
    8/23/2024
    Updated
    8/23/2024
    Deleted
    -

    Files

    REYAMix-v1.fp16.safetensors

    Mirrors

    ShakkerAI (1 mirrors)
    TensorArt (1 mirrors)