CivArchive
    epi_noiseoffset - v2
    NSFW
    Preview 167154
    Preview 178275
    Preview 167153
    Preview 167152
    Preview 178280

    UPDATE: v2-pynoise released, read the Version changes/notes

    LORA based on the Noise Offset post for better contrast and darker images.

    Weighting depends often on Sampler, kept it in the low-middle range (Maybe i will put up a stronger one).

    Custom weighting is needed sometimes.

    Works better if u use good keywords like: dark studio, rim lighting, two tone lighting, dimly lit, low key etc.

    Didn't like the color saturation and image compositions from the theovercomer8sContrastFix_sd15 so i made this one.

    UPDATE: v2 is slightly stronger in the whole composition

    Update 2: added safetensor file

    Quick Comparsion

    Description

    bigger and a bit better

    FAQ

    Comments (42)

    PolygonFeb 28, 2023
    CivitAI

    This LORA will not show up in the additional networks list of A1111.

    Am I doing anything wrong?

    Does it show up under a completely different name?

    osakaharkerFeb 28, 2023

    just a guess but maybe you have the "Show only safetensors" On and this is a PickleTensor

    PolygonFeb 28, 2023

    @osakaharker That was a good idea. I took a look and it's not checked, so it should be showing other formats.

    epinikion
    Author
    Feb 28, 2023

    @Polygon does that help u? https://civitai.com/models/13941?commentId=40193&modal=commentThread

    PolygonFeb 28, 2023· 1 reaction

    @epinikion Thank you, I did see that, and it seemed to work when used in the prompt. I'm trying to figure out how to show up normally in the drop down with the other LORA, though. Any ideas?

    epinikion
    Author
    Feb 28, 2023

    @Polygon Nope, I don’t use the additional network extension. Can have a look tomorrow maybe. Do u see it in the extra Networks after refresh?

    epinikion
    Author
    Mar 1, 2023· 1 reaction

    @Polygon I've taken a look into it and i can see the Lora from the dropdown. Only thing that could be possible, is that u have not set "Only show models that have/don't have user-added metadata" under the Settings to disabled because i've not set metadata. Otherwise it should show up. Maybe it could also be the wrong path where u put it. Since Automatic1111 hast Lora build in natively i set the Setting for extra path to scan for Lora files to "C:\A1111 Web UI\stable-diffusion-webui\models\Lora", wherever your Lora files are.

    PolygonMar 1, 2023

    @epinikion I didn't see it after a refresh. I have that option disabled. I'll try again today with the new one, thank you!

    DanzelusFeb 28, 2023
    CivitAI

    I need to merge it with model?

    sheevlordFeb 28, 2023· 3 reactions

    No, it's a LoRA. You need to either use the additional networks extension or update Automatic1111 to the latest version since it now supports LoRA natively. It works in a similar way to embeddings and hypernetworks, in the sense that it's applied to the model on the fly without modifying the file

    sheevlordFeb 28, 2023· 2 reactions

    See the "Extra network" section of this page
    https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features

    epinikion
    Author
    Feb 28, 2023· 1 reaction

    Yes, download, put it in the "A1111 Web UI\stable-diffusion-webui\models\Lora" Folder, and use <lora:epiNoiseoffset_v2:1> in the prompt.

    U can adjust the weighting with the number after : like <lora:epiNoiseoffset_v2:1.5> for stronger embedding and so on

    DanzelusFeb 28, 2023· 2 reactions

    @epinikion Thx you very much

    DanzelusFeb 28, 2023· 2 reactions

    @sheevlord Thank you 

    IIIXMMar 1, 2023
    CivitAI

    Why write like this? <lora:epi_noiseoffset2:1>

    In the Tags tab, there's not this tag.

    When do you use a Lora, you need to do it like this? Thanks

    epinikion
    Author
    Mar 1, 2023· 2 reactions

    That’s my Lora name how I named it, when u download it from civitai it will rename the Lora to the Title and version e.g. epiNoiseoffset_v2 so u have to use <lora:epiNoiseoffset_v2:1> The easiest way to chose a Lora is from the „extra Networks“ tab https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Features

    GalogaloMar 1, 2023
    CivitAI

    It is supposed to work with anime models?

    epinikion
    Author
    Mar 1, 2023

    Have none of them, why not test it out? It should.

    TroublesomeAMar 1, 2023· 6 reactions
    CivitAI

    please add safetensor

    epinikion
    Author
    Mar 1, 2023

    Added.
    Had to add a version for safetensor, because Lora files don't support two modelfiles in different formats like it works for Checkpoints

    epinikion
    Author
    Mar 1, 2023

    edit: got it working for same version with a workaround :D So ignore what I've just said

    AiliMar 1, 2023
    CivitAI

    How many dims is this ? i've been trying to merge it with other LORAs but i think the difference in dim is causing an issue, if you can tell me how many dims i might work around it

    epinikion
    Author
    Mar 1, 2023

    v2 was about 96 Dimension if i remeber it correct

    truefalseMar 1, 2023
    CivitAI

    doesn't work RuntimeError: mat1 and mat2 shapes cannot be multiplied (154x1024 and 768x96)

    epinikion
    Author
    Mar 1, 2023

    It's not the Lora. Has something to do with Model u use, maybe u selected a 2.x Model or a script or controlNet etc.

    truefalseMar 1, 2023· 1 reaction

    @epinikion thank you for clearing it up! i was using a 2.x model yes thanks!

    mdvMar 1, 2023
    CivitAI

    I installed it and it shows up in the Lora tab, I use a checkpoint based on 1.5, I use this in my prompt <lora:epi_noiseoffset2:1> and yet I get this error message when I run my prompt.

    "Couldn't find Lora with name epi_noiseoffset2"

    epinikion
    Author
    Mar 1, 2023

    please read here https://civitai.com/models/13941?commentId=41138&modal=commentThread

    mdvMar 1, 2023· 1 reaction

    Okay thank you! I'll give it a shot. Just that I put the image in the png info and sent it to txt2img and I expected it to run as is.

    epinikion
    Author
    Mar 1, 2023

    @mdv yeah, sadly that doesn't work :(

    Tim_DMar 2, 2023· 1 reaction

    @epinikion <lora:epi_noiseoffset_v2:1>

    network name has to match the file name. OP gave the wrong example of the tag

    epinikion
    Author
    Mar 2, 2023· 1 reaction

    @Tim_D <lora:epiNoiseoffset_v2:1> thats the correct spelling after you download it from civitai, so it is a well known "problem" for Lora names and sample prompts

    Update: I edited the prompts in the samples to the namings from CivitAi

    SparklingAppsMar 3, 2023

    @epinikion In <lora:epiNoiseoffset_v2:1> is it necessary to mention lora or can we directly use <epiNoiseoffset_v2:1>? I am using the .pt files like embedding with my model using the textual inversion

    epinikion
    Author
    Mar 3, 2023

    @SparklingApps You have to use <lora: so Automatic1111 knows, that it has to use a lora here. (Looking in the right folder) Doesn‘t matter if u use .bin, .pt or .safetensor file.

    SparklingAppsMar 3, 2023

    @epinikion actually I am not using Automatic1111. I have integrated it with an App named ImagAIne so there is no need to use lora right?

    epinikion
    Author
    Mar 3, 2023

    @SparklingApps don't know what that should be

    SparklingAppsMar 3, 2023

    @epinikion Thank you.. Feel free to try out in this app: https://www.sparklingapps.com/imagaine and try using the lora by tagging <epiNoiseoffset_v2> in the prompt and see if its giving the results.

    farcaller899Mar 3, 2023
    CivitAI

    Just from the samples shown, V3 does look better. It is darker, but more importantly, the pictures made with it look more 'real', and the overall result is more cinematic and dramatic. V3 is the way!

    epinikion
    Author
    Mar 3, 2023

    Had an error in my settings, so i have to redo and see if it works out :(

    eric847Mar 3, 2023
    CivitAI

    Need some assistance. Trying to generate hugh laurie image and get similar pose but without the highly detailed face. Using same image A1111 settings. model: epiHyperphotogodess_v1CLIPFix.safetensors

    Lora: epiNoiseoffset_v2.safetensors

    VAE: vae-ft-mse-840000-ema-pruned

    prompt, seed and other same

    did you use ersgan or gfpgan to increase details?

    epinikion
    Author
    Mar 4, 2023

    Do you have turned on face restoration? Sounds like that, I don't use it.

    AgathornMar 3, 2023
    CivitAI

    Any tips for darker scenes but with a soft bounce light on the model's face?

    LORA
    SD 1.5
    by epinikion
    Download (Beta) View on CivitAI
    dark
    noise offset
    high contrast
    tool

    Details

    Downloads
    208,875
    Platform
    CivitAI
    Platform Status
    Available
    Created
    2/28/2023
    Updated
    5/13/2026
    Deleted
    -

    Files

    epi_noiseoffset2.pt

    Size:
    77.87 MB
    SHA256:
    3775aad8a02dc5ca4e411f812b9b68b7539db2a4074b0577ce9b033b5779c0b1

    Mirrors

    HuggingFace (9 mirrors)
    CivitAI (1 mirrors)
    TensorFiles (1 mirrors)
    TensorArt (1 mirrors)

    epi_noiseoffset2.safetensors

    Size:
    77.71 MB
    SHA256:
    81680c064e9f50dfcc11ec5e25da1832f523ec84afd544f372c7786f3ddcbbac

    Mirrors

    HuggingFace (135 mirrors)
    CivitAI (1 mirrors)
    TensorFiles (1 mirrors)
    CivitasBay (1 mirrors)

    Available On (1 platform)

    Same model published on other platforms. May have additional downloads or version variants.

    CivitasBay
    epi_noiseoffset - v2