CivArchive
    Preview 3266444
    Preview 3266451
    Preview 3266446
    Preview 3266445
    Preview 3266624
    Preview 3266632
    Preview 3266455
    Preview 3266449
    Preview 3266450
    Preview 3266452
    Preview 3266504
    Preview 3266447
    Preview 3266619
    Preview 3266609
    Preview 3266456
    Preview 3266457
    Preview 4052579
    Preview 4052581
    Preview 4052584
    Preview 4052585

    model for asians

    Description

    t3(ver-09) final version

    1) Although the model has a large capacity, it can easily create good pictures.

    2) Please refer to the fromm and options in the sample picture and add a better fromm.

    Pick a lot of great pictures

    3) The next model has already been made in version 10, but I have made it a little more

    I will draw a picture and reveal it later. (It will be much better)

    FAQ

    Comments (10)

    shaozelong584Oct 30, 2023
    CivitAI

    更新速度杠杆的,Well done

    darksagaq12926
    Author
    Oct 30, 2023

    Thank you~

    Anarchism_007Nov 1, 2023
    CivitAI

    大大,用做底模训练真人LORA效果如何哈?

    NoTimeToGenerateNov 5, 2023· 2 reactions
    CivitAI

    Love your model. You’re doing a fantastic job!!

    darksagaq12926
    Author
    Nov 6, 2023

    Thank you~^^;

    civitai2022Nov 7, 2023· 1 reaction
    CivitAI

    well,anybody test which is the best? It looks pretty much the same!

    darksagaq12926
    Author
    Nov 7, 2023· 2 reactions

    1. Change sampling method: Try changing it for each type. The feeling is different. I can give you a recommendation, but it's hard to pick your own picture, so if you're trying to find civitai2022's own feel, try changing the sampling method one by one to the same fromm and the same settings.

    2. CFG Scale change: The image may be broken depending on the CFG Scale change. Since it is difficult to do all the requirements (Recommendation 1), a higher value of 4.5~8 or 16~25 is also good, but if you use it like that, most of the pictures will be broken. So at this time, activate Dynamic Thresholding (CFG Scale Fix) and set Mimic CFG Scale value smaller than CFG Scale. Top percentile of latents to clamp is around 99.95~99.8? Find the Mimic Scale Scheduler and CFG Scale Scheduler that you like. Minimum value of the Mimic Scale Scheduler The value is between 1 and 3. Minimum value of the CFG Scale Scheduler is also between 1 and 3. This also changes the feel of the picture.

    3. Enable ADetailer I use this, but the ADetailer model uses face_yolov8n.pt and mediapipe_face_full, and there is no difference.

    Please look at my picture and add or change the prom below.

    darksagaq12926
    Author
    Nov 30, 2023· 8 reactions
    CivitAI

    We are currently testing t3 (10-test ver) and will upload it soon.

    This time, I will also attach settings to create a more realistic-looking picture.

    The t3(10ver-test) picture has been added to the gallery below.

    darksagaq12926
    Author
    Dec 4, 2023

    @giants It takes a long time to keep testing and revising on your own, so if you upload the picture you made, we can make it a little faster. Please create various pictures and upload them.

    hagane_no_kintamaJan 26, 2024

    General question: Are you training upon your merged models (for refinement) or are you just merging?

    Checkpoint
    SD 1.5

    Details

    Downloads
    4,240
    Platform
    CivitAI
    Platform Status
    Available
    Created
    10/30/2023
    Updated
    5/14/2026
    Deleted
    -

    Files

    t3_Ver09.safetensors

    Mirrors

    HuggingFace (1 mirrors)
    CivitAI (1 mirrors)
    ShakkerAI (1 mirrors)

    Available On (1 platform)

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