---
model:
  base_learning_rate: 1.0e-04
  target: ldm.models.diffusion.ddpm.LatentDiffusion
  params:
    linear_start: 0.00085
    linear_end: 0.0120
    num_timesteps_cond: 1
    log_every_t: 200
    timesteps: 1000
    first_stage_key: "jpg"
    cond_stage_key: "txt"
    image_size: 64
    channels: 4
    cond_stage_trainable: false   # Note: different from the one we trained before
    conditioning_key: crossattn
    monitor: val/loss_simple_ema
    scale_factor: 0.18215
    use_ema: False

    scheduler_config: # 10000 warmup steps
      target: ldm.lr_scheduler.LambdaLinearScheduler
      params:
        warm_up_steps: [ 10000 ]
        cycle_lengths: [ 10000000000000 ] # incredibly large number to prevent corner cases
        f_start: [ 1.e-6 ]
        f_max: [ 1. ]
        f_min: [ 1. ]

    unet_config:
      target: ldm.modules.diffusionmodules.openaimodel.UNetModel
      params:
        image_size: 32 # unused
        in_channels: 4
        out_channels: 4
        model_channels: 320
        attention_resolutions: [ 4, 2, 1 ]
        num_res_blocks: 2
        channel_mult: [ 1, 2, 4, 4 ]
        num_heads: 8
        use_spatial_transformer: True
        transformer_depth: 1
        context_dim: 768
        use_checkpoint: True
        legacy: False

    first_stage_config:
      target: ldm.models.autoencoder.AutoencoderKL
      params:
        embed_dim: 4
        monitor: val/rec_loss
        ddconfig:
          double_z: true
          z_channels: 4
          resolution: 256
          in_channels: 3
          out_ch: 3
          ch: 128
          ch_mult:
          - 1
          - 2
          - 4
          - 4
          num_res_blocks: 2
          attn_resolutions: []
          dropout: 0.0
        lossconfig:
          target: torch.nn.Identity

    cond_stage_config:
      target: ldm.modules.encoders.modules.FrozenCLIPEmbedder
---

Model name: H&A 3DKX
Model version: 1.0b

## Description: 

SFW model with limited nsfw capabilities (suggestive nsfw) that is highly versatile for 3D renders.
The model has the particularity of splitting itself into two different well balanced styles. 
If you'd like to have your 3D characters have a more "Cartoony" face, you simply start your prompt 
with "3d cartoon of", and if you want the classic 3D render style, you write "a 3d render of".

## Dataset:
 - between 140 and 180 pictures of 3D render of all kind

## Has a high success rate at:
- sfw portraits, full body poses, close ups, etc 
- high versatility in terms of outputs, it isn't locked to perform well on portraits
- Landscapes, cyberpunk, steampunk, natural, scifi, etc
- 2B Nier Automata (Don't ask us why)
- different body types - different ethnicity
- nsfw portraits, full body poses, close ups, etc 

## What it "In theory" shouldn't exceed at:
- anything outside the scope of portraits, people, landscapes, game artworks, 3D sculptures, 3D fantasy, 3D film stills, etc
- celebrities
- highly specific animated cartoon characters
- multiple subjects 
- highly specific video-game characters
- pornography, genitalia and highly explicit materials