<h3></h3><h3 id="this-is-just-a-test-lora-trained-on-only-10-pussy-spread-clips-so-please-don't-expect-great-results.">This is just a test LoRA, trained on only 10 pussy spread clips, so please don't expect great results.</h3><h3 id="no-dataset-changes-in-v0.2.-now-all-training-video-captions-follow-the-minimax-h3-prompt-writing-guide.-the-prompts-of-all-showcase-videos-also-follow-the-guide-and-are-in-english.">No dataset changes in V0.2. Now all training video captions follow the Minimax H3 prompt writing guide. The prompts of all showcase videos also follow the guide and are in English.</h3><p></p><p></p><p>Minimax H3 test LoRA for pussy spread<br /><br /><br /></p><p></p><p>Training setting:</p><pre><code>job: "extension"
config:
  name: "minimax_v3"
  process:
    - type: "diffusion_trainer"
      training_folder: "/path/to/output"
      sqlite_db_path: "./aitk_db.db"
      device: "cuda"
      trigger_word: null
      performance_log_every: 10
      network:
        type: "lora"
        linear: 16
        linear_alpha: 16
        conv: 16
        conv_alpha: 16
        lokr_full_rank: true
        lokr_factor: -1
        network_kwargs:
          ignore_if_contains:
            - "adaln_proj"
      save:
        dtype: "bf16"
        save_every: 250
        max_step_saves_to_keep: 4
        save_format: "diffusers"
        push_to_hub: false
      datasets:
        - folder_path: "/path/to/datasets/my_dataset"
          mask_path: null
          mask_min_value: 0.1
          default_caption: ""
          caption_ext: "txt"
          caption_dropout_rate: 0
          cache_latents_to_disk: true
          is_reg: false
          network_weight: 1
          resolution:
            - 1024
            - 768
          controls: []
          shrink_video_to_frames: true
          num_frames: 39
          flip_x: false
          flip_y: false
          num_repeats: 1
          do_i2v: true
          fps: 24
          auto_frame_count: true
      train:
        batch_size: 1
        bypass_guidance_embedding: false
        steps: 3000
        gradient_accumulation: 1
        train_unet: true
        train_text_encoder: false
        gradient_checkpointing: true
        noise_scheduler: "flowmatch"
        optimizer: "automagic3"
        timestep_type: "shift"
        content_or_style: "balanced"
        optimizer_params:
          weight_decay: 0.0001
        unload_text_encoder: false
        cache_text_embeddings: true
        lr: 0.0001
        ema_config:
          use_ema: false
          ema_decay: 0.99
        skip_first_sample: false
        force_first_sample: false
        disable_sampling: false
        dtype: "bf16"
        diff_output_preservation: false
        diff_output_preservation_multiplier: 1
        diff_output_preservation_class: "person"
        switch_boundary_every: 1
        loss_type: "mse"
        do_guidance_loss: true
        guidance_loss_target: 3.5
        audio_loss_multiplier: 1
        do_differential_guidance: true
        differential_guidance_scale: 4
      logging:
        log_every: 1
        use_ui_logger: true
      model:
        name_or_path: "Comfy-Org/MiniMax-H3"
        quantize: true
        qtype: "convrot8"
        quantize_te: true
        qtype_te: "nvfp4"
        arch: "minimax_h3"
        low_vram: false
        model_kwargs: {}
        compile: false
        layer_offloading: false
        layer_offloading_text_encoder_percent: 1
        layer_offloading_transformer_percent: 1
        assistant_lora_path: "ostris/minimax_h3_training_adapter/minimax_h3_training_adapter_v1.safetensors"</code></pre><p></p>

MiniMax H3 - Pussy spread

<p>Still sample dataset as V0.1, but video captions now follow the MiniMax H3 prompt writing guide</p>

Files with SHA256: 4670e3bd2c30faf3d733d9568c7fa38e8a38482eeef58f842affc7053372bef7