| acceleration: |
| load_text_encoder_in_8bit: false |
| mixed_precision_mode: bf16 |
| quantization: null |
| checkpoints: |
| interval: 250 |
| keep_last_n: 10 |
| precision: bfloat16 |
| data: |
| num_dataloader_workers: 2 |
| preprocessed_data_root: .precomputed |
| flow_matching: |
| timestep_sampling_mode: shifted_logit_normal |
| timestep_sampling_params: {} |
| hub: |
| hub_model_id: Alissonerdx/BFS-LTX2 |
| push_to_hub: true |
| lora: |
| alpha: 256 |
| dropout: 0.005 |
| rank: 256 |
| target_modules: |
| - attn1.to_k |
| - attn1.to_q |
| - attn1.to_v |
| - attn1.to_out.0 |
| - attn2.to_k |
| - attn2.to_q |
| - attn2.to_v |
| - attn2.to_out.0 |
| - ff.net.0.proj |
| - ff.net.2 |
| - adaln_single.linear |
| - video_embeddings_connector.transformer_1d_blocks.0.attn1.to_q |
| - video_embeddings_connector.transformer_1d_blocks.0.attn1.to_k |
| - video_embeddings_connector.transformer_1d_blocks.1.attn1.to_q |
| - video_embeddings_connector.transformer_1d_blocks.1.attn1.to_k |
| - scale_shift_table |
| model: |
| load_checkpoint: /train/LTX-2/packages/ltx-trainer/outputs/ltx2_v2v_ic_lora_head_swap_128_prodigy_4/checkpoints |
| model_path: /train/models/ltx-2-19b-dev.safetensors |
| text_encoder_path: /train/models/gemma |
| training_mode: lora |
| optimization: |
| batch_size: 1 |
| enable_gradient_checkpointing: true |
| gradient_accumulation_steps: 1 |
| learning_rate: 4.0e-05 |
| max_grad_norm: 1.0 |
| optimizer_type: lion8bit |
| scheduler_params: {} |
| scheduler_type: cosine |
| steps: 3000 |
| output_dir: /train/LTX-2/packages/ltx-trainer/outputs/ltx2_v2v_ic_lora_head_swap_128_prodigy_5 |
| seed: 42 |
| training_strategy: |
| condition_on_last_frame: false |
| first_frame_conditioning_p: 1.0 |
| name: video_to_video |
| num_conditioning_frames: 1 |
| reference_latents_dir: reference_latents |
| validation: |
| frame_rate: 25.0 |
| generate_audio: false |
| guidance_scale: 4.0 |
| images: |
| - dataset/test/351912466498942.png |
| - dataset/test/6403624465969299.png |
| include_reference_in_output: true |
| inference_steps: 30 |
| interval: 250 |
| negative_prompt: worst quality, inconsistent motion, blurry, jittery, distorted |
| prompts: |
| - 'head_swap: replace the head in the video.' |
| - 'head_swap: replace the head in the video.' |
| reference_downscale_factor: 1 |
| reference_videos: |
| - dataset/test/351912466498942.mp4 |
| - dataset/test/6403624465969299.mp4 |
| seed: 42 |
| skip_initial_validation: true |
| stg_blocks: |
| - 29 |
| stg_mode: stg_v |
| stg_scale: 1.0 |
| video_dims: !!python/tuple |
| - 576 |
| - 1024 |
| - 49 |
| videos_per_prompt: 1 |
| wandb: |
| enabled: true |
| entity: null |
| log_validation_videos: true |
| project: ltx-2-trainer-31-01-2026-v1 |
| tags: |
| - ltx2 |
| - ic-lora |
| - video-to-video |
|
|
