{ "model_type": "diffusion_cond", "sample_rate": 44100, "model": { "conditioning": { "cond_dim": 768, "configs": [ { "id": "prompt", "type": "qwen_text", "config": { "model_path": "text_encoder", "hidden_dim": 1024, "max_length": 512, "enable_grad": false, "project_out": true, "padding_mode": "zero", "caption_region_embedding": false, "source_region_embedding": false, "sceneplan_role_embedding": true, "sceneplan_max_sources": 4, "sceneplan_role_scale": 1.0, "fail_on_truncation": true } }, { "id": "sceneplan_44", "type": "sceneplan_44", "config": { "output_dim": 256, "max_sources": 4, "trajectory_feature_dim": 5, "event_embedding_dim": 32, "trajectory_embedding_dim": 32 } } ], "pre_encoded_keys": [ "source_foa_latent" ] }, "diffusion": { "diffusion_objective": "rectified_flow", "mask_padding_attention": true, "cross_attention_cond_ids": [ "prompt" ], "global_cond_ids": [], "type": "dit", "config": { "io_channels": 64, "embed_dim": 1024, "depth": 15, "num_heads": 16, "cond_token_dim": 768, "global_cond_dim": 0, "project_cond_tokens": true, "transformer_type": "continuous_transformer", "input_concat_dim": 320, "input_concat_cond_cfg": true, "require_explicit_negative_input_concat": true }, "input_concat_ids": [ "sceneplan_44", "source_foa_latent" ] }, "io_channels": 64 } }