config.json

{
  "_gradient_checkpointing": false,
  "architectures": [
    "VisualForesight"
  ],
  "attn_implementation": null,
  "diffusion_model_id": "Efficient-Large-Model/Sana_1600M_512px_diffusers",
  "dtype": "bfloat16",
  "in_channels": 32,
  "input_size": [
    15,
    20
  ],
  "max_input_text_tokens": 256,
  "mllm_id": "google/gemma-2-2b-it",
  "model_type": "visualforesight",
  "modules_to_freeze": [
    "vae",
    "mllm_backbone"
  ],
  "modules_to_unfreeze": [],
  "noise_scheduler_id": "Efficient-Large-Model/Sana_1600M_512px_diffusers",
  "scheduler_id": "Efficient-Large-Model/Sana_1600M_512px_diffusers",
  "system_prompt": "You are a robot and should focus on your actions. Generate a new image that meets the user's instruction while maintaining consistency with the original input where appropriate.",
  "transformers_version": "4.57.1",
  "vae_downsample_f": 32,
  "vae_id": "Efficient-Large-Model/Sana_1600M_512px_diffusers"
}