Instructions to use Miayan/freelit-checkpoint with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use Miayan/freelit-checkpoint with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("Miayan/freelit-checkpoint", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
| experiment: | |
| project_name: relighting_indoors | |
| output_dir: train_r1_diff | |
| logging_dir: logs | |
| report_to: tensorboard | |
| model: | |
| pretrained_model_name_or_path: stable-diffusion-2-1 | |
| revision: null | |
| variant: null | |
| tokenizer_name: null | |
| hf_repo_id: Miayan/project | |
| hf_version: null | |
| pretrain_unet_path: scribblelight_controlnet/checkpoint-10000 | |
| resume_from_checkpoint: latest | |
| enable_ambient_cond: true | |
| light_encoder: | |
| cross_dim: 768 | |
| K: 4 | |
| Bi: 4 | |
| Bc: 4 | |
| Ba: 4 | |
| albedo_estimator: | |
| enabled: false | |
| version: v2 | |
| load_stage: 3 | |
| frozen_components: | |
| - ord_model | |
| - iid_model | |
| - col_model | |
| trainable_components: [] | |
| data: | |
| dataset_cache_dir: ./.hf_cache | |
| data_hf_repo_id: Miayan/physical-relighting-dataset | |
| data_split: test | |
| resolution: 512 | |
| dataloader_num_workers: 0 | |
| condition_mode: v1 | |
| relighting_impl: ori | |
| use_ambient_in_controlnet: false | |
| use_color_on_lightmap: true | |
| colors: | |
| - - 255 | |
| - 255 | |
| - 255 | |
| - - 255 | |
| - 0 | |
| - 0 | |
| - - 0 | |
| - 255 | |
| - 0 | |
| - - 0 | |
| - 0 | |
| - 255 | |
| - - 255 | |
| - 255 | |
| - 0 | |
| - - 255 | |
| - 165 | |
| - 0 | |
| - - 128 | |
| - 0 | |
| - 128 | |
| - - 255 | |
| - 192 | |
| - 203 | |
| - - 0 | |
| - 255 | |
| - 255 | |
| - - 255 | |
| - 0 | |
| - 255 | |
| intensities: | |
| - 0.0 | |
| - 0.2 | |
| - 0.4 | |
| - 0.6 | |
| - 0.8 | |
| - 1.0 | |
| training: | |
| batch_size: 1 | |
| num_epochs: 20 | |
| max_train_steps: 100000 | |
| max_train_samples: null | |
| seed: 42 | |
| mixed_precision: 'no' | |
| allow_tf32: true | |
| gradient_accumulation_steps: 1 | |
| checkpointing_steps: 5000 | |
| checkpoints_total_limit: 4 | |
| learning_rate: 5.0e-06 | |
| scale_lr: false | |
| lr_scheduler: constant | |
| lr_warmup_steps: 500 | |
| adam_beta1: 0.9 | |
| adam_beta2: 0.999 | |
| adam_weight_decay: 0.01 | |
| adam_epsilon: 1.0e-08 | |
| max_grad_norm: 1.0 | |
| use_8bit_adam: false | |
| set_grads_to_none: false | |
| losses: | |
| phys_loss: | |
| enabled: false | |
| weight: 1.0 | |
| latent_loss: | |
| enabled: true | |
| weight: 1.0 | |
| image_loss: | |
| enabled: false | |
| weight: 1.0 | |
| sup_structure_loss: | |
| enabled: false | |
| weight: 1.0 | |
| enable_decay: true | |
| start_weight: 1.0 | |
| end_weight: 0.1 | |
| decay_steps: 5000 | |
| self_recon_loss: | |
| enabled: false | |
| mode: teacher | |
| weight: 2.0 | |
| affine_alignment: true | |
| epsilon: 0.05 | |
| staging: | |
| switch_epoch: 5 | |
| stage1_weight: 2.0 | |
| stage2_weight: 0.5 | |
| consistency_loss: | |
| enabled: false | |
| weight: 1.0 | |