Wish You Were Here - a Stable diffusion 1.5 LORA for vintage postcard replication

<!-- Provide a quick summary of what the model is/does. -->

Wish you were here is a LORA model developped to create vintage postcard images. The model was trained on Stable Diffusion 1.5.

Model Description

<!-- Provide a longer summary of what this model is. --> Wish You Were Here (WYWH) is a LORA model developped to replicate the look and feel of vintage postcards. This is done via harvesting public domain images from WikiMedia via manual review and using a combination of manual and automated annotation to describe the images. The specific feature desired to extract were: color, damage and printing technique. The model was developped over a duration of 2 days over 100 epochs of which one epoch was taken as resulting image.

• Developed by: calm-and-collected
• Model type: LORA
• License: CC-BY 4.0
• Finetuned from model [optional]: Stable diffusion 1.5 pruned

Bias, Risks, and Limitations

The model is trained of images from ~650 images. From observation, the majority of these images are from american origins. The model is thus excelent at replicating USA destinations. The model will also replicate damage seen in the images.

Recommendations

<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->

To use the WYWH model, use your favorite Stable Diffusion model (the recommended model is a realistic model) and use the LORA along with the following triggers:

• WYWH (the base trigger)
• Photograph (for photography postcards)
• Drawing (for drawn postcards)
• Damage (to add scratch and water damage to the generation)
• Monochrome (for black and white images)

For negatives, your can use the following:

• White border (if you do not want a white border)

How to Get Started with the Model

You can use this model with automatic1111, comfyui and sdnext.

Training Details

Training Data

The Wish You Were Here dataset consists out of ~650 images of postcards from 1900-1970. Dataset: origional dataset.

Training Hyperparameters

<details> <summary>Kohya_SS paramaters</summary>

{
"LoRA_type": "Standard",
"adaptive_noise_scale": 0,
"additional_parameters": "",
"block_alphas": "",
"block_dims": "",
"block_lr_zero_threshold": "",
"bucket_no_upscale": true,
"bucket_reso_steps": 64,
"cache_latents": true,
"cache_latents_to_disk": true,
"caption_dropout_every_n_epochs": 0.0,
"caption_dropout_rate": 0,
"caption_extension": ".txt",
"clip_skip": 2,
"color_aug": false,
"conv_alpha": 1,
"conv_block_alphas": "",
"conv_block_dims": "",
"conv_dim": 1,
"decompose_both": false,
"dim_from_weights": false,
"down_lr_weight": "",
"enable_bucket": true,
"epoch": 1,
"factor": -1,
"flip_aug": false,
"full_bf16": false,
"full_fp16": false,
"gradient_accumulation_steps": 1,
"gradient_checkpointing": false,
"keep_tokens": "0",
"learning_rate": 0.0001,
"logging_dir": "/home/glow/Desktop/ml/whyw_logs",
"lora_network_weights": "",
"lr_scheduler": "constant",
"lr_scheduler_args": "",
"lr_scheduler_num_cycles": "",
"lr_scheduler_power": "",
"lr_warmup": 0,
"max_bucket_reso": 2048,
"max_data_loader_n_workers": "1",
"max_resolution": "512,650",
"max_timestep": 1000,
"max_token_length": "75",
"max_train_epochs": "100",
"max_train_steps": "",
"mem_eff_attn": true,
"mid_lr_weight": "",
"min_bucket_reso": 256,
"min_snr_gamma": 0,
"min_timestep": 0,
"mixed_precision": "bf16",
"model_list": "custom",
"module_dropout": 0.2,
"multires_noise_discount": 0.2,
"multires_noise_iterations": 8,
"network_alpha": 128,
"network_dim": 256,
"network_dropout": 0.3,
"no_token_padding": false,
"noise_offset": "0.05",
"noise_offset_type": "Multires",
"num_cpu_threads_per_process": 2,
"optimizer": "AdamW8bit",
"optimizer_args": "",
"output_dir": "/home/glow/Desktop/ml/whyw_logs/model_v2",
"output_name": "final_model",
"persistent_data_loader_workers": false,
"pretrained_model_name_or_path": "runwayml/stable-diffusion-v1-5",
"prior_loss_weight": 1.0,
"random_crop": false,
"rank_dropout": 0.2,
"reg_data_dir": "",
"resume": "",
"sample_every_n_epochs": 0,
"sample_every_n_steps": 0,
"sample_prompts": "",
"sample_sampler": "euler_a",
"save_every_n_epochs": 1,
"save_every_n_steps": 0,
"save_last_n_steps": 0,
"save_last_n_steps_state": 0,
"save_model_as": "safetensors",
"save_precision": "bf16",
"save_state": false,
"scale_v_pred_loss_like_noise_pred": false,
"scale_weight_norms": 1,
"sdxl": false,
"sdxl_cache_text_encoder_outputs": false,
"sdxl_no_half_vae": true,
"seed": "1234",
"shuffle_caption": false,
"stop_text_encoder_training": 1,
"text_encoder_lr": 5e-05,
"train_batch_size": 3,
"train_data_dir": "/home/glow/Desktop/wyhw",
"train_on_input": true,
"training_comment": "",
"unet_lr": 0.0001,
"unit": 1,
"up_lr_weight": "",
"use_cp": true,
"use_wandb": false,
"v2": false,
"v_parameterization": false,
"v_pred_like_loss": 0,
"vae_batch_size": 0,
"wandb_api_key": "",
"weighted_captions": false,
"xformers": "xformers"
}

The final model is the 50th iteration of the model. </details>

Hardware

The model was trained on two GTX 4090 for a duration of 2 days to extract 100 epochs of the model.

Software

The model was trained via the Kohya_SS gui.

Model Card Contact

Use the community section of this repository to contact me.