Human Focus Photography is a general-purpose LoRA trained on diverse photographic portraits and lifestyle images. It is intended to reduce bias, enhance realism, lighting, and natural detail across a wide range of subjects, making it well-suited for portraits, street scenes, fashion, and candid photography styles.Workflow is included in the sample images.Training for v1.0 took around 17 hours on my 5090SimpleTuner training config:<pre><code>{
 "aspect_bucket_rounding": 2,
 "base_model_precision": "int8-quanto",
 "caption_dropout_probability": 0.05,
 "checkpointing_steps": 250,
 "checkpoints_total_limit": 40,
 "compress_disk_cache": false,
 "data_backend_config": "/data/datasets/config/Human_Focus_Photography_v1.0.json",
 "disable_benchmark": true,
 "disable_bucket_pruning": true,
 "flow_schedule_shift": 1.73,
 "gradient_accumulation_steps": 8,
 "gradient_checkpointing": true,
 "ignore_final_epochs": true,
 "learning_rate": 0.000025,
 "lora_type": "lycoris",
 "lr_scheduler": "cosine",
 "lr_warmup_steps": 100,
 "lycoris_config": "/data/config/lycoris_config_283.19M.json",
 "max_grad_norm": 0.5,
 "max_train_steps": 0,
 "minimum_image_size": 0,
 "mixed_precision": "bf16",
 "model_family": "qwen_image",
 "model_flavour": "v1.0",
 "model_type": "lora",
 "num_eval_images": 25,
 "num_train_epochs": 100,
 "optimizer": "optimi-lion",
 "optimizer_config": "weight_decay=0.0",
 "output_dir": "/data/output/loras/Human_Focus_Photography_v1.0",
 "push_checkpoints_to_hub": false,
 "push_to_hub": false,
 "quantize_activations": false,
 "quantize_via": "cpu",
 "report_to": "none",
 "resolution_type": "pixel_area",
 "resolution": 1024,
 "resume_from_checkpoint": "latest",
 "seed": 42,
 "skip_file_discovery": false,
 "train_batch_size": 1,
 "use_ema": true,
 "vae_batch_size": 2,
 "validation_disable": true
}</code></pre>And the Lycoris config:<pre><code>{
 "algo": "lokr",
 "multiplier": 1.0,
 "linear_dim": 100000,
 "linear_alpha": 1,
 "factor": 12,
 "apply_preset": {
 "target_module": [
 "Attention",
 "FeedForward"
 ],
 "module_algo_map": {
 "Attention": {
 "factor": 12
 },
 "FeedForward": {
 "factor": 6
 }
 }
 }
}</code></pre>

Human Focus Photography

My plan was to wait until step 12,000 for this version to keep the same interval, but I feel the results for 10,000 are worthy of the bump.

Files with SHA256: fc509551da95ee5506255dbab0dca7848acba0d45fa0aa949ff6a61d19c2d5a6