Budget: 2000 USD Deadline: 25 days
Ready to develop a model for generating portraits based on 8-10 user photos. I will use Stable Diffusion + LoRA to speed up model adaptation and support various styles (classic portrait, corporate style, street photography, etc.). To improve quality, I will apply Real-ESRGAN or CodeFormer. For face recognition, I will use Mediapipe or InsightFace to determine smile, eye openness, and other parameters.
Model training will be optimized, but speed depends on GPU power. What specific GPU is planned for training? If it's mid-range, LoRA will significantly speed up the process, but we still need to clarify what the time and resource constraints are.
Development will be conducted on PyTorch with the possibility of containerization via Docker for simplified deployment. I will propose an efficient pipeline, but for an accurate estimate, more details are needed: what photo format will be used, whether full support for all styles is required from the start, and how important is control over the training process?