I have high-quality product shots of necklaces, earrings, and rings and I need them blended onto female model photos so convincingly that they look like in-studio fashion shoots. My plan is to run an open-source diffusion model—Flux 2-dev or a similarly capable checkpoint—inside a rented RunPod GPU environment. Here’s the flow I need you to build and, once proven, hand off to me: • Spin up and configure the RunPod instance (A10 / A100 class or better), install the model, required weights, and supporting libraries (Diffusers, ControlNet, LoRA, Automatic1111 or ComfyUI—whichever you feel gives the cleanest control). • Design an image-to-image pipeline that takes my jewelry PNGs and merges them with a mix of model photographs I will supply, plus additional royalty-free images you will source when my library doesn’t cover a needed angle, pose, or skin tone. • Calibrate prompts, masks, and inpainting so lighting, perspective, and shadows align flawlessly; no floating pendants or clipped earrings. • Output final composites in lossless PNG, 4K where possible, along with the seed, prompt, and settings for each frame so I can reproduce or iterate later. Once you have the workflow stable, deliver: 1. A short video walkthrough (screen-record or quick Loom) showing the RunPod setup and one example generation from start to finish. 2. The environment file or launch script that recreates the node exactly. 3. Ten sample composites—three necklaces, three earrings, four rings—demonstrating varied poses and skin tones. Acceptance criteria: every sample passes a casual “is this a real shoot?” eye test at 100 % zoom, with crisp edges around the jewelry, matching reflections, and no AI artefacts. If you’ve already tweaked Flux 2-dev, Stable Diffusion XL, or similar for apparel or accessory placement, that experience will let us move fast.