<h2>Jessy</h2><p>A photoreal character LoRA for <strong>Krea 2</strong>, trained on the Raw base. Trigger it with <code>Jessy_lds</code>.</p><p>Like all my character LoRAs it is <strong>face-centred</strong>: the training locks the facial identity and deliberately leaves body, build and framing free for your prompt, so she stays recognisable whether you ask for a close-up portrait or a full-body shot, and you keep full control of everything else.</p><h3>Settings that work</h3><ul><li><strong>Strength 1.0</strong> on the Krea 2 Raw base - that is what the samples use.</li><li>Around <strong>16 steps, CFG 1.0</strong>, 9:16 for portraits.</li><li>It carries across community Krea 2 checkpoints too. The sample sheets show the same seed and prompt run through five different bases side by side, at strength 0.0 and 1.0, so you can see exactly what the LoRA adds rather than guessing.</li></ul><h3>Training</h3><p>2500 steps, 59 epochs, Krea 2 Raw. Dataset built and captioned with <a href="https://github.com/perfectgf/lora-dataset-studio" rel="ugc">LoRA Dataset Studio</a>, a free and self-hosted tool I build for exactly this: turning a handful of references into a clean, captioned dataset and training the LoRA from it. The comparison sheets above were produced by its Test Studio.</p><p>She is a synthetic character - not a real person.</p>

Jessy - Krea 2 character LoRA

<p>Final save at <strong>2500 steps</strong> (59 epochs), trained on the Krea 2 Raw base. Face-centred identity lock; body and framing stay prompt-driven.</p>

Files with SHA256: f017ad8f13a05587f65acaa8fe9f1db7e0ce058561075d288d8c45d8b53274cd