<p>Close to my other lora : <a target="_blank" rel="ugc" href="https://civarchive.com/models/523485/kikis-delivery-service-studio-ghibli-style">Ghibli style (Kiki's Delivery Service)</a>, but with a retro vibe, more grainy and less vibrant.</p><h3 id="klein-9b"><span style="color:rgb(250, 176, 5)">Klein 9B</span></h3><p>Showcase images :</p><ul><li><p>Distilled model</p></li><li><p>768x1152</p></li><li><p>Euler / normal / 4 steps / CFG 1</p></li><li><p>Lora strength of 1</p></li></ul><h3 id="qwen"><span style="color:rgb(121, 80, 242)">Qwen</span></h3><p><strong><u>2512</u></strong></p><p>It was trained on 2512, but I prefer to use the original Qwen-Image model for inference.</p><p>Showcase images :</p><ul><li><p>Original Qwen-Image (BF16)</p></li><li><p>Lightning lora (4 steps)</p></li><li><p>5 steps / 1 CFG</p></li><li><p>DDIM &amp; Beta</p></li><li><p>Lora strength 0.9</p></li></ul><p><strong><u>v1</u></strong></p><p>All the showcase images have been generated without upscaling :<br />1584x1056 (portrait and landscape) with ddim / beta + Lightning lora</p><p>5 steps and CFG 1, and a lora weight of 1.<br />I've only generated 2 images for each prompt, and picked one, so it gives great results without cherry picking nor upscaling !</p><p></p><h3 id="flux"><span style="color:rgb(190, 75, 219)">Flux</span></h3><p>Trained on 29 carefully handpicked images from the movie <a target="_blank" rel="ugc" href="https://www.imdb.com/title/tt0104652/">Porco Rosso</a>.<br />With a weight of 0.8, I start the prompt with :<br /><span style="color:rgb(230, 73, 128)">ghibli style image of</span><br />All images are x1.5 upscaled with 4x_NMKD-Siax_200k (denoising strength 0.35 - 0.5).<br /><br /></p><p></p><p></p><h3></h3><h3></h3><h3></h3><h3></h3><p></p>

Retro Ghibli style (Porco Rosso)

retro_ghibli_style_klein9b.safetensors

Files with SHA256: f0db102ef29fe0cce1b552db6b3fc5df45f5f44685f2c2e89a75b02f2c535c64