This model is a reimagining of the fictional character <a target="_blank" rel="ugc" href="https://civarchive.com/models/234362/manuela-vicious">Manuela Vicious</a>, by <a target="_blank" rel="ugc" href="https://civarchive.com/user/Tower13Studios">Tower13Studios</a>.<h2 id="flux.2-klein-9b">Flux.2 Klein 9B</h2><ul><li>Checkpoint: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2322332/flux2-klein">Flux.2 Klein 9B</a></li><li>LoRA strength: 0.8–1.6</li><li>Sampler: euler, euler_ancestral, res_multistep, or dpmpp_2s_ancestral</li><li>Scheduler: simple or beta</li><li>Steps: 4–12; 30–50 for base model</li><li>CFG Scale: 1, or try 1.2–3; 2–5 for base model</li><li>Trigger word (for superior results): ATRX_MANUELA_FLUX2KLEIN9B</li></ul><h2 id="z-image-turbo">Z-Image —Turbo</h2><ul><li>Checkpoint: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2168935/z-image-turbo">Z-Image Turbo</a>, or try <a target="_blank" rel="ugc" href="https://civarchive.com/models/2342797/z-image-base">Z-Image [Base]</a></li><li>LoRA strength: 1–1.6</li><li>Sampler: dpmpp_2s_ancestral</li><li>Scheduler: simple or beta</li><li>Steps: 4–12</li><li>CFG Scale: 1, or try 1.2–3; CFG Norm: 1–1.12</li><li>Shift: 7 (first sampler) and 0.6 (second sampler)</li><li>Trigger word (for superior results): ATRX_MANUELA_ZIMAGE</li><li>Negative prompt (as needed): "Asian woman, black hair"</li></ul><h2 id="qwen-image-2512">Qwen-Image 2512</h2><ul><li>Checkpoint: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2268063/qwen-image-2512">Qwen-Image 2512</a></li><li>LoRA strength: 0.8–1</li><li>Sampler: res_multistep</li><li>Scheduler: simple</li><li>Steps: 20+, or 4–8 with <a target="_blank" rel="ugc" href="https://huggingface.co/lightx2v/Qwen-Image-2512-Lightning">lightning LoRA</a></li><li>CFG Scale: 2–5, or 1 with lightning LoRA; CFG Norm: 0.94–0.98</li><li>Shift: 3.1</li><li>Trigger word (strictly required!): ATRX_MANUELA_QWEN2512</li></ul><h2 id="qwen-image">Qwen-Image</h2><ul><li>Checkpoint: <a target="_blank" rel="ugc" href="https://civarchive.com/models/1864281">Qwen-Image</a></li><li>LoRA strength: 0.75–1</li><li>Sampler: euler</li><li>Scheduler: simple</li><li>Steps: 15+, but 20+ is better, or 4–8 with <a target="_blank" rel="ugc" href="https://huggingface.co/lightx2v/Qwen-Image-Lightning">lightning LoRA</a></li><li>CFG Scale: 2–5, or 1 with lightning LoRA; CFG Norm: 0.8–0.9</li><li>Shift: 3.1</li></ul><ul><li>IMG-2-IMG denoise: 0.2–0.6</li></ul><h2 id="flux.1">Flux.1</h2><ul><li>Checkpoint: <a target="_blank" rel="ugc" href="https://civarchive.com/models/618692/flux">Flux.Dev 1.0</a></li><li>LoRA strength: 0.8–1.4</li><li>Sampler: deis or euler</li><li>Scheduler: beta or simple (or ddim_uniform for different look, but add steps)</li><li>Steps: 15+, but 20+ is better, or 8–12 with <a target="_blank" rel="ugc" href="https://civarchive.com/models/876388/flux1-turbo-alpha">turbo LoRA</a></li><li>CFG Scale: 2–5</li></ul><ul><li>IMG-2-IMG denoise: 0.65–0.85</li></ul><h2 id="pony">Pony</h2><ul><li>Checkpoint: <a target="_blank" rel="ugc" href="https://civarchive.com/models/153568?modelVersionId=832353">Real Dream SDXL Pony 12</a></li><li>LoRA strength: 0.6–1</li><li>Sampler: DPM++ SDE or euler</li><li>Scheduler: turbo or simple</li><li>Steps: 20–30+, or 8–12 with <a target="_blank" rel="ugc" href="https://civarchive.com/models/1181892?modelVersionId=1330126">turbo LoRA</a></li><li>CFG Scale: 2–5</li><li>ADetailer recommended</li></ul>

Manuela —Reimagined

Pair with a realism LoRA. Can produce artistic nudes pretty well on its own. Works well for img-2-img with denoise 0.2–0.6.Trained using AI Toolkit following <a target="_blank" rel="ugc" href="https://www.youtube.com/watch?v=gIngePLXcaw">this official tutorial</a>, but the learning rate seemed to be too low so I continued rerunning with additional steps until it was something over 20,000, I think? At some point in the training, I also recall changing the linear rank from 16 to 32 and some other adjustments.The dataset included 44 diverse images—different settings, hair styles, emotions, etc.—with almost all being upper body or closer and about one-fourth nudes. The captions were drafted mostly by <a target="_blank" rel="ugc" href="https://chat.mistral.ai/">Mistrel Le Chat</a>—which didn't seem to mind the nudity—and <a target="_blank" rel="ugc" href="https://chat.qwen.ai/">Qwen</a> at about 100 words, including the trigger word ATRX_MANUELA_QWEN, although the LoRA seems to respond without it. The captions described all of the elements not associated with the character or of temporal nature (e.g., hair style and clothing).

Files with SHA256: 7aed721c06921eeb659339d89d5f41480ae57c89918f3ba7e7d9d458d3fec3d4