<h3 id="overview-518s0fuma">Overview</h3><p><em>This F1 LoRA aims to reproduce the distinctive style of </em><a target="_blank" rel="ugc" href="https://en.wikipedia.org/wiki/Jean_Giraud"><em>Jean Giraud</em></a><em>, a famous French artist who made significant contributions to the Franco-Belgian bandes dessinées (BD) tradition and is better known under the pseudonym Moebius. His unique style, characterized by intricate line work, expansive worlds and visionary, surreal and imaginative designs left an indelible mark on the genre.</em></p><p></p><h3 id="usage-kwayt171b">Usage</h3><p>All showcase images were generated with the following settings, which are recommended:</p><pre><code>Prompts: started with "In style of Moebius."
Model: flux1-dev (fp8e4m3fn)
Text Encoder: t5pxxl_fp16
Sampler: euler
Scheduler: 24 steps (normal)
Flux Guidance: 4
LoRA Strength: 1</code></pre><p></p><h3 id="training-li63gnit5">Training</h3><p>LoRA was created using the <a target="_blank" rel="ugc" href="https://github.com/ostris/ai-toolkit">ai-toolkit</a> on an RTX 3090. The fine-tuning process utilized 362 images of varying quality, all autocaptioned with Florence-2 ("&lt;MORE_DETAILED_CAPTION&gt;" task), with addition of "<span style="color:rgb(34, 34, 34)">In style of Moebius. " into the beginning of each caption</span>. Training hyperparameters were:</p><pre><code>Rank/alpha: 32/32
Batch Size: 1
Steps: 6000
Learning Rate: 1e-4
Optimizer: adamw8bit</code></pre>

Files with SHA256: 5ed49b8335146ee728a5d4f1db17f88caa21890f2e297399a761479626851e93