I originally wanted to recreate my ropicsXL style LoRA on Krea 2 because I was curious to see how the same general style mix would behave on a model other than Illustrious XL.For this version, I changed the training recipe and used a much smaller dataset. I do not think it captures the style as strongly or as consistently as the original ropicsXL, but it still has its own appeal and can produce some very nice western comic-style results.<h2 id="recommended-strength">Recommended Strength</h2>Recommended LoRA strength:0.65–1.5I personally use it between:1.0–1.5Lower strengths preserve more of Krea 2’s default look. Higher strengths push the line work, rendering, and comic-book influence more strongly, but they may also reduce prompt flexibility.<h2 id="trigger-phrase">Trigger Phrase</h2><pre><code>ropicsK2, a digital comic illustration with high-contrast rendering and ink hatching</code></pre>The high-contrast rendering and ink hatching phrases may be optional. I kept them because they are key parts of the look I wanted from the mix, but I have not fully tested how much the result changes if you remove them.The dataset does not contain specifically tagged characters, so the LoRA is not intended to reproduce any particular character from either artist’s work.<h2 id="prompting-notes">Prompting Notes</h2>Krea 2 behaves differently from Illustrious XL and many Stable Diffusion models. At least for style mixes, it is not really a simple plug-and-play case where you add the trigger token and instantly get the exact intended style.The output can change quite a lot depending on the rest of the prompt. I found that adding a more detailed style phrase helps the LoRA produce the look I wanted:<pre><code>polished semi-realistic comic-book style, modern western comic cover finish, clean line art, selective hatching, painterly shading, smooth airbrushed rendering, glossy highlights, rich saturated colors, crisp silhouettes, elegant pin-up anatomy, and local-color hair detailing with darker same-hue strand lines</code></pre>Feel free to shorten, modify, or experiment with this phrase. You may prefer stronger line art, heavier hatching, flatter comic coloring, or a more painterly finish.It also helps to place the trigger phrase and important style instructions near the beginning of the prompt rather than burying them at the end.<h2 id="basic-prompt-example">Basic Prompt Example</h2><pre><code>ropicsK2, a digital comic illustration with high-contrast rendering and ink hatching, polished semi-realistic comic-book style, modern western comic cover finish, clean line art, selective hatching, painterly shading, smooth airbrushed rendering, glossy highlights, rich saturated colors, crisp silhouettes, elegant pin-up anatomy, and local-color hair detailing with darker same-hue strand lines. An adult female superhero stands on a rooftop at night with a confident expression, her hair and cape moving in the wind, surrounded by dramatic city lights.</code></pre><h2 id="comic-panels">Comic Panels</h2>The LoRA can produce comic pages and multi-panel compositions. Krea 2 already seems capable of understanding comic-panel layouts fairly well, so you can describe the number of panels and what should happen in each one.Example:<pre><code>A three-panel western comic page. Panel one shows the heroine entering an abandoned laboratory. Panel two is a close-up of her noticing claw marks on the wall. Panel three shows a large creature attacking from behind.</code></pre>Panel consistency will still vary, especially when several characters or more complicated actions are involved.<h2 id="known-limitations">Known Limitations</h2>Krea 2 can sometimes simplify distant characters in scenes with multiple people. Background characters may become generic, cartoony, or resemble low-detail AI meme-style figures.I suspect this is mostly a Krea 2 behavior rather than something caused purely by the LoRA, but I cannot confirm that for sure.For better results:<ul><li>Keep important characters closer to the camera.</li><li>Avoid placing too many detailed characters in a single scene.</li><li>Describe each character clearly.</li><li>Use separate panels when multiple actions or characters are needed.</li><li>Generate at higher resolutions when possible.</li><li>Expect hands, faces, and costumes to lose detail at a distance.</li></ul>The LoRA may also vary a lot depending on the prompt. Some prompts show the style strongly, while others fall back more toward Krea 2’s default rendering.<h2 id="training-information">Training Information</h2><ul><li>Dataset: 100 images</li><li>Base model: Krea 2</li><li>Training resolution range: 512–1024</li><li>Batch size: 4</li><li>Learning rate: 1e-4</li><li>Weight decay: 1e-4</li><li>Training steps: 2,400</li><li>Network dimension: 32</li><li>Network alpha: 32</li><li>Optimizer: AdamW8Bit</li><li>Timestep type: Linear</li><li>Timestep bias: Balanced</li><li>Loss type: Mean Squared Error</li><li>Caption dropout: 0.05</li><li>Trainer: Ostris AI Toolkit</li><li>Remaining settings: AI Toolkit defaults</li></ul>The training was completed on a rented GPU pod because Civitai’s trainer did not allow me to use a batch size higher than 1.<h2 id="final-notes">Final Notes</h2>This is an experimental recreation rather than a direct conversion of ropicsXL. Krea 2 interprets prompts and style training differently, so it will not behave exactly like the Illustrious XL version.Try different LoRA strengths, shorten or expand the style phrase, and experiment with line art, hatching, coloring, lighting, and rendering instructions. The model is highly prompt-dependent, and even small wording changes can produce noticeably different results.

RopicsK2 — Kenneth Rocafort × Predondo Western Comic Style Mix

Prompting NotesKrea 2 behaves differently from Illustrious XL and many Stable Diffusion models. At least for style mixes, it is not really a simple plug-and-play case where you add the trigger token and instantly get the exact intended style.The output can change quite a lot depending on the rest of the prompt. I found that adding a more detailed style phrase helps the LoRA produce the look I wanted:<pre><code>polished semi-realistic comic-book style, modern western comic cover finish, clean line art, selective hatching, painterly shading, smooth airbrushed rendering, glossy highlights, rich saturated colors, crisp silhouettes, elegant pin-up anatomy, and local-color hair detailing with darker same-hue strand lines</code></pre>Feel free to shorten, modify, or experiment with this phrase. You may prefer stronger line art, heavier hatching, flatter comic coloring, or a more painterly finish.It also helps to place the trigger phrase and important style instructions near the beginning of the prompt rather than burying them at the end.

Files with SHA256: 5240cf8018f8a8fb587efe12420b38d5abab9e716c44a7b8f0face250c7596a2