<h1 id="dandd-battlemaps-af2wa21j1">D&amp;D battlemaps</h1><p>The model is trained on fantasy battlemaps, so it works the best for medieval fantasy setting but not only limited to this.</p><p>To install just put both checkpoint and config files to your model folder and keep same names for them.</p><h2 id="resolution-26g4ngorr">Resolution</h2><p><strong>You need to use width and height not less than resolution 1024</strong> but you can make it larger and that works perfectly just producing bigger battlemap.</p><h2 id="dimensions-9qh5wh4tx">Dimensions</h2><p>You can assume battlemaps dimensions as approximately 24x24 squares (120x120 feet) for normal dimensions of specific model, for example:</p><ul><li><p>1024x1024 generation is 24x24 battlemap</p></li><li><p>1536x1024 generation is 36x24 battlemap</p></li><li><p>2048x1024 generation is 48x24 battlemap</p></li></ul><p>And so on. But it can be depending on concept.</p><p>Grid can be produced unintentionally but it very unlikely and also depends on concept.</p><h2 id="prompting-o8vcvos1l">Prompting</h2><p>This model has no trigger words but you can refer to the graph below. X axis is tag words and Y axis is how often it was used in dataset. It is not complete tag list but these should work more or less reliable. <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/bb14e5b2-a7bb-46da-b743-41443c5e13d9/width=525/bb14e5b2-a7bb-46da-b743-41443c5e13d9.jpeg" />Captioning on dataset was made not with natural language but with tags, for example "Forest, river, bridge, road". However I did not found significant difference between these prompting types on tests. Feel free to try both and share what you prefer.</p><h2 id="settings-iwoj23bri">Settings</h2><p>Steps: 30</p><p>Sampler: DPM++ 2M SDE</p><p>Schedule type: Karras</p><p><strong>CFG scale: 2.5</strong></p><p>Size: 1024x1024</p><p>Sampler - I did not tried, so you can experiment and share your favorite.</p><p>CFG - you have to use pretty low values between 2 and 3. On 7 it feels really overtrained and burnt but hopefully later i will fix it. It is especially bad with high contrast concepts, like "Lava" or "Winter" and in these cases go CFG 2. In others I prefer 2.5 or 3.</p><h2 id="tips-and-tricks-wvsq56o8w">Tips and Tricks</h2><ul><li><p>Try <a target="_blank" rel="ugc" href="https://github.com/hako-mikan/sd-webui-regional-prompter">regional prompter</a>. You can produce something like this in one generation:</p></li><li><p><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/f09ca6b8-5f33-4d5a-8af4-9e824fe00178/width=525/f09ca6b8-5f33-4d5a-8af4-9e824fe00178.jpeg" />Try draw a simple scheme of map and img2img it with denoise about 0.75</p></li></ul><p><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/4f295f28-6b9c-49e3-8078-39dda4f0c6a4/width=525/4f295f28-6b9c-49e3-8078-39dda4f0c6a4.jpeg" />-&gt;</p><ul><li><p><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/550f9abf-72bd-453e-b647-9d9187741b1d/width=525/550f9abf-72bd-453e-b647-9d9187741b1d.jpeg" />Try make scale "Smaller" with same composition. Resize generated image (for example x2) with denoise 0.5.</p></li></ul><p>This one looks like 24x24</p><p><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/faeb442d-12c0-4576-94df-eb0906ff5c15/width=525/faeb442d-12c0-4576-94df-eb0906ff5c15.jpeg" />And this one is around 48x48 -&gt;</p><h2 id="p.s.-nb6feqdae"><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/42c1d10a-cda8-4f6d-86f8-27d761080001/width=525/42c1d10a-cda8-4f6d-86f8-27d761080001.jpeg" />P.S.</h2><p>If you want to support new models release, you can share battlemaps you have. That would be perfect if they will be same style, good quality and no grid. Also I really want to try something not medieval fantasy, for example cyberpunk or postapocalypse like fallout.</p>

Files with SHA256: 0776bd3d06c59292d133249bada6a0516589712f1c8135513756d614623f4905