<p>A style lora from the work of criis-chan. My intent was to create a booster pack of sorts to push how well illustrious models would handle Danganronpa characters. Criis-chan is, in my opinion, one of the highest quality artists who both specializes in Danganronpa and has a huge library of public works to sample from. While criis-chan is typically represented with artist tags in Illustrious models, it's not to the extent to which lesser known characters will be usable. This lora should take care of that and provide a wider range of Danganronpa characters while also providing options for changing costumes, looks, style, etc to go along with criis-chan's work. It won't provide 'every' Danganronpa character, but it at least expands the range of them while pushing for authenticity and quality.</p><p></p><h1 id="chroma-notes">Chroma notes</h1><p>I wasn't satisfied with my Chroma 1.0 lora so I tried again and the result is the chroma 1.1 lora. I do think the 1.1 version is a big improvement over the original chroma 1.0. However, it does have some issues. The biggest is that text will often appear if you're not using a negative prompt. Of course using negative prompts will also constrain you a bit in terms of lowered step loras. If you want to use one of these loras I'd suggest the <a target="_blank" rel="ugc" href="https://civarchive.com/models/2032955?modelVersionId=2300950">rank 48 Chroma Flash Heun</a> lora. Then go for a cfg of 1.3 or at most 1.4. And for the negative prompt use at least the following: "signature, watermark, signature, twitter_username, artist_name,watermark, subtitle,tumblr_username". I've had acceptable results with a sampler of Heun/Beta and step count of 10 though obviously more is better. Aside from style improvements, the 1.1 Chroma lora also has improved ability to prompt for Danganronpa characters over the 1.0 version. Still not great but a nice push in the right direction at least. You can push the style of the 1.1 lora into a more polished form by adding something like "A modern digital anime illustration in the style of a visual novel key image. The lighting is dramatic. The art features colored line art soft cel-shading mixed with airbrushed gradients and highly detailed expressive eyes with multiple highlights." to the prompt.</p><p></p><h1 id="usage">Usage</h1><p>Illustrious already has a strong knowledge of danganronpa characters so compatibility there is already a given. The chroma lora comes close to being able to replicate characters from the dataset rather than just the style alone. Not perfectly, there'll inevitably be issues, but close. For example when using 'only' the chroma lora Chiaki Nanami's distinctive backpack might just be a normal backpack, Junko Enoshima's bear hair orniments might not come out correctly, etc. The gallery should have some examples of how mixing in even a standard flux character lora with this style lora can work in tandem with this one. But in general 0.1 to 0.3 strength for a character lora, even a flux one, should help there.</p><p></p><p>However, in general, this lora really is best seen as a style lora and not something specifically for danganronpa characters in isolation.</p><p></p><h1 id="tags:">Tags:</h1><p>The main tag you'll need is ccdanganart. Using criis-chan as an artist tag isn't required, it might help or it might hurt. The arist has made a number of pieces emulating the official Danganronpa style, and you can push in that direction with a tag of <span style="color:rgba(255, 255, 255, 0.85)">komatsuzaki_rui_(style). Whether you use the rui tag will heavily influence how "danganronpa" the results are.</span></p><p></p><h1 id="compatibility:"><span style="color:rgba(255, 255, 255, 0.85)">Compatibility:</span></h1><p><span style="color:rgba(255, 255, 255, 0.85)">It seems to do well with my Danganronpa character loras. When doing so I'd suggest lowering the character lora strength to around 0.3 while keeping this lora at 1.0.</span></p><p></p><h1 id="negative-tags:"><span style="color:rgba(255, 255, 255, 0.85)">Negative tags:</span></h1><p><span style="color:rgba(255, 255, 255, 0.85)">This lora does have a tendency to overlay text on top of the images. You can lower the chances of that happening by using these tags as negatives: </span>signature, watermark, signature, twitter_username, artist_name,watermark, subtitle,tumblr_username</p>

Ccdanganart (style and Danganronpa booster)

<p>Retrain of the Chroma 1.0 lora to see if I can boost the quality.</p>

Files with SHA256: 7959849318ef89965c962cbb834a9685e99ac370b64e403c768997618744751e