This is a successor to my original Chiaki Nanami loras. I go into more detail on the intent and how this lora differs from the typical approach in the 'technical details' section. But the gist of this is that it's a lora for Chiaki Nanami as she's presented in Danganronpa 2, Danganronpa 3, Danganronpa S, and as it becomes more feasible Danganronpa 2x2.<h1 id="prompting-style">Prompting style</h1>I aim to have some level of compatibility with both danbooru tags and narrative style prompts. Along with using both tags and narrative prompts at the same time. That said a lot comes down to the nature of the model. For example, anima and chroma were trained heavily on both styles so will adapt to mixed prompts better than z-image. I provide prompts in the mixed format but you can just remove the tags and keep the descriptive text, the reverse, or just use the full prompt is and in theory at least any of them should work.<h1 id="prompts">Prompts</h1>Main keywords - (try to place at start of prompt): nanami_(danganronpa), nanami_chiakiThe reason I added the "nanami_(danganronpa)" is to offer an option to lower reliance on Chiaki style data in models that have some training data about her already. Anima in particular. I'd suggest using both triggers at once, but just going with nanami_(danganronpa) should push toward the Chiaki data in this lora and away from built-in information about her that might be in the diffusion model itself.Danganronpa 2 uniform: Chiaki Nanami is wearing her uniform from Danganronpa 2. Chiaki Nanami wearing a teal-gray hooded cardigan over a white button-up shirt with a dark teal-gray front placket. Chiaki's hair is colored a very light brown with a slight pinkish tint. Chiaki's hair ends at neck length. Chiaki's hair slightly curls outwards, though the locks on either side of her face curl inwards. Chiaki has a white Galaga hairclip pinned to a lock of her hair. She has blossom pink eyes. Chiaki wears a white button-up shirt with a dark teal-gray front placket. Chiaki's shirt has a dark line running from her collar to her pale beige pleated skirt. The black placket line runs from her shirt collar to the bottom of her shirt where it ends as it's covered by her skirt. She wears a light pink ribbon tied around her neck in a bow. This also includes a one-button dark teal-gray hooded cardigan with gray horizontal lines running across the hem, the Hope's Peak crest on the chest pocket. She also wears black thigh-high socks and light pink Mary-Jane style shoes with white soles.Danganronpa 3 Hope's Peak uniform: nanami_(danganronpa), Chiaki Nanami's wearing her brown uniform from Danganronpa 3. Chiaki Nanami's wearing her uniform from Danganronpa 3. Chiaki's hair is colored a very light brown with a slight pinkish tint. Chiaki's hair ends at neck length. Chiaki's hair slightly curls outwards though the locks on either side of her face curl inwards. Chiaki has short straight bangs split towards the right, a white Galaga hairclip pinned to a lock of her hair. Chiaki Nanami has blossom pink eyes. Chiaki's uniform consists of a white button-up dress shirt with a red ribbon tied in a bow at the collar, with a warm brown blazer with a cream-colored hooded cardigan underneath her blazer. , 1girl,blunt_bangs,brown_skirt,collared_shirt,danganronpa_(series),danganronpa_3_(anime),dress_shirt,flipped_hair,galaga,hair_ornament,hope's_peak_academy_school_uniform,jacket,long_sleeves,medium_hair,nanami_chiaki,neck_ribbon,open_clothes,pink_eyes,pleated_skirt,red_ribbon,ribbon,uniform,shirt,shirt_tucked_in,short_hair,skirt,spaceship_hair_ornamentDanganronpa S swimsuit: nanami_(danganronpa), Chiaki Nanami is a woman whose hair is colored a very light brown with a slight pinkish tint. Chiaki's hair ends at neck length. Chiaki's hair slightly curls outwards though the locks on either side of her face curl inwards. She has bright pink eyes and wears a white and red galaga spaceship shaped hair clip on the right side of her head. She's wearing her Danganronpa S swimsuit which is a white bikini top with ruffled trim along the edges. danganronpa_(series), danganronpa_2:_goodbye_despair, nanami_chiaki, 1girl, bikini, breasts, cleavage, collarbone, frilled_bikini, frills, large_breasts, medium_hair, mole, mole_on_breast, spaceship_hair_ornament,swimsuitDanganronpa 10th anniversary outfit: nanami_(danganronpa), Chiaki Nanami is wearing her Danganronpa 10th anniversary dress. Her hair has short bangs and two small buns on either side of her head. Her eyes are light pink. very light brown with a slight pinkish tint. She is wearing a dark pink off-the-shoulder gown with an overlapping sweetheart neckline a wide dark red sash around her waist tied into a bow at the back and a full skirt. She also wears a simple red choker around her neck. She wears a single small galaga spaceship themed hair clip, 1girl,bare_shoulders,blunt_bangs,braid,breasts,choker,collarbone,danganronpa_(series),danganronpa_10th_anniversary_costume,danganronpa_2:_goodbye_despair,dress,emerjk-9,galaga,hair_ornament,medium_breasts,nanami_chiaki,off-shoulder_dress,off_shoulder,official_alternate_costume,pink_choker,pink_dress,pink_eyes,pink_hair,short_hair,short_sleeves,spaceship_hair_ornamentNote, the following Asuka Zero outfit seems to have not been picked up very well in the training. The Chroma lora generally does a bad job of rendering it. The Z-Image turbo lora is serviceable but not especially accurate with it. The Z-Image Base lora can render it properly in base if you use long hair as a negative prompt. But if you're using the z-image base lora in turbo her hair end up too long and it'll stick out through the back of her hood.Asuka Zero outfit (The ninja looking outfit): nanami_(danganronpa),nanami_chiaki, asuka_zero, Chiaki Nanami is wearing her Asuka Zero, outfit. She is seated in a kneeling position on a wooden floor her torso facing forward her head is tilted slightly forward. Her right arm is positioned forward bent at the elbow her right hand grips the hilt of a katana that rests on the floor angled diagonally across her lap. Her left arm hangs relaxed at her side. She wears a vibrant red hooded cloak adorned with a leaf pattern the hood has cat ears and red ties hanging down her chest. Beneath the cloak she wears a dark red bra top and matching dark red shorts. Her top leaves her cleavage exposed. The skin of her shoulders and upper arms are exposed. Her midriff and arms are bare other than a covering of thin fishnet material. She has dark red armored forearm guards though her shoulders and upper arms are bare. She wears knee-high black boots accented with gold trim. Her expression is calm her eyes are wide and pink her gaze directed straight forward. The scene is illuminated by the soft glowing yellow light of two lanterns creating strong highlights and deep shadows suggesting a nighttime setting. A full moon shines brightly in the upper right background. The art style is anime with detailed shading and soft colors emphasizing pink cherry blossoms and the dark blues of the distant traditional Japanese-style buildings and mountains. , detailed_background, danganronpa_(series), alternate_costume, 1girl, solo, medium_hair, brown_hair, pink_eyes, japanese_armor, red_armor, lamellar_armor, shoulder_armor, arm_guards, leg_armor, fishnet_stockings, fishnet_clothing, thigh_highs, exposed_shoulders, midriff, navel, tassets, ninja, armor_skirt, segmented_skirt, mon_(crest), clan_symbol, breasts, medium_breasts, cleavage, bikini_armor, mole_on_breast, hood_up, hair_ornament, hairclip, galaga, green_crest, green_mon, ancient_japanese_castle, cherry_blossoms, falling_petals, stone_lanterns, holding_katana, moonlight, mist, traditional_architecture, kneeling<h1 id="style-prompts">Style prompts</h1>I worked in a few styles that work with varying levels of success on a model by model basis. Sadly, the Chroma lora didn't seem to pick them up especially well. But it's worth experimenting with.The style from Danganronpa games: The image is in the danganronpa sprite style.The style from the Danganronpa 3 anime: danganronpa_3_(anime)_(style), This is an anime screencap from Danganronpa 3.Danganronpa S game interface (kinda unpredictable in adherence to the actual Danganrompa S GUI layout). This works best with the Krea 2 version: nanami_(danganronpa), Chiaki Nanami, nanami_chiaki, komatsuzaki_rui_(style), Chiaki Nanami wearing her uniform from Danganronpa 2. The image is in the danganronpa sprite style and danganronpa S GUI style. Chiaki has short, dusty pink hair that slightly curls outwards, though the locks on either side of her face curl inwards. She has short straight bangs split towards the right, a white Galaga hairclip pinned to a lock of her hair. She has blossom pink eyes. Chiaki Nanami is wearing a white button-up shirt with a dark teal-gray front placket and a light pink ribbon tied around her neck in a bow. She's also wearing a dark teal-gray hooded cardigan with gray horizontal lines running across the hem and sleeves. INSERT IMAGE BACKGROUND AND OTHER ELEMENTS HERE. There is the standard danganronpa S GUI style dialogue box at the bottom of the image. 'Chiaki Nanami' is at the top of the dialogue box with a blue background and white stars. The text within the dialogue box reads "INSERT YOUR TEXT HERE" The speech bubble has a black speech bubble at the top with the text "Chiaki Nanami" in white, and the speech bubble has blue and white stars around its perimeter. The image is from the game Danganronpa S.<h1 id="model-specific-notes">Model Specific Notes</h1>Shared issues: So far both the Chroma and Z-image loras have terrible performance trying to render Monomi. I found that pretty disappointing since she's heavily represented in the training data 'and' she has a strong narrative link to Chiaki.Chiaki's backpack is a recurring problem. It's easily the most common visual issue I've seen when using both the Chroma and Z-Image version of this lora. It's both heavily represented in the training data 'and' erratically drawn even in official art. Plus it's paired with her so often that a model's justified in just assuming it'll be there. So it might be drawn incorrectly or appear when you didn't prompt for it.Krea 2: The most significant issue with the Krea 2 lora is Chiaki's hairclip. Krea 2 wound up picking up details in the training data more clearly than expected. Good for some things, but there's a good deal of inconsistency with the design and presentation of the hairclip in both official and unofficial images. The end result is that it's essentially a roll of the dice on how well it'll come out. Krea 2 is promptable to the point where one can almost force it with proper descriptive text. But in general I've found that just going with the "pixel art galga spaceship hairclip" phrasing tends to still work best.Z-Image: For whatever reason I've noticed that I seem to need more steps with the z-image version of this lora than usual when generating images. So if you're getting results that seem subpar compared to the example images in the gallery then try pushing up your step number.I generally use a strength of 1 for this lora with z-image but it seems to hold together pretty well with lower strengths while gaining a bit of extra flexibility. It might be worth trying with a default strength of 0.8 depending on your preferences. Possibly lower if you're working with additional style loras.I've also found that I can get some nice results pairing this lora with my <a target="_blank" rel="ugc" href="https://civarchive.com/models/1738285/nononononoway-style">Nononononoway Style</a> lora and using the Danganronpa 3 anime style keywords. Using both loras with reduced strength. Something like: &lt;lora:Nononononoway_zImageTurbo:0.7&gt;, 86_(nononononoway), &lt;lora:chiakiNanamiZimage:0.4&gt;, nanami_(danganronpa), danganronpa_3_(anime)_(style), This is an anime screencap from Danganronpa 3.Obviously multi-character gens are pretty hacky at the best of times. I was initially optimistic about how well this dataset and z-image might work there and have a manageable number of images in the set with Chiaki and other characters. And it was working during much of the training....and then right about the time it finally caught on to Chiaki's hairclip design the multicharacter proficiency burned out. In particular Hajime, Tsumugi, and Nagito are in there to the point where I felt there was a good chance they'd be usable alongside Chiaki. But the results seem pretty sporatic.Chroma: I also like how the Chroma lora pairs with the <a target="_blank" rel="ugc" href="https://civarchive.com/models/1738285?modelVersionId=2146071">Nononononoway Style</a> lora. The impact isn't nearly as dramatic though. A strength of 1 for both the Chiaki chroma lora and the Nononononoway lora provides an aesthetic boost but to a lesser degree than with the z-image combo. I've found that to be the case in general with this lora and styles. It 'really' likes the general averaging out of Chiaki from the base chroma model used and the dataset as a whole. Much harder to push out of that comfort zone.<h1 id="technical-details">Technical Details</h1>My original Flux Chiaki Nanami lora was heavily built around official material. The vast majority of it is still in this lora's dataset. However, I wanted to push up the amount of fan art as well if I felt it reached a similar quality level and consistency of the official works. I also worked in some synthetic data. In essence I rebuilt the dataset to a large extent. I think the only elements that I carried over totally unmodified were the sprites from Danganronpa 2.My main goal is trying to maintain the element of non-verbal storytelling found in the games. There's a tremendous range of emotions on display in danganronpa's sprites so I took considerable care in hand labeling every sprite in hopes of being able to better evoke that from prompts. Of course theory and practice are often pretty distinct from each other. But I'm fairly happy with the results after some early testing.I tried two different approaches to the training with the initial Chroma build which is why there's a small version. Ultimately I feel the standard (largest) Chroma lora is the best but I'm also unsure if there's enough of a difference to matter for anyone who's low on vram. If in doubt with chroma I suggest just using the larger standard lora.For the Chroma lora, I've done most of my testing using the Cat Collar model as it's fine tuned for anime. But the chroma lora works fine with the standard Chroma model as well.The Krea 2 lora came after a huge revision to the dataset. Though I don't think there would be enough impact for the other models to be worth retraining for them. That said it should be a little more versatile as a result.

Chiaki Nanami [Danganronpa]

chiaki_nanami_MOSS-image_512_r64.safetensors

MOSS-Image v2.0

Krea v2.0

Files with SHA256: 47d11f203706d40d980ba49a297f9115a7dcd429b79702224b03f8fcb05a465c