Guide 2 - Hassaku/Grapefruit

How to make good use of Hassaku – Stable Diffusion Anime Prompts.

In this guide, I will also mention our new feature of mixing different loras (mini models) together with a main model.

New? Check out our first guide here to learn about the basics of rendering on AnimeMaker.ai.

Hassaku (formerly known as Grapefruit) is a very popular anime model included in our anime maker AI. It can achieve both great and terrible results, which requires us to spend some time learning how to use it properly. A good tip from the author of this model is to include (masterpiece) and (best quality) in positive together with [worst quality] and [low quality] in negative (this also works well for other models). It is important to remember that Hassaku is strongly oriented towards NSFW content, although making great SFW renders is possible as well. Just make sure to include [NSFW], [nude], and other negative prompts like those if you don’t want a surprise.

Use this model by adding <hassaku> into your prompts.

Pros:

Wide range of achievable effects and styles.
Works well with various loras.
Vibrant colors.
Great for characters.
Perfect for NSFW content.

Cons:

Finding a good combination of prompts to obtain the desired result can be challenging.
Tends to provide bland backgrounds.
Hates rendering at sizes bigger than 640×640. It can be done, but takes patience, provides tons of garbage, and requires a lot of specific prompts.

This checkpoint merge was created by Ikena.

1. Let’s begin with a simple prompt featuring an adorable panda girl.

/render <hassaku> cute girl enjoying a picnic, panda ears, sakura flowers

How about a beautiful blue fairy in the city?

/render <hassaku> 1 girl, fairy wings, shimmering blue dress, modern city, streetlights, skyscrapers, night sky, masterpiece, cinematic [[extra limbs, extra fingers]]

Sometimes, the simplest prompts provide unexpectedly good results.

/render /seed:776439 /size:832x1024 /sampler:dpm2m /guidance:7 /images:7 little fairy in a glass jar, <hassaku>

2. Time for more advanced renders. Also, from now I will enhance the quality of every picture by using one of our upscaling commands. Either ‘/facelift’, ‘/facelift /photo’, or ‘/facelift /anime’. The prompts provided beneath every picture will render it in its original size, which can be later upscaled.

Starting prompts with ‘oil portrait‘ is becoming one of my favorites.

/render /seed:403482 /size:832x832 /images:6 /sampler:dpm2m /guidance:7 Oil portrait featuring broken automata girls, steampunk machines ((gear pupils)), machine eyes, masterpiece, best quality, cinematic lighting, highres, good saturation, color gradient, multicolored hair, [low quality, human_eyes, worst quality, lowres, NSFV, nsfv, naked, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, monochrome, sepia, muted colors, sepia, monochrome, greyscale] <hassaku>

Simple and elegant. Also, remember that you can always remove ‘seed‘ and render without it to obtain more varied results. Just keep in mind that rendering with /size:832×1024 will produce only one good picture out of ten.

/render /seed:776439 /size:832x1024 /sampler:dpm2m /guidance:7 /images:7 masterpiece, best quality, little fairy in a glass jar, good saturation, [worst quality, low quality, lowres, blurry, disfigured, deformed, extra limbs, sepia, monochrome, bad anatomy] <hassaku>

Including the names of various artists can positively affect the outcome.

/render /seed:806660 /size:1024x832 /images:6 /sampler:dpm2m /guidance:8 official game art, 1 girl, solo, pouting gamer girl with cat ear ((headphones)), deep space behind spaceship window, sitting in a gaming chair, in front of computer, furrowed forehead, brews closely together, (orange blunt_bangs), full body seen from (above), focused on playing a game, computer nerd, future room, neon lights, gamer aesthetic, strong crisp lineart, by ilya kuvshinov, krenz cushart, Greg Rutkowski, (trending on artstation, intricate details, caustics, masterpiece, detailed body, cute long shirt cropped_jacket), cell shading, [lowres, blurry, worst quality, naked, washed out, extra limbs, extra digits, distorted, disfigured, bad saturation, sepia, watercolor, monochrome, pastel, logo, signature, watermark, cloned face, two or more characters, more than two legs, more than two arms, black hair,] <hassaku>

Hassaku can easily render men if required, but it likes girls more, so make sure to use negative prompts to prevent them from appearing – increasing guidance also helps. I really like how it tried to put this guy into a suit, even gave him a necktie.

/render /seed:597282 /images:6 /guidance:12 /sampler:k_euler_a /size:832x1024 glorious visage of a victorious king, a tired man after bloody battle, (sitting on a old rock), blood splotches on armor, ruins of a gothic tower in the background, 1 man, worn out silver armor, ((mysterious and dark forest at night, rotting vegetation, thorny vines)), gloving vertical eyes, full body seen from (the side), short blond hair, (casual clothes, creepy atmosphere, horror), (biolumiescent mushrooms, good quality, detailed background, high resolution, hdr, good contrast, good saturation, ray tracing, deep shadows, sharp quality, the rule of thirds, masterpiece, official art, keyvisual, post processing), cinematic lighting, [girl, female, lowres, blurry, worst quality, extra limbs, bad quality, ugly, text, logo, signature, greyscale, sepia, monochrome, disfigured, bad anatomy, extra limbs, bokeh] <hassaku>

The negative prompts below the following picture are an excellent base to use and modify. They will increases our chances of obtaining good results.

/render /seed:836170 /size:832x1024 /images:6 /sampler:dpm2m /guidance:8 scenery, oil painting of a dark room, petite girl eating cookies, hugging a plush rabbit under one arm, full body view from ((distance)), a dark room at night, holding a small lit candle in one hand, scared, trembling hands, (stunning environment, lavishly decorated room, intricate details, ray_tracing, post processing, good saturation, (pink and yellow gradient hair), good quality, best quality, highres, ultrasharp 8k, masterpiece, highres, fascinating, keyvisual, official-art), high resolution, vivid colors, lush colors, crepuscular rays, moon light, [lowres, blurry, worst quality, NSFV, naked, extra limbs, bad quality, ugly, text, logo, signature, greyscale, bokeh, sepia, monochrome, disfigured, bad anatomy, extra limbs, bokeh, poorly drawn, bad lighting, washed out] <hassaku>

Describing face details in combination with certain words like ‘portrait‘ move the subject closer to the front. On the other hand, starting with defining the scenery and background tends to give us more scenic results.

/render /seed:746188 /images:6 /guidance:8 /sampler:dpm2m /size:832x1024 oil portrait of a wicked fae princess (sitting on a fallen tree), glowing big fairy wings, expressive angry big eyes, ((mysterious and dark forest at night, rotting vegetation, thorny vines)), gloving vertical eyes, orange pigtails, (casual clothes, creepy atmosphere, horror), (biolumiescent mushrooms, good quality, detailed background, high resolution, hdr, good contrast, good saturation, ray tracing, deep shadows, sharp quality, the rule of thirds, masterpiece, official art, keyvisual, post processing), cinematic lighting, [lowres, blurry, worst quality, extra limbs, bad quality, ugly, text, logo, signature, greyscale, sepia, monochrome, disfigured, bad anatomy, extra limbs, bokeh] <hassaku>

Making dynamic poses and action scenes is most certainly possible. Prompts like ‘running‘, ‘action‘, ‘dynamic‘, ‘movement lines‘, tend to give good results.

/render /seed:78349 /images:6 /guidance:10 /sampler:dpm2m /size:832x1024 scared monster girl, running away from monster hunters, yelling for help, dark alley, (fiery fairy wings), ((layered skirt, striped_legwear)), (creepy and horror atmosphere, sci-fi), hooded man chasing in the background, biolumiescence, whole body viewed away, detailed background, big expressive eyes, scared and hopeless, dark and overwhelming presence, feeling trapped, running for life, high resolution, hdr, great contrast, good saturation, green goo, glowing details, ray tracing, deep shadows, sharp quality, cinematic lighting, gradient yellow and black colored hair, the rule of thirds, masterpiece, intricate details, trending on artstation, [lowres, blurry, worst quality, extra limbs, bad quality, ugly, text, logo, signature, greyscale, sepia, monochrome, disfigured, bad anatomy, extra limbs] <hassaku>

Monster girls are your thing? Most of the renders I’ve made with them are 18+, but here’s a SFW example. If joined together with loras specializing in them the effect can be even better.

/render /seed:452079 /images:6 /guidance:10 /sampler:dpm2m /size:832x1024 scared monster girl ((trapped inside a giant alien device)), hitting a glass wall, yelling for help, inside alien spaceship, (fiery fairy wings), gothic-lolita dress, organic pods with eggs, (creepy and horror atmosphere), sci-fi, biolumiescence, whole body viewed from the side, strange alien machines in the background, detailed background, big expressive eyes, scared and hopeless, dark and overwhelming presence, feeling trapped, high resolution, hdr, great contrast, good saturation, green goo, glowing details, ray tracing, deep shadows, sharp quality, cinematic lighting, gradient yellow and black colored hair, the rule of thirds, masterpiece, intricate details, trending on artstation, [lowres, blurry, worst quality, extra limbs, bad quality, ugly, text, logo, signature, greyscale, sepia, monochrome, disfigured, bad anatomy, extra limbs] <hassaku>

Of course, this wouldn’t be a Hassaku guide if I hadn’t included at least one or two NSFW example. Try rendering them with the prompts provided beneath for an uncensored version.

/render /seed:379001 /images:6 /size:1024x832 /sampler:dpm2m /guidance:9 oil painting, shaming of a naked princess, little golden tiara, judging crowd in the background, close shot at distressed face from the side, crawling on the street, 1 girl with collar, 1 hooded man holding a chain, beautiful expressive eyes, crying for help, ((on all fours lead through city streets)), tension, dangerous atmosphere, torch light, city square at night after revolution, masterpiece, cinematic light, [lowres, blurry, worst quality, sword, more than 1 girl, poorly drawn, pixelated, compressed, low resolution, normal quality, low quality, bad anatomy, bad proportions, cloned face, duplicate, extra arms, extra limbs, extra legs, fused fingers, gross proportions, long neck, malformed limbs, extra legs, fused fingers, watermark, signature, username, jpeg artifacts, jpg artifacts, bad saturation, bad colors, bad contrast, sepia, monochrome] <hassaku>

/render /seed:562461 /size:832x1024 /images:1 /sampler:dpm2m /guidance:8 head, (cinematic lighting, masterpiece, best quality, perfect cinematic lighting, red bioluminescent moss and mushrooms in the background, color grading), petite ginger girl ((squating while masturbation)) on a flower field, ((a-cup breasts, small tits)), spreading legs, fingers massaging pussy, female ejaculation, moaning and panting, sexy vibe, sex on a flower field, full body view (from below), vaginal peneteration, dark forest in the background, gothic-lolita clothes, red underwear, (intricate details, astonishing background details, good saturation, ektachrome, heavy panting, blushing, NSFV, details excited expression, [futa, futanari, worst quality, low quality, greyscale, white background, more than 1 girl. big breasts, tight clothes, blurry, mutant, disfigured, sepia, monochrome, extra limbs, lowres, bad anatomy, muted colors, dog, prebuscent, kid] <hassaku>

Although I wanted three, I will gladly take four as well. What will happen next with those girls is up to your imagination.

/render /seed:360966 /images:6 /guidance:8 /sampler:dpm2m /size:1024x832 three sadistic fae (laughing behind a tree), hiding behind a tree, fairy wings, mysterious and dark forest at ((night)), rotting vegetation, vines, three small fae with gloving vertical eyes, yellowfae with curly hair, red fae with short hair, blue fae with pigtails, (creepy atmosphere, horror), (biolumiescent mushrooms, good quality, detailed background, high resolution, hdr, good contrast, good saturation, ray tracing, deep shadows, sharp quality, the rule of thirds, masterpiece, official art, keyvisual, post processing), cinematic lighting, [lowres, blurry, worst quality, extra limbs, bad quality, ugly, text, logo, signature, greyscale, sepia, monochrome, disfigured, bad anatomy, extra limbs, bokeh] <hassaku>

3. Our new feature, using loras together with a model (checkpoint). This feature opens a brand new world for everyone interested in rendering. Moreover, most loras perform unexpectedly well outside their specified area. A detailed guide can be found here.

Remember, you can mix together as many loras as you want, but only one model is used while rendering.

/render /seed:103267 /size:832x832 /images:6 /sampler:dpm2m /guidance:7 vivid digital photo of a broken automata girl, steampunk machines ((gear irises, gear pupils)), machine_eyes, mechanical eye, masterpiece, best quality, cinematic lighting, highres, good saturation, color gradient, multicolored hair, red and white, intricate background, busy victorian street, [low quality, human_eyes, normal eyes, worst quality, lowres, NSFV, nsfv, naked, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, monochrome, sepia, muted colors, sepia, monochrome, greyscale, cloned faces] <crazy-expressions:0.3> <hassaku>

Placing loras either at the beginning or at the end of prompts can affect the outcome, although some are so influential that simply including them changes everything. For example, the one below. I’ve put it at the end and lowered the weight, but the result is clear to see.

/render /seed:455712 /size:832x832 /images:6 /sampler:dpm2m /guidance:7 Oil portrait featuring broken automata girls, steampunk machines ((gear pupils)), machine eyes, masterpiece, best quality, cinematic lighting, highres, good saturation, color gradient, multicolored hair, [low quality, human_eyes, worst quality, lowres, NSFV, nsfv, naked, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, monochrome, sepia, muted colors, washed out, washed colors, greyscale, cloned faces] <cutescrap:0.1> <hassaku>

However, loras aren’t everything. Proper prompts, as always, are essential. For example, the two pictures below use the same loras, but are entirely different in style.

/render /seed:122974 /size:832x1024 /sampler:dpm2m /guidance:7 /images:6 masterpiece, best quality, pastel painting of a little fairy in a glass jar, color sketch, wearing headphones, lofi vibes, chilling on a sofa, submerged in water, wet hair, wet clothes, swimming, two butterfly wings, colorgradient, lush colors, vivid colors, hard lines, (ray_tracing, cinematic lighting, intricate details, gradient red and aqua hair), [low quality, worst quality, NSFV, naked, white background, nipples, blurry, lowres, deformed, extra limbs, greyscale, disfigured, monochrome, cloned] <girls-in-glass-jars:0.5> <animelineart:0.5> <hassaku>

/render /seed:866977 /size:832x1024 /sampler:dpm2m /guidance:7 /images:6 masterpiece, best quality, little fairy in a glass jar, 4 fairy wings, colorgradient, [low quality, worst quality, blurry, lowres, deformed, extra limbs, greyscale, disfigured, monochrome] <girls-in-glass-jars:0.7> <animelineart:0.7> <hassaku>

Setting a lora’s weight low can give us interesting results. For example, this fascinating panda Darth Vader:

/render /seed:735531 /size:832x1024 /images:6 /sampler:dpm2m /guidance:8 <darthvader:0.4> oil painting of a petite girl eating cookies, hugging a teddy bear under one arm, full body view from ((distance)), a dark room at night, holding a small lit candle, scared, trembling hands, (stunning environment, lavishly decorated room, intricate details, ray_tracing, post processing, good saturation, (pink and yellow gradient hair), good quality, best quality, highres, ultrasharp 8k, masterpiece, highres, fascinating, keyvisual, official-art), high resolution, vivid colors, lush colors, crepuscular rays, moon light, [lowres, blurry, worst quality, NSFV, naked, extra limbs, bad quality, ugly, text, logo, signature, greyscale, bokeh, sepia, monochrome, disfigured, bad anatomy, extra limbs, bokeh, poorly drawn, bad lighting, washed out] <hassaku>

Below are some unexpectedly good results from combining three different loras with <hassaku>. They’ve made this model much more friendly towards rendering in bigger sizes.

/render /seed:133504 /size:1024x1024 /images:6 /sampler:dpm2m /guidance:8 (masterpiece, best quality, perfect cinematic lighting, red bioluminescent moss and mushrooms growing in an abandoned city, color grading, intricate details, astonishing background details, good saturation, ektachromea,) shellshocked teenage girl sitting on a ruined tank, mad smile, crounching behind a concrete obstacle, (short blonde pigtails), dark alley, (a-cup breasts), techwear jacket, black gloves, freckles, shimmering hair, ongoing battle in the background, explosions, fire and destruction, holeless atmosphere, detailed eyes without life in them, dark circles under eyes, [NSFV, naked, nipples, panties, black hair, mask, worst quality, face-shot, zoomed in, low quality, greyscale, white background, more than 1 girl. big breasts, blurry, mutant, disfigured, sepia, monochrome, extra limbs, lowres, bad anatomy, muted colors, prebuscent] <urbansamurai:0.2> <crazy-expressions:1> <arknights-texas:0.3> <hassaku>

/render /seed:494580 /size:832x1024 /images:6 /sampler:dpm2m /guidance:8 vivid photo of a shellshocked teenage girl sneaking through abandoned city, crounching behind a concrete obstacle, (short blonde pigtails), dark alley, (a-cup breasts), techwear jacket, black gloves, freckles, shimmering hair, (masterpiece, best quality, perfect cinematic lighting, red bioluminescent moss and mushrooms growing in an abandoned city, color grading, intricate details, astonishing background details, good saturation, ektachromea,) ongoing battle in the background, explosions, fire and destruction, holeless atmosphere, detailed eyes without life in them, dark circles under eyes, [NSFV, black hair, mask, worst quality, face-shot, zoomed in, low quality, greyscale, white background, more than 1 girl. big breasts, blurry, mutant, disfigured, sepia, monochrome, extra limbs, lowres, bad anatomy, muted colors, prebuscent] <urbansamurai:0.2> <crazy-expressions:0.8> <arknights-texas:0.3> <hassaku>

Sometimes when you strike a perfect balance of prompts and loras, you can render literal masterpieces. The next two pictures were rendered either at /size:1216×1216 or /size:1344×1344, which shouldn’t make anything except for garbled garbage, and yet! Unfortunately, one of the loras had its tokens changed slightly due to an update, and now they’re impossible to replicate despite using the same prompts. I still decided to show them as an example that it’s worth experimenting with everything.

/render /seed:383685 /size:1344x1344 /images:3 /sampler:dpm2m /guidance:8 (masterpiece, best quality, perfect cinematic lighting, red bioluminescent moss and mushrooms growing in an abandoned city, color grading, intricate details, astonishing background details, good saturation, ektachromea), ((shellshocked teenage girl sitting on a destroyed machine)), mad smile and eyes, crounching behind a concrete obstacle, (((short blonde pigtails))), dark alley, (a-cup breasts), techwear jacket, black gloves, freckles, shimmering hair, ongoing battle in the background, explosions, fire and destruction, hopeless atmosphere, detailed eyes without life in them, dark circles under eyes, fire reflecting in her eyes, [NSFV, naked, nipples, panties, black hair, mask, worst quality, face-shot, zoomed in, low quality, greyscale, white background, more than 1 girl. big breasts, blurry, mutant, disfigured, sepia, monochrome, extra limbs, lowres, bad anatomy, muted colors, prebuscent] <urbansamurai:0.3> <crazy-expressions:0:8> <arknights-texas:0.2> <hassaku>

/render /seed:354958 /size:1224x1224 /images:4 /sampler:dpm2m /guidance:8 (masterpiece, best quality, perfect cinematic lighting, red bioluminescent moss and mushrooms growing in an abandoned city, color grading, intricate details, astonishing background details, good saturation, ektachromea), ((shellshocked teenage girl sitting on a destroyed machine)), mad smile and eyes, crounching behind a concrete obstacle, (((short blonde pigtails))), dark alley, (a-cup breasts), techwear jacket, black gloves, freckles, shimmering hair, ongoing battle in the background, explosions, fire and destruction, hopeless atmosphere, detailed eyes without life in them, dark circles under eyes, fire reflecting in her eyes, [NSFV, naked, nipples, panties, black hair, mask, worst quality, face-shot, zoomed in, low quality, greyscale, white background, more than 1 girl. big breasts, blurry, mutant, disfigured, sepia, monochrome, extra limbs, lowres, bad anatomy, muted colors, prebuscent] <urbansamurai:0.3> <crazy-expressions:0:8> <arknights-texas:0.2> <hassaku>

4. Something a little bit nerdy now, comparison of how Hassaku deals with different samplers and guidance levels. Since I was getting interesting results, I’ve decided to ramp it up all the way to 40. All tests were done by using variations of this:

/render /steps:more /seed:196340 /sampler:ddim /guidance:5 /images:1 /size:832x1024 oil painting of a cute girl from sun, detailed face, expressive detailed eyes, detailed body, great quality, deep shadows, global illumination, masterpiece, wallpaper, 8h, [[lowres, blurry, worst quality, disfigured, bad quality, mutant, mutation, ugly, extra limbs, more than two legs, more than two arms, extra digits, monochrome, sepia, logo, text, signature]] <hassaku>

Sampler:dpm2m: