🐰 Mello's Definitive NSFW Zootopia LoRA (Minimax H3) (Please Read) 🦊
All showcase renders were made using my new site! https://app.melloai.art/ (If you feel like supporting my efforts while getting free/reduced access to minimax, paid apis, and upscalers, use and share my site!) (All demos showcased were t2v)
Trained on 250 clips from "a certain animated film" as well as an additional 33 clips on nsfw animations (Was meant to be published alongside the sfw version but never got around to proper documentation). It can do mostly any action you could want now. Honestly this one performs a little better than the sfw version even with sfw prompts. Its gets rid of most of the white fur bugs Judy would have around the arms and is now fully colored in. (Note: this version's NSFW stuff was more trained on human x Judy & tiger x Judy, no Nick was included so furry female autonomy will be a lot stronger than furry male, use a penis lora for better results if that is what you are after. I was going to train with nick stuff but boy does it take a while to fully caption nsfw stuff. Thus there will be further iterations in the future if there is demand or I get more free time.)
⚡ Quick Settings & Info
Trigger Word: z00t0p1a
Recommended Weight: 0.85 (Tested between 0.8 – 1.0, Motion becomes a little more stiff beyond .85)
Base Architecture: Minimax H3
Dataset: 283 video clips
Training Specs: 16,600 steps (~25 epochs)
🔗 Links & Mirrors
💾 Direct Mirrors:
The workflow I used (Still using an old version, will change workflows at some point): Link
📋 Modular Prompt Reference Guide
💡 How to use this guide: Text inside (parentheses) is meant to be replaced with your specific details. Plain text outside parentheses can stay as-is, unless you combine with a style lora.
z00t0p1a, (Character 1 ex. Judy Hopps), (Character 2 ex. Nick Wilde, Pawbert, etc.). A (shot type ex. wide side-angle, medium close-up, high-angle tracking, over-the-shoulder) shot of (describe the main focus of the video ex. a grey 3D rendered anthropomorphic rabbit and a red 3D rendered anthropomorphic fox) (NSFW DETAILS: if a character is nude or has little amount of clothing, add "nude" to the beginning of the character, ex. "nude grey 3D rendered anthropomorphic rabbit" "nude 3D rendered muscular orange anthropomorphic tiger" "nude 3D rendered white human male" etc.). (Describe their primary action, posture, and movement without detailing their specific anatomy yet ex. stand side-by-side inside a small white ZPD vehicle, runs frantically towards the right side of the camera, leans in close to the rabbit's face) (NSFW DETAILS: If the rabbit is having sex, clarify it here, ex. "The rabbit is having sex in cowgirl position with a 3D rendered white human male. The rabbit moves in a rhythmic motion up and down on the man's penis in a medium pace as she is crouched leaning slightly back on top of the man" ["pace" is customizable, slow, medium, or fast, you can also do slow-medium if you want an in-between pace. The sex position is also customizable]. Here's another ex. "The rabbit is having sex in missionary position with a 3D rendered orange anthropomorphic tiger, with the tiger's large penis visible at the bottom of the frame. The tiger slowly thrusts his penis back and forth at a slow pace into the rabbit's pussy. After 4 seconds the tiger then slowly pulls his penis out and moves it down out of frame and lets go of the rabbit's foot. The rabbit keeps her legs spread as the camera slowly moves down and towards her pussy." Another example: "One white man is at the bottom of the frame performing missionary sex on the rabbit, rhythmically thrusting his lower body and penis back and forth at a slow-medium pace into the rabbit's pussy while he is handcuffed on his right arm to the rabbit's right leg. Another white man is at the top of the frame performing oral sex on the rabbit, rhythmically thrusting his lower body and penis at a slow-medium pace as she sucks on his penis, his right arm covers the left side of the frame."). (If a character is in frame and is wearing clothes or gear, describe it here ex. The grey rabbit is wearing a blue police uniform with a black bulletproof vest. The red fox is wearing a pink floral short-sleeved shirt with a patterned tie and light grey pants) (NSFW DETAILS: If the characters are nude but wearing little clothing, add "only" to the clothing, ex. "The grey rabbit is only wearing underwear". If a character is fully nude, just say the character is "nude", ex. "The grey rabbit is nude".
IF JUDY HOPPS (THE GREY RABBIT) IS IN FRAME, ADD THESE TO THE PROMPT, IF NOT SKIP:
The rabbit has detailed fluffy white and grey fur. It has a round head with two large ears and a small nose (Optional: and a small puff tail poking out of her upper pants). Her purple eyes are large and round and (Describe eye state and expression ex. wide open with a proud, content expression; half closed with a deeply exhausted, sleepy expression; slightly narrowed with a curious expression). (NSFW DETAILS: add this sentence if her bare chest is visible "She has a bare flat chest" , if her nipples are visible then add on to the flat chest description [you can make them grey or pink], ex. "She has a bare flat chest with grey nipples visible". If the rabbit's pussy or anus is visible, you can add customizable features, ex. "Her pussy is visible and is penetrated by the man's penis as she moves up and down in a rhythmic motion." Another example: Her pussy is visible and is stretched wide as the tiger slowly thrusts his penis inside." And another example: "Her anus is visible and is penetrated by the man's penis as she moves up and down in a rhythmic motion." If the rabbit or whoever is having sex with a big penis, you can add on that her stomach slightly buldges out when the penis is deep inside her, ex. "Her belly slightly bulges out at times when the penis is deep inside her."). The rabbit (Describe where and how the rabbit is looking ex. looks straight ahead at the fox's chest level, looks back over her shoulder, looks directly at the phone screen). She (Describe her mouth and expression ex. has a closed mouth with a subtle smile expression, is talking with a big cheerful open-mouthed smile, has a slightly open mouth with a gasping expression). (If the rabbit blinks, add "The rabbit blinks [multiple times/slowly]"). The rabbit's ears are (Describe ear position ex. up, down, pinned back, in a ball, blowing back in the wind).
IF NICK WILDE (THE RED FOX) IS IN FRAME, ADD THESE TO THE PROMPT, IF NOT SKIP:
The fox has detailed fluffy reddish-orange fur. It has a tapered head with two large, dark-tipped ears and a small black nose (Optional: and a large fluffy tail poking out of his upper pants). His green eyes are large and round and (Describe eye state and expression ex. half closed with a relaxed, affectionate gaze; wide open with a completely terrified, panicked expression; hidden behind reflective sunglasses). The fox (Describe where and how the fox is looking ex. looks down toward the rabbit, looks forward unconcerned of the background action, surveys his surroundings). He (Describe his mouth and expression ex. has a closed mouth with a subtle smile expression, is talking with a smug smile expression, has an open mouth baring his teeth). (If the fox blinks, add "The fox blinks [multiple times]"). The fox's ears are (Describe ear position ex. up, back, down, flattened back).
(NSFW DETAILS: If a male character has their penis out or is visible, add these controls and characteristics in a new paragraph: The [character]'s penis is erect [can also be semi-erect or flaccid]. The [character]'s penis is light brown [any color you want]. The [character]'s firm light-brown balls are visible at the base of his penis [only add if the man's balls are supposed to be visible]. Here's a full example: "The tiger's penis is erect. The tiger's penis is white with multiple veins visible and with its skin pulled back. The tiger's pink-redish penis head is partially visible is he slowly thrusts. The tiger's huge light-brown balls are visible at the base of his penis, slightly bouncing as he thrusts.").
IF ANOTHER CHARACTER OR ANIMAL IS IN FRAME (EX. LYNX, SNAKE, BEAVER), APPLY THE SAME CHARACTERISTICS AS THE PREVIOUS SUBJECTS:
(Describe their fur/scales texture ex. The lynx has detailed, thick grey and white fur / The snake has detailed, shiny blue scales). (Describe their head shape, ears, and nose/fangs ex. It has a broad head with two large, pointed ears topped with black tufts and a small pink nose / It has a broad head with two large yellow eyes and one white fang). (Describe their eye color and state ex. His eyes are large and round and wide open with an eager expression / His yellow eyes with vertical pupils are half closed). (Describe their gaze direction). (Describe their mouth, teeth, and overall expression). (Note if they are blinking). (Describe their ear position if applicable).
IF A CHARACTER IS SPEAKING, ADD THIS SECTION (OPTIONAL):
The [character] says:
(0:00 - 0:02) "(Add tone/action in parentheses if needed ex. Slowly/Sarcastically) [Insert dialogue here]."
(0:03 - 0:05) "[Insert dialogue here]."
CAMERA, BACKGROUND, AND LIGHTING (ALWAYS INCLUDE AT THE END):
The camera (describe the camera movements here ex. is stationary with a subtle zoom in, tracks backwards rapidly with intense handheld shake to keep the running characters in frame, pans to follow the fast-moving car). The background is (describe the environment ex. a sunlit park field with soft-focus on a fountain, a dimly lit rustic wooden interior hallway, a bustling sun-drenched fish market pier). The lighting is (describe the lighting conditions ex. from soft daylight illuminating the stage evenly, from a bright warm offscreen setting sun casting long shadows, from cool blue ambient party lights). (Optional: Add post-processing effects ex. shallow depth of field, motion blur, vivid).
overall_soundscape: (Describe/add any sounds to the sound scape ex. "Nick Wilde speaking off-screen in a calm, casual tone: "You play it smart, stay off the..." accompanied by paper and binder rustles as Judy holds her notebooks." If there are no sounds just add "none")
non_diegetic_music: (Describe the soundtrack/score ex. "Lighthearted, upbeat funky score." If there is no score just add "none".)🛠️ Tech-Help & Troubleshooting
"Speech is cutoff midway when generated video finishes"
If the dialogue between characters are cut off before the characters are supposed to finish, add-on to the prompt that a post-action happens or that there is a second of pause after they finish their conversation (honestly, it's my bad for not refining the dataset for minimax interms of speech but the workaround works for now).
Also, of note is if the voices don't quite match up in terms of sounding like them, calm down on either the vocalization styles and tone or if you are doing a stylized render, do one where it's not stylized on the same seed and sync up the speech after the fact (only voice guaranteed to work is judy, all the others do have a much higher chance of failing).
"Characters are too stiff"
Yeah, minimax hates adding unprompted actions unlike wan. If you do write out your own prompts, make sure that you use an llm to help add micro-movements and expressions into the prompts to give more life to the characters.
Just feeding the 'master prompt' into an llm won't give you great out of the box gens from my findings, you should have a stylistic idea or concept feed alongside the master prompt to give better and more unique results. Make sure to make use of involved camera movements ("Camera has handheld movements as it tracks with the subject", etc.).
"Character's faces/bodies are distorted/undefined at a distance"
This seems to be a problem with minimax in general. I did most of my gens at 1280x512 and ran into this issue quite a bit even when increasing steps.
Best recommendation I have is to just zoom in the shot from a wide to either medium-wide or medium. If you like that shot a lot however, rendering at 1920x804 (or its 1080p equivalent) does seem to help a lot, tho motion has the potential to suffer (I never found myself going this route since it would take over 45 mins to do 30 steps at that quality for me).
Some Other Notes:
Speed-Up Workflows: I never tested this with any speed-up stuff so whatever optimization you do to get loras to work with those workflows, just apply that to here.
Video Duration: Going beyond 12-13 seconds can be rough, especially 15 seconds, actions become repeated and/or stiff (the official dataset only goes up to 10 seconds with most being sub 6 second clips, mostly because it was made for wan).
Repetitive Animation: Since the nsfw stuff was trained on 3D animation, most of the actions were just looped every 30 or so frames which can lead to a very reparative output in terms of the pace and roboticness of it. Easiest fix is to prompt for pace changes, camera movements, and more specific prompts for micromovements and such.