text2vid img2vid H3 Minimax - Fully uncensored local AI video generator released today

As others have stated, prompt structure is crucial. Here's a distilled version of the prompt writing guides we can refer to for both the base and ref models. Good luck, have fun.



Base Model
The base model uses a two-part structure: a specific alignment instruction (for keyframe tasks) followed by three core fields.

Supports zero, one, or two input images.

No image input: Text-to-video mode

One image input: First-frame-to-video or last-frame-to-video generation

Two image inputs: First-and-last-frame-to-video generation

T2VA (Text-to-Video-Audio)
No alignment instruction is used for T2VA.

integrated_multimodal_description: [Shot 1] [Style], [Initial Composition], [Subject Appearance/Position], [Environment/Props]. [Camera Motion Type] with [Amplitude] at [Speed] [Direction/Target]. [Subject ID] ([Speaker ID]) [Action/Delivery]: [Language] [Verbatim Dialogue]. [Shot 2] At [00:SS.mmm], the camera cuts to [Composition]. [Action/Audio Continuity].



overall_soundscape: [1–4 sentences summarizing ambient sounds, physical action sounds, and non-verbal human sounds across the entire video].



non_diegetic_music: [1–3 sentences describing instrumentation, speed, rhythm, and dynamic changes. No abstract mood words].

I2VA (Image-to-Video-Audio)
For the target video, at 0.00 seconds into the target video, (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] [Style derived from image], [Composition from ]. [Subject/Scene anchors from ]. [Action onset] $\rightarrow$ [Continuous development] $\rightarrow$ [Result/Reaction]. [Camera Motion Type] with [Amplitude] at [Speed]. [Subject ID] ([Speaker ID]) [Action]: [Language] [Verbatim Dialogue].



overall_soundscape: [1–4 sentences of ambient and physical sounds].



non_diegetic_music: [1–3 sentences of audience-only score].

FL2VA (First-and-Last-frame-to-Video-Audio)
How the reference pictures align with the target video — Picture 1 (from Shot 1) aligns with the 0.00-second mark of the target video; Picture 2 (from Shot [N]) aligns with the [S.SS]-second mark of the target video.



integrated_multimodal_description: [Shot 1] [Style derived from images], [First-frame state from Picture 1]. [Observable intermediate changes] $\rightarrow$ [Progressively narrowing differences] $\rightarrow$ [Last-frame state from Picture 2]. [Camera Motion Type] with [Amplitude] at [Speed].



overall_soundscape: [1–4 sentences of ambient and physical sounds].



non_diegetic_music: [1–3 sentences of audience-only score].

L2VA (Last-frame-to-Video-Audio)
How the reference pictures align with the target video — (from [Shot N]) aligns with the [S.SS]-second mark of the target video.

integrated_multimodal_description: [Shot 1] [Style derived from image], [Plausible preceding state]. [Explicit action and transition path] $\rightarrow$ [Gradual convergence in final shot] $\rightarrow$ [Landing on ]. [Camera Motion Type] with [Amplitude] at [Speed].



overall_soundscape: [1–4 sentences of ambient and physical sounds].



non_diegetic_music: [1–3 sentences of audience-only score].



Ref Model (Full-Reference Mode)

The omni-reference model uses a six-section structure. It replaces the "Instruction" line with explicit reference labels (``, ``, etc.) throughout the analysis and description.

Supports multi-modal reference inputs:

Images: ≤ 9 images

Videos: ≤ 3 clips; each clip must be 2–15 seconds long; total duration ≤ 15 seconds

Audio: ≤ 3 clips; each clip must be 2–15 seconds long; total duration ≤ 15 seconds

Mixed inputs: Maximum number of files across all input types is 12

General Full-Reference Scaffold
This scaffold covers Reference Generation, Keyframe Completion, and Video Editing/Continuation.

subject_definitions:

is [description of person/object/scene] from [/

is [first frame/keyframe/anchor] for [Shot N], showing [compositional detail].

is [voice-timbre/signal reference] for ([Speaker ID]), containing [audio characteristics].



summary:

[[Task Type: keyframe completion / reference generation / video editing / video continuation / audio reuse / audio reference]] [Short paragraph summarizing target video, main reference relationships, and shot flow using defined labels].



retention_analysis:

(appears in [Shot X]): [marker: fully_preserved / partially_preserved / attribute_transfer / weak_reference] - [explanation of what is kept/changed].

([Shot X] anchor): [marker] - [explanation].

: [marker: fully_copy / partially_copy / reference / weak_reference] - [explanation].



detailed_description:

[1-2 sentences establishing overall style, lighting, and color palette].

[Shot 1] [Initial composition]. [First appearance of ] [referenced characteristics, position, action]. [Camera Motion Type] with [Amplitude] at [Speed]. ([Speaker ID]) [Action]: [Language] [Verbatim Dialogue].

[Shot 2] At [00:SS.mmm], the shot cuts to [Composition]. [Reference to or



overall_soundscape:

[1-4 sentences of ambient/physical sounds. Cite if signal is copied/referenced].



non_diegetic_music:

[1-3 sentences of audience-only score. Cite if signal is copied/referenced].

Scenario-Specific Adaptations for Ref Model
1. For Keyframe Completion (I2VA/FL2VA/L2VA logic):

  • Summary: Use `[keyframe completion + reference generation]`.
  • Detailed Description: Instead of the "Instruction" line, use phrasing like "the shot begins from ", "the shot's keyframe corresponds to ", or "the shot ends on ".
2. For Video Editing/Continuation:

  • Summary: Use `[video editing]` or `[video continuation]`. Begin the summary text with: "The target video is an edited version of
  • Detailed Description: Cite `
3. For Audio-Heavy Tasks:

  • Subject Definitions: Ensure every vocal source has a mapping: ` is the voice-timbre reference for `.
  • Detailed Description: Use `` for physically speaking characters and `` for cues within a reused BGM/Soundtrack.
 
Funny thing that I noticed when using male character sheets was that the model completely ignored their dicks. The dick in the sheet was clear and detailed but the video output had the generic meat tube it defaults to without any references. Kinda annoying needing to add another reference image for it when it's already in the character sheet
 
not even close to sd 2.0, but very promising development considering wan came out last yr and took many training to avoid sloppy motion
 
nah, i think model limitation, some ref are 100 fine all day long, then one is not ((curious if that scene or likeness is in the data))
-- blurring faces -- helps
 
Thanks it's good to know that I'm not going crazy about not knowing what I'm doing. Ref videos are a great tool but the bleeding issue kinda messes things up. Like it always transfers over the dick and vagina over even though I have reference images for both.

Any sex scenes with penetration are hard to prompt for. I can't get decent hard fucking motion and facial expressions. One thing I ask working with though is using reference images to set them up in the correct positions. Was finally able to get a good 69 vid with it which is something I never could do with grok
 
This time I tried prompts from 3 categories:

  • TikTok Trends
  • Comic effects
  • Magic effects
Because of the limitation of 15.000 characters per post I can't add all prompts in one post. So this one is only the first part of the comic part.
These are all Image2Video prompt. The starting Images are AI created with Krea2. No loras exept turbo loras where used for the examples.
This post is not about workflows or loras but just about prompt ideas.
I only now noticed that only 6 videos are allowed per post. So most of them are only linked.

Comic effects
Chaos Magic



For the target video, at 0.00 seconds into the target video, (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] Live-action, cinematic, first-person POV. The shot begins from , showing the woman exactly as she appears in the reference image. From the bottom edge of the frame a hand glowing with red Chaos Magic energy rises into view. [Shot 2] At 00:01.500, the hand releases a surge of Scarlet Witch-style red Chaos Magic. Bright crimson energy tendrils and swirling red particles rapidly envelop the woman’s outer clothing. The fabric distorts, breaks apart and dissolves into red magical energy within seconds, leaving the woman in elegant lace lingerie while her body and face remain completely unchanged and identical to . [Shot 3] At 00:03.500, the red energy fades. The woman stands in lingerie in the same position, still looking toward the camera, until the end of the clip at 00:05.000.

overall_soundscape: Soft ambient room tone, the deep humming and crackling of Chaos Magic, followed by the distorted dissolving sound of the clothing, then quiet natural breathing.

non_diegetic_music: N/A

Eldritch Magic



For the target video, at 0.00 seconds into the target video, (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] Live-action, cinematic, first-person POV. The shot begins from , showing the woman exactly as she appears in the reference image, her gaze resting on the camera with a calm, unaware expression, her hands relaxed and completely still at her sides. The camera holds a static shot for a beat. From the bottom edge of the frame a hand forming glowing orange-red Eldritch Magic mandalas rises into view. [Shot 2] At 00:01.500, the hand unleashes classic Doctor Strange Eldritch Magic. Multiple glowing magical circles and whip-like energy tendrils shoot forward and wrap around the woman's outer clothing. The moment the magical energy wraps around the fabric, her eyes snap wide open and her mouth falls open in a gasp of shock, and she flinches with an involuntary startle. The fabric is sliced and pulled away by the magical energy entirely by the magic alone, vanishing into sparks of orange light, while the woman stays completely passive, her hands remaining relaxed at her sides, leaving the woman in elegant lace lingerie while her body and face remain completely unchanged and identical to . [Shot 3] At 00:03.500, the Eldritch Magic fades. The woman stands in lingerie, frozen in place with a wide-eyed, open-mouthed expression of pure shock, staring in disbelief at the camera, her hands still relaxed at her sides, until the end of the clip at 00:05.000.

overall_soundscape: Soft ambient room tone, the mystical whooshing and ringing of Eldritch Magic mandalas, followed by the sharp slicing and dissolving sounds of the clothing being removed, punctuated by the woman's sharp gasp of shock, then quiet, shaky breathing.

non_diegetic_music: N/A

From the deep



For the target video, at 0.00 seconds into the target video, (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] Live-action, cinematic, first-person POV. The shot begins from , showing the woman exactly as she appears in the reference image, her gaze resting on the camera with a calm, unaware expression, her hands relaxed and completely still at her sides. The camera holds a static shot for a beat. From the bottom edge of the frame several wet, slimy octopus tentacles and a few small fish suddenly rise into view, controlled by The Deep. [Shot 2] At 00:01.500, the tentacles lunge forward in a comically absurd manner and wrap around the woman's outer clothing. The moment the slimy tentacles make contact with the fabric, her face twists in disgust, her lip curls, her eyes squint shut, and she flinches back with an involuntary recoil. With exaggerated, over-the-top movements the tentacles and fish pull, yank and strip the clothing off her body entirely by the sea creatures alone in a ridiculous and slightly chaotic fashion, while the woman stays completely passive, her hands remaining relaxed at her sides. Within seconds the outer clothing is completely removed and dragged out of frame by the tentacles and fish, leaving the woman in a two-piece bikini — a bikini top and matching bikini bottom — while her body and face remain completely unchanged and identical to . The tentacles then awkwardly retreat. [Shot 3] At 00:03.500, the last tentacle disappears from view. The woman stands in her two-piece bikini, the top and bottom clearly separate, her face still twisted in disgust with a curled lip and scrunched nose, shaking her head slowly in revulsion, her hands still relaxed at her sides, until the end of the clip at 00:05.000.

overall_soundscape: Soft ambient tone, wet squelching and slapping sounds of tentacles, faint fish flopping, exaggerated cartoonish pulling and fabric-ripping noises, punctuated by the woman's short, disgusted vocalization and her sharp recoil gasp, then quiet, shaky breathing.

non_diegetic_music: N/A


Laser Beams/Eyes



For the target video, at 0.00 seconds into the target video, (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] Live-action, cinematic, first-person POV. The shot begins from , showing the woman exactly as she appears in the reference image, her gaze resting on the camera with a calm, unaware expression, her hands relaxed and completely still at her sides. The camera holds a static shot for a beat. From the bottom edge of the frame a pair of glowing red eyes rises into view, indicating the POV of Homelander. [Shot 2] At 00:01.500, intense red Heat Vision beams shoot from the eyes toward the woman. The moment the beams first trace across the fabric, her eyes snap wide open and her mouth falls open in a gasp of shock, and she flinches with an involuntary startle. The precise laser-like beams continue to trace across her outer clothing entirely by the magic alone, burning and slicing the fabric cleanly without ever touching her skin. Within seconds the outer clothing is completely burned away and falls off as ash and scorched remnants while the woman stays completely passive, her hands remaining relaxed at her sides, leaving the woman in a two-piece lace set — a lace bra and matching lace panties — while her body and face remain completely unchanged and identical to . [Shot 3] At 00:03.500, the Heat Vision beams stop. The woman stands in her two-piece set, the bra and panties clearly separate, frozen in place with a wide-eyed, open-mouthed expression of pure shock, staring in disbelief at the camera, her hands still relaxed at her sides, until the end of the clip at 00:05.000.

overall_soundscape: Soft ambient tone, the high-pitched charging sound of Heat Vision, followed by the sharp sizzling and burning of fabric and the soft crumble of ash falling to the ground, punctuated by the woman's sharp gasp of shock, then quiet, shaky breathing.

non_diegetic_music: N/A


Shapeshifter



integrated_multimodal_description: [Shot 1] Live-action, cinematic, a medium close-up portrait frames a woman facing the camera, smiling softly. The camera holds a static shot for a beat, then a classic Mystique-style transformation effect from the X-Men films begins: a rippling wave of blue, scale-like energy and shifting texture spreads rapidly across her body and clothing. Within the transformation the clothing dissolves and vanishes. [Shot 2] At 00:02.500, the shot cuts to a tighter close-up. The Mystique effect completes. The woman is now fully nude while remaining the exact same person with the same face and expression. She continues looking into the camera with a slight smile until the end of the clip at 00:05.000.

overall_soundscape: Soft ambient room tone, followed by the subtle shimmering and rippling energy sound of the transformation effect, then quiet natural breathing.

non_diegetic_music: N/A


Glove Snap



For the target video, at 0.00 seconds into the target video, (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] Live-action, cinematic, first-person POV. The shot begins from , showing the woman exactly as she appears in the reference image, her gaze resting on the camera with a calm, unaware expression, her hands relaxed and completely still at her sides. The camera holds a static shot for a beat. From the bottom edge of the frame a hand wearing the Infinity Gauntlet is raised into view. The golden gauntlet with the Infinity Stones is clearly visible in the lower foreground. [Shot 2] At 00:01.500, the fingers of the gauntlet close and perform the iconic snap. Instantly the classic Avengers "Blip" effect activates: the woman's outer clothing breaks apart into dark, ashy particles and fine dust that float upward and dissolve into the air, exactly like the people vanishing in Avengers: Infinity War. The moment the ashy particles reach her clothing, her eyes snap wide open and her mouth falls open in a gasp of shock, and she flinches with an involuntary startle. The outer clothing crumbles and disperses entirely by the snap alone, while the woman stays completely passive, her hands remaining relaxed at her sides. Only her elegant lace lingerie remains. Her body and face stay completely unchanged and identical to . [Shot 3] At 00:03.500, the last particles fade away. The woman stands in her lace lingerie, frozen in place with a wide-eyed, open-mouthed expression of pure shock, staring in disbelief at the camera, her hands still relaxed at her sides, while the Infinity Gauntlet remains visible in the lower foreground until the end of the clip at 00:05.000.

overall_soundscape: Soft ambient tone, the metallic sound of the Infinity Gauntlet moving, the sharp iconic snap, followed by the quiet, eerie crumbling and dispersing particle sound from the Blip effect, punctuated by the woman's sharp gasp of shock, then quiet, shaky breathing.

non_diegetic_music: N/A


Spider Webs
https://turbo.cr/embed/Y2H4X0-zdtvyk

For the target video, at 0.00 seconds into the target video, (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] Live-action, cinematic, first-person POV. The shot begins from , showing the woman exactly as she appears in the reference image, her gaze resting on the camera with a calm, unaware expression, her hands relaxed and completely still at her sides. The camera holds a static shot for a beat. From the bottom edge of the frame a hand in a Spider-Man suit raises into view. [Shot 2] At 00:01.500, the hand shoots several thin, white Spider-Man webs toward the woman. The moment the webs grip the fabric, her eyes snap wide open and her mouth falls open in a gasp of shock, and she flinches with an involuntary startle. The webs rapidly wrap around her outer clothing, tighten, and forcefully pull the fabric directly toward the camera, yanking it forward and down beneath the frame, entirely by the webs alone, while the woman stays completely passive, her hands remaining relaxed at her sides. The clothing is torn off her body within seconds, flying toward and out of the bottom edge of the frame. The woman is left in a two-piece lace set — a lace bra and matching lace panties — while her body and face remain completely unchanged and identical to . [Shot 3] At 00:03.500, the webs retract and disappear. The woman stands in her two-piece set, the bra and panties clearly separate, frozen in place with a wide-eyed, open-mouthed expression of pure shock, staring in disbelief at the camera, her hands still relaxed at her sides, until the end of the clip at 00:05.000.

overall_soundscape: Soft ambient tone, the signature Spider-Man web-shooting thwip sounds, followed by the stretching and tearing of fabric as the clothing is yanked toward the camera and flung out of the bottom of the frame, punctuated by the woman's sharp gasp of shock, then quiet, shaky breathing.

non_diegetic_music: N/A

Older posts
Part 1
 
More comic effects and some tries to replicate tiktok trends.

Rapid Shrinking


For the target video, at 0.00 seconds into the target video, (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] Live-action, cinematic. The shot begins exactly from the first frame of : the woman stands in the same pose, lighting, and framing as the reference image, fully dressed as she appears there, her hands relaxed and completely still at her sides. The camera holds a static shot for a beat. Then her body suddenly begins to shrink rapidly and dramatically. The moment the fabric starts to billow loose around her contracting frame, her eyes snap wide open and her mouth falls open in surprise. Every layer of clothing loses all its stretch at once and begins to sag outward into enormous, deflated folds, hanging limp and dragging along the ground around her rapidly shrinking body, while her hands stay relaxed and still at her sides. [Shot 2] At 00:02.000, the shot cuts to a closer view, and the camera pushes in with large amplitude at fast speed, tracking her shrinking figure and keeping her in frame as she continues to shrink at a breathtaking pace, her body contracting to a tiny fraction of its original size. The surplus fabric billows and pools around her in massive, limp folds, straps slipping and seams going slack. The surprised woman gasps, her wide eyes darting around as she watches her own body shrink, her head tilting forward in disbelief. Her face and features remain completely unchanged and identical to , just enormously smaller. Then her two-piece lace set — the lace bra and the matching lace panties — tightens and stretches under the same strain: the lace strains taut, its seams split and burst, and the bra and the panties rip away and flutter down on their own as well, leaving her fully nude. Her face and features remain completely unchanged and identical to , just enormously smaller. [Shot 3] At 00:04.000, the shot cuts to a tight shot from a high angle, looking down at her. The shrinking stops. She stands fully nude, now extremely tiny, her hands still relaxed at her sides, frozen with a wide-eyed, open-mouthed expression of stunned surprise, gazing upward in disbelief, breathing quietly, until the end of the clip at 00:05.000.

overall_soundscape: Soft ambient tone under a fast, descending whoosh as her body rapidly shrinks to an extremely small scale. Fabric flutters, slacks, and rustles as it sags into loose folds and drags across the ground, punctuated by the sharp crack of overstretched seams bursting and fabric tearing itself apart, the woman's sharp gasps of surprise layered through it, then everything settles into quiet, stunned breathing.

non_diegetic_music: N/A

Rapid Growing


For the target video, at 0.00 seconds into the target video, (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] Live-action, cinematic. The shot begins exactly from the first frame of : the woman stands in the same pose, lighting, and framing as the reference image, her outer clothing still intact, her hands relaxed and completely still at her sides. The camera holds a static shot for a beat. Then her body suddenly begins to grow rapidly and dramatically taller. The moment the fabric tightens around her expanding torso, her eyes snap wide open and her mouth falls open in surprise. Her outer clothing is overstretched and starts to tear on its own — hairline cracks run through the material and the first seams begin to pop, while her hands stay relaxed and still at her sides. [Shot 2] At 00:01.000, the camera cuts to a closer shot and pedestals up with large amplitude at fast speed, struggling to keep her in frame as she grows at a breathtaking pace, her body stretching to many times its original height. Her outer clothing stays the same size while her body outgrows it, and it rips apart entirely by itself: seams burst, buttons pop off, and the torn pieces peel away from her body and flutter down to the ground on their own. The surprised woman gasps, her wide eyes darting up and down as she watches her own body stretch, her head tilting back in disbelief as she towers higher and higher. Then her lace lingerie tightens and stretches under the same strain: the lace strains taut, its seams split and burst, and the pieces rip away and flutter down to the ground on their own as well, leaving her fully nude. Her face and figure remain completely unchanged and identical to , just enormously taller. [Shot 3] At 00:03.500, the camera cuts to a wide shot from a low angle, looking up at her. The growing stops. She stands fully nude, now enormous and towering over the entire space around her, her hands still relaxed at her sides, frozen with a wide-eyed, open-mouthed expression of stunned surprise, gazing upward in disbelief, breathing quietly, until the end of the clip at 00:05.000.

overall_soundscape: A soft, deep, resonant rushing sound swells as her body rapidly expands to an enormous scale, punctuated by the sharp crack of overstretched seams bursting and fabric tearing itself apart, buttons popping off, and torn cloth flapping to the ground, the woman's sharp gasps of surprise layered through it, then everything settles into quiet, stunned breathing.

non_diegetic_music: N/A

Telekinesis


For the target video, at 0.00 seconds into the target video, (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] Live-action, cinematic, first-person POV. The shot begins exactly from the first frame of : the woman stands in the same pose, lighting, and framing as the reference image, her hands relaxed and completely still at her sides. The camera holds a static shot for a beat. Then a powerful telekinetic force suddenly grips the woman and lifts her off the ground. The moment the force takes hold, her eyes snap wide open and her mouth falls open in surprise, while her hands stay relaxed and still at her sides. [Shot 2] At 00:01.000, the shot cuts to a wider view, and the camera pedestals up, keeping her in frame as she is raised higher into the air. The surprised woman begins a smooth 360-degree rotation in mid-air, her wide eyes darting around in disbelief as she spins. While she rotates, invisible telekinetic energy tears her outer clothing apart entirely by the force alone: the fabric rips and is forcefully pulled away to the left and right sides of the frame, flying out of view, while the woman stays completely passive, her hands remaining relaxed at her sides. By the time she completes the full rotation, she is left in a two-piece lace set — a lace bra and matching lace panties — while her body and face remain completely unchanged and identical to . [Shot 3] At 00:03.500, the shot cuts to a tighter framing. The telekinetic force gently lowers her back to the ground. She stands in her two-piece set, the bra and panties clearly separate, her hands still relaxed at her sides, frozen with a wide-eyed, open-mouthed expression of stunned surprise, staring at the camera in disbelief, until the end of the clip at 00:05.000.

overall_soundscape: Soft ambient tone, a low humming of telekinetic energy, a gentle whoosh as she is lifted and rotated in mid-air, punctuated by the sharp crack of fabric tearing and being forcefully pulled away and the woman's sharp gasp of surprise, then quiet, stunned breathing as she is set down.

non_diegetic_music: N/A

TikTok Trends

#outfittransition


For the target video, at 0.00 seconds into the target video, (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] Live-action, cinematic, vertical social-media try-on style, a medium-full shot frames the same bright room and the same young woman shown in , preserving her face, hair, body type, and the top she wears in the image. The camera holds a static shot as she reaches down out of the bottom of the frame and lifts a black lace bra into view, holding it with both hands at a distance in front of her torso. She then makes several quick, rhythmic movements: pressing the lace bra to her chest with both hands and pulling it back out to a distance in front of her, pressing it to her chest a second time and pulling it forward again, and on the third press to her chest the shot cuts. [Shot 2] At 00:04.000, the camera cuts to the same framing, and in that instant her outfit has changed — the top she was wearing is gone, the bra is no longer in her hands, and she is now wearing the matching black lace bra. She lowers her empty hands, flashes a quick satisfied smile at the camera, then reaches down out of the bottom of the frame again and lifts the matching black lace panties into view, holding them with both hands in front of her waist. She repeats the same quick, rhythmic effect with the panties: pressing them to her lower belly and hips with both hands and pulling them back to a distance, pressing them a second time and pulling them forward again, and on the third press against her body the shot cuts. [Shot 3] At 00:08.500, the camera cuts to the same framing, and in that instant the panties are suddenly on — the matching black lace set is now complete on her body and the panties are no longer in her hands. She lowers her empty hands to her sides, settles into a confident pose with a small smile, and the camera eases into a slow push-in with small amplitude at slow speed as she looks at the camera; the push-in settles as the clip ends.

overall_soundscape: Quiet indoor room tone runs under the whole clip. Soft lace and fabric rustle as the bra is lifted and pressed to the chest, a quick bright whoosh and a short impact tick land exactly at the first flash change, then rustling again as the panties are lifted and pressed to her lower body, a second whoosh and tick at the second flash change, and a light satisfied breath follows as she settles into the final look.

non_diegetic_music: An upbeat pop-electronic beat at a fast tempo with a four-on-the-floor kick, a plucky synth bassline, and bright chord stabs, punctuated by a sharp hit and a brief sparkle riser exactly at the bra change and repeated at the panties change near the end.

#wipeitdown

This one is reference2video. Picture 1 is the person, audio 1 is the song.


subject_definitions:

is the young woman whose appearance comes from ; she wears an oversized black hoodie, black shorts, and white ankle socks at the start, later changing into a black lace bra and matching black lace panties.

is the modern bedroom environment featuring a large full-length white-framed mirror, a bed with white duvet and pillows, a grey sofa in the background, a dark wooden door with colorful circular decorations, a white ceiling with a multi-bulb light fixture, light wooden floor, and a large green leafy plant in the left foreground partially framing the shot.

is the complete original audio track containing the upbeat female vocal hook "Wipe, wipe, wipe it down, wipe" and accompanying beat, to be reused 1:1.

summary:

[reference generation + audio reuse] The target video recreates the #wipeitdown TikTok trend in which stands in front of the mirror in , sprays cleaner, wipes the glass, and transforms from casual loungewear into lingerie while continuously wiping the mirror to the fully reused .

retention_analysis:

(appears throughout all shots): fully_preserved - appearance from is retained exactly; only clothing changes as part of the trend action.

(appears throughout all shots): fully_preserved - the bedroom layout, mirror, bed, plant, lighting and decorative elements remain consistent.

: fully_copy - is reused 1:1 as the target video’s complete final audio track with no additions or alterations.

detailed_description:

The target video is a vertical smartphone-style TikTok clip with natural indoor daylight mixed with soft ceiling lighting, slightly warm color grade and shallow depth of field that keeps the green plant softly blurred in the foreground.

[Shot 1] A static medium-wide shot framed through the large white-framed full-length mirror in . , the young woman from , stands barefoot in white ankle socks on the light wooden floor, wearing an oversized black hoodie and tight black shorts. She holds a crumpled white cleaning cloth in her right hand and looks toward her reflection with a playful expression. A hand in a black sleeve (her own arm from outside the mirror frame) enters from the right holding a blue spray bottle and mists the mirror surface several times. then raises the white cloth and begins wiping the glass in continuous circular and up-and-down motions, her body swaying lightly to the rhythm. The swiping motion gets larger and slower each time. At 00:08.200 the wiping motion continues without interruption and her oversized black hoodie and black shorts visually dissolve and wipe away while her arm hides the refelction in the mirror, revealing a black lace bra with small decorative details and matching black lace panties. She keeps wiping the mirror vigorously with the white cloth in her right hand for the entire remaining duration, never stopping the cleaning motion, while looking at her reflection with a playful expression. The shot remains continuous with no cuts until the audio ends cleanly; the final frame shows still actively wiping the mirror while standing in lingerie in front of it with the plant in the left foreground.

overall_soundscape:

Quiet indoor bedroom ambience with faint soft room tone; no additional diegetic sounds beyond those already present in the reused track.

non_diegetic_music:

N/A — the complete musical and vocal content is supplied by the fully copied .

This prompt was very hard to enginner (It's Text2Video):


a naked woman dancing

Older Posts:
Part 1
Part 2
 
really good with tiktok trend ones lol. Underwear sticks like glue haha
 
That said, just tried this model https://huggingface.co/TenStrip/10Eros-Max/tree/main 21gb one, disabled the lora node since...apparently this model has loras baked in; and so far it handles NSFW stuff pretty good with good quality at 6 steps, before I had to do 8. If you're gonna try it you need to bypass/delete H3 AdaLN LoRA Fix and maybe ModelSamplingMiniMaxH3 nodes in Conditioning - Sampler subgraph of the above workflow. Also bypass the lora node, either bypass it or remove the lora/s or at least the turbo lora which is definitely baked in.
 
I've tried different generations but the dick always comes out like some radioactive blob, it gets better with sageattention disabled, have you noticed this issue as well and if yes what worked?
 
You need to be super detailed with your prompting. It's incredibly tedious to write an essay per prompt. I found a decent system prompt for minimax h3 ref2va flows and run it against gemma 4b uncensored. you can attach your image inputs and it does a great job of keeping characters consistent and realizing your vision.
 
I gave my old grok creampie method a try on some vids and they surprisingly work the same and well. Using "while the dick is inside of her mouth thick clear white gel begins to leak from her mouth/vagina/anus". The result is something that looks like convincing cum just flowing out realistically unlike the paper glue white stuff the model generates when describing cum.

If the male is sitting down while getting a bj you can add "and runs down the shaft of the penis and the scrotum" for a better effect. You can even have it pooling on the seat of whatever they're sitting on.
 
Any help with clothes remove edit?? I found this workflow https://civitai.red/models/2855941/minimax-h3-character-replacement?modelVersionId=3238780 for characters replacement.
They use SAM to mask the character and I edit the SAM node to let it mask the clothes so it can be edit out. But it still doesn't looks 100% identical to the original vid...

It did indeed edit the clothes out and keeps element of the vid almost the same, but the face and movement always got changed in some part of the video, either the expression or arms movement. It always got changed a little. (Looking in the preview node, it seems like while generating there's always some part that the model add unnecessary noise and made the edit looks different from source vid T_T)
Is there no way to edit only the clothes and make the unmaked/unsegmented part stay exactly the same as original vid???

There use to be a workflow that someone released on here that was made for this job but that was for Wan 2.1 Vace. Is it possible for this model tho?? Iirc it was someone name Singh... Something
 
can you share it with those of us that don't have (and don't want) a civitai account please. there's no huggingface mirror on civarchive
 
You must log in or register to reply here.
Back
Top