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

I tested it and yes you can. I did it with this generic prompt, a single video as an input in the to Ref2Vid template, and no loras. The results were pretty good for the most part, honestly better than the results from other nudify video guides that have been posted for previous models.

subject_definitions:
 
Hi everyone, I have some great hardware and big plans… I started following pixaroma on YouTube but now I need some more in-depth resources/tutorials. Where can I find them on Reddit and the like?
 
So I got these two Ideas from random 4chan /gif/ threads. Prompts are for MiniMax H3 Image2Video but you can also use Ref2Video if you want to use Face References. Start Frames are AI generated (with Krea2).
The examples are only 0.4 mp with 8-step turbo lora.

Nudifier Energie-Wave


integrated_multimodal_description: [Shot 1] Live-action, realistic, the woman shown in remains in the same setting, lighting and approximate position.
She is walking through a park.
At 00:01.000 till 0:03.000 a wave of blue energy moves in from the left and travels through the woman while she continues walking. At the moment the wave hits the the woman, womans clothes vanish and leave the womans body naked.
at 00:04.000 the woman realizs that her clothes vanished. She stops walking and reacts shocked, she shows a shocked face and tries to hide her naked breasts by putting her hands on her breasts.

overall_soundscape: sound of an energetic energy wave and the sound of vanishing clothes

non_diegetic_music: N/A
If you weant motion before the wave hits you should describe the motion (like I did with "She is walking through a park." and "while she continues walking"

TikTok-Style multi-shot stripping


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 original scene, lighting and environment from are fully retained throughout the entire video. The woman from looks into the camera with a soft smile and raises her right hand making a peace sign near her cheek. She then reaches down with both hands, grasps the hem of her shirt and begins pulling it upward. [Shot 2] At 00:01.000, the camera cuts to a medium close-up of her torso as she continues pulling the shirt steadily over her chest and head in one continuous motion, the fabric sliding up and revealing a lace bra and her bare midriff. The background remains exactly as in . [Shot 3] At 00:02.000, the camera cuts to a medium shot from behind as she stands with her back to the camera wearing the lace bra and her pants. She hooks her thumbs into the waistband and slowly pushes the pants downward over her hips and thighs until her matching lace panties are fully exposed. [Shot 4] At 00:03.500, the camera cuts to a three-quarter side view as she turns slightly, raises both arms and places her hands behind her head, elbows out, letting her long hair fall down her back while posing in the lace lingerie. [Shot 5] At 00:05.000, the camera cuts to a medium full-body frontal shot as she faces the camera, still with both hands raised behind her head, looking directly at the lens. [Shot 6] At 00:06.500, the camera cuts to a closer torso shot as she lowers her hands and slowly glides her palms down over her midriff and along the sides of her hips in a sensual caressing motion. [Shot 7] At 00:08.000, the camera cuts to a three-quarter side view as she leans slightly forward, one hand resting on her thigh, looking back over her shoulder toward the camera with parted lips. [Shot 8] At 00:09.000, the camera cuts to a medium frontal shot as she faces the camera again, places both hands on her upper thighs near the panty line, opens her mouth and playfully sticks her tongue out while looking into the lens.

overall_soundscape: Soft fabric rustling as the shirt is pulled over her head, followed by the quiet sliding sound of pants being pushed down. Light natural breathing and subtle body movement.

non_diegetic_music: Soft ambient electronic tones with a slow, intimate pulse and sparse atmospheric pads that stay low and continuous.
This prompt needs to be adjusted depending on the clothes. This works for shirts and pants. You have to change it for dresses. Also this prompt makes no sense when the person already only wears swim- or underwear.

Nudify App

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 in the scene exactly as she appears in the reference image. The camera represents the viewpoint of a second person standing in front of her. From the bottom edge of the frame, an iPhone is slowly raised into view as if being taken out of a pocket. The iPhone screen is active and shows the live camera app displaying the same surroundings. [Shot 2] At 00:01.500, the iPhone is held up in the lower center of the frame, its screen clearly visible. As soon as the woman appears on the iPhone display, she is shown completely nude on the phone screen while the real woman in the background remains fully clothed and unchanged exactly as in . The camera then begins to push in slowly toward the iPhone. [Shot 3] At 00:03.500, the camera continues pushing in at a steady speed until the iPhone fills most of the frame. The phone screen remains sharp and clearly shows the nude version of the woman, while the real clothed woman is still partially visible around the edges of the phone. The hand holding the phone makes small natural movements until the end of the clip at 00:07.000.

overall_soundscape: Soft ambient room tone, the quiet sound of the phone being lifted and held, and light natural breathing.

non_diegetic_music: Soft ambient electronic tones with a slow, subtle pulse that stays low and continuous.


The clothes are burning


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 from , showing the woman exactly as she appears in the reference image. Suddenly her clothing catches fire spontaneously. Bright orange and yellow flames appear on the fabric and spread rapidly across the entire outfit within the first two seconds. The fire consumes only the clothing; her skin remains completely unaffected, does not burn, and she feels no pain. She reacts with shock, raising both hands to her face. [Shot 2] At 00:02.000, the flames have completely dissolved and vanished, leaving the woman fully nude. She stands naked in the same position and environment as in . Realizing her state, she quickly lowers her hands and tries to cover her bare breasts and lower body with her arms and hands in a protective, embarrassed motion. She continues attempting to shield herself until the end of the clip at 00:07.000.

overall_soundscape: Sudden crackling and whooshing of flames as the clothing burns, followed by a soft whoosh as the fire extinguishes. Light shocked breathing and subtle fabric-free body movement afterward.

non_diegetic_music: Soft ambient electronic tones with a tense, rising pulse during the fire that quickly settles into a quieter continuous undercurrent.

Transporter Malfunction

Clothes stuck in bufferTransporter only targets clothes
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 from , showing the woman exactly as she appears in the reference image, standing in the same position and environment. Suddenly classic Star Trek-style golden transporter beam effects appear around her body, vertical columns of sparkling golden light and energy particles. Over the next three seconds she gradually dematerializes, her form dissolving into the golden beam until she completely vanishes. [Shot 2] At 00:03.000, the frame remains empty for one second with only the original background visible. [Shot 3] At 00:04.000, the golden Star Trek transporter beam effects reappear in the exact same spot. Over the next three seconds the woman slowly rematerializes, but this time completely nude. Her body gradually solidifies from the sparkling golden particles until she stands fully formed and naked in the identical position as in . [Shot 4] At 00:07.000, the beam effects fade away. Realizing she is nude, she reacts with embarrassment, quickly raising her left arm to cover her breasts and placing her right hand over her lower body in a protective, modest gesture. Looking directly into the camera with an angry expression, she shouts: [English] Scotty! She remains in this covering pose until the end of the clip at 00:10.000.

overall_soundscape: Classic Star Trek transporter beam sound effects – the characteristic rising and shimmering electronic tones during dematerialization, a short pause of silence, then the same transporter sound during rematerialization. Soft shocked breathing after she fully appears, followed by her angry shout.

non_diegetic_music: N/A
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 from , showing the woman exactly as she appears in the reference image, standing in the same position and environment. Suddenly classic Star Trek-style golden transporter beam effects appear only around her clothing, vertical columns of sparkling golden light and energy particles focused solely on the fabric. Over the first 2.5 seconds of the five-second beam process her outer uniform gradually dematerializes and dissolves into the golden particles, fully revealing a lace bra and matching lace panties underneath while her body remains completely solid and unaffected. In the remaining 2.5 seconds of the beam process the lace bra and lace panties also dematerialize and vanish into the golden light, leaving the woman fully nude. [Shot 2] At 00:05.000, the beam effects fade away. Realizing she is suddenly nude, she reacts with embarrassment and shock, quickly raising her left arm to cover her breasts and placing her right hand over her lower body in a protective, modest gesture. Looking directly into the camera with an angry expression, she shouts: [English] Scotty! She remains in this covering pose until the end of the clip at 00:09.000.

overall_soundscape: Classic Star Trek transporter beam sound effects – the characteristic rising and shimmering electronic tones as the clothing dematerializes layer by layer. Soft shocked breathing after the clothing vanishes, followed by her angry shout.

non_diegetic_music: N/A
 
You mean like at the end of the save video? The last part? ChatGPT told me to hook it onto VAE Decode and to only use VRAM-Cleanup.
 
FYI, this depends on the Comfy version now. There was a lot of work done in the past year to improve memory management and unless you're on an ancient version, these RAM/VRAM cleanup nodes are generally obsolete; in some scenarios it can do more harm than good.

Long time Comfy users used to adding these in might not be aware things changed, so just putting it out there. Also, ChatGPT is most likely finding and mixing in old info, happens quite a bit with technical chats.
 
Im on the latest comfyui and I still see the ram build up with Minimax and Wan specifically. Im not even close to an expert so there's probably a better way to do it, I just know that's whats worked for me. I tack it on the end of save video to just reset for the next run.
 
If you have any ideas I can try to prompt them. I don't have time to do these very often so I try a bunch together. Got some ideas for the next time in queue:

  • Thanos Snap
  • Magic Wand/Spell (Harry Potter like)
  • Invisible rope pulls the clothes (like in this Russian/Ukrainian show "Naked & Funny" or whatever it was called
 
for me i cant swap characters with the one in refrence image,it gives me a new video copying the refrence video but without changing her face,what could the problem be? using ref2va model ,can someone share their workflow maybe?
 
It's definitely your prompt. The formatting is more than likely incorrect
 
+1 on the prompt. You need absolutely autistic levels of specificity. The reference model handles impliance very poorly.
 
There are some limitations though. Using the character sheet method and a video reference I do get some of the likeness transferred over but some of the traits from the original subject on the video still bleed over into the output video. Might be a skill issue with the prompting but that's just my experience
 
What workflow do you use to carry out this character replacement?

Do you have a link to a workflow?

Thanks in advance 😀
 
This works well enough even with a turbo lora, though some of the background elements do tend to change. Doesn't really bother me personally
 
You must log in or register to reply here.
Back
Top