Workflow Venice AI Workflows and Models (DON'T POST REAL PEOPLE!!)

do you have the exact wording on the prompt? I've tried this but got mixed results like the original girl but the outfit of the target woman, etc.
 
it's hit and miss. depends on the scene too but i've used something like this with vary degrees of success. Obviously tweak it to fit your material.

---

5 seconds, 16:9 aspect ratio. The woman from @Image1 performs the professional scene from video1.

Subject: The woman defined by @Image1 with her specific face - smiling expression, dark wavy shoulder-length hair, nose stud piercing, hoop earrings - replicates the exact movements, poses, and actions from video1. She maintains the body, lingerie, and positioning from video1 while her face and hair match @Image1.

Action: Recreate every motion from video1 precisely - same timing, body language, poses, gestures, and professional adult scene interactions. The woman with face/hair from @Image1 becomes the performer executing the scene.

Environment: The exact setting from video1 with identical background, props, lighting, and atmosphere.

Lighting: Matches video1 lighting conditions - same sources, shadows, and color temperature.

Camera: Follows video1 camera work exactly - same angles, framing, movement, and professional adult scene cinematography.

Style/Mood: Seamless face and hair replacement maintaining the original video's professional adult tone, pacing, and nature. Face and hair from @Image1, body and lingerie from video1.

Audio: Matches video1 ambient sounds and professional scene audio.
 
I have had no luck with 3.0 and think it sucks. Wan 2.7 enhanced is the best model by far
 
If that wallet is deep I suggest getting an ai workstation instead and generating locally. It's even more addicting
 
I've been using this prompt and Wan 3.0 to replace someone in a video with a totally fictional character from reference images, works best with continuous single shot scenes (720p input video in either 9:16 or 16:9):

Replace the woman in video1 with the woman from the references.

Face: @Image1 @Image2
Body: @Image3 @Image4

Use the reference woman's exact face, body, breasts, hips, and proportions. Do not keep the source woman's body or measurements.

Keep video1 camera, motion, environment, timing, and audio. Remove all clothing and UI. Same scene, new woman. Extend video1 backward, generate
 
wow, Wan 3.0 reference can do extreme fetish. using a porn vid reference, a nude reference of my chosen woman, and then a face reference of my chosen woman, I was able to watch a woman I know anal fuck a dildo the size of a leg and then spread her ass for a huge prolapse. No tricks, no bullshit full page prompting, just this:

replace woman in video1 with the woman in @Image1 but retain all actions of video1. use @Image2 for face reference.
 
The 50% discount on Wan 3.0 isn't showing for me anymore. Wondering if that's the same for everyone?

A few nights ago it disappeared for 12 hours then came back so I'll wait till tomorrow but I suppose it had to end at some point.
 
I tried something similar to this it just regenerated the same reference video with the same woman from the reference video basically eating the credits for nothing
 
Its working great. Thanks mate.
I'm wondering, if I add the girl's voice as a reference, can it clone it using a prompt to say exactly what's in the reference video? I mean not writing down what she say just adding to the prompt.
 
Hello!
attached video as a reference,+ photo of model to change, and then add this prompt with Wan 3.0 ofc. But for some reason it's showing me an error…already tried different type od video but same error. Do you know what the problem might be?
thanks!
 
I just noticed something important likely the reason I couldn't get anything to work with WAN 3.0. The videos have to be in 480p; if I do them in 720p, the output is a mess, but at 480p, it follows my instructions. Do you guys experience this too? (I pick a face, a naked body, a clothed girl, and a setting; then she undresses and I fuck her.)
 
I mean I'm able to generate NSFW stuff without a plan, just buy credits and make shit happen lol
 
Prime is supposed to be faster,but regular Wan 3.0 doesn't take that long. Prime is also more expensive. If you're using references, maybe you have to wait 12-15 mins depending on the length of your video and the prompt.

I have had Wan 3.0 gens take 3-5 mins, sometimes less. I just click another tab or do something else, I don't mind waiting to save some tokens.
As far as the output, I don't see too much of a difference. I started out using prime, but now I use 480p Wan 3.0 and the vids are just as good.
 
Given the thread we are in I assume most of us are using Venice AI, fully uncensored on there
 
This is the Venice AI thread, so... we're talking about Venice AI. However, as I mentioned earlier, in my case I have to create 480p videos so it follows my instructions best. At 720p, it just does random stuff and the result is completely rubbish.
 
Does anyone know how to fix the video reference on wan 3.0 no matter what I try to add it says it’s not compatible even other videos I made on Venice. Too many pixels or too little too long or too short
 
For references on Wan 3.0, videos have to be 1-15 seconds. If they're too long trim them to 14.50 or .60 seconds for it to work.
Also, vids can't be low resolution and pics can't be under a certain size. The solution for this is to find a cheap way to upscale them.

For example, I couldn't use my 480p grok references I made in Seedance 2.0 R2V because it says they're too low resolution. They had to be 720p.

I wouldn't use Topaz video Upscaler on Venice to upscale videos, 120 tokens to upscale is a rip off. just find another free tool or a program on the net.

-------------------------------------------------------------------------------------------------------------------------------------------------------------

Another tip that has been making my gens better is I type about 5- 6 sentences of what I want:
1-2 sentences for the action, 1 for woman or man's emotions,1 sentence for the audio, and 1 for what I want the camera to do.

Then I use the prompt enhancer (the little wand icon in studio) on Venice to do the heavy lifting. Don't know why I wasn't doing this before, but I'm getting more of what, especially camera wise. Don't do this with references because it will remove your @ and doesn't tag references.
 
You can you use the movie editor in Venice to simply import the 480p footage and export it from there at 1080p. Just make sure to trim 15 videos by half a second. If your video reference is 15 seconds (14.5) you can generate a 15 second video. If your video reference is 10 seconds you can do a 20 second video. And so on.
 
You must log in or register to reply here.
Back
Top