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

I’ve honestly never been able to get any reference gens to work. I’m assuming I’m just not prompting right. I haven’t seen how to prompt them explained in depth.
 
I literally just write on Wan 3.0 Reference

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. the woman from @Image1 is wearing something similar like in video1.

For 5/10/15 second videos and the results are genuinely insane. It even works somewhat on SFM videos. Out of 76 total generations around 5 failed.
 
What’s happened with H3 R2V
Look like i2v
Video doesn’t work like before
 
Face consistency is definitely hit or miss but if it works, it's great. The devs should publish a prompting guide, like the devs of Minimax H3 did.
 
Anyone got any tip of reference videos for 3.0 wan? It seems like no matter what I screen record or save it’s too much for the reference even when I cut the videos down? Where do yall get your reference videos from? How do you prep them how long are they?
 
So sorry for the newb question but how do you generate a new image using an image model (like Grok or Seedream) while using reference images? I've used plenty of platforms before (Higgsfield, Freepik--now magnific) and what I'm about to describe has never happened there.

I'm in the chat, and if I upload an image first, it shows up in the chat box, but as soon as I switch to the model I want, the uploaded image disappears.
If I go to the model I want first, then upload the image, it just switches to a completely different model.

Any help would be greatly appreciated.
 
I'm actually looking to create a whole new image using multiple references, not edit one image specifically. (Like put this person and this person in this room or whatever).
It's so intuitive to do that with nearly every other image generation site except this one for some reason.
 
You click the plus icon in the bottom left corner of the chat (it's called Actions when you hover over it) and then click Remix Images. It allows you to upload multiple reference images. Then it's just a case of prompting. I usually just refer to them in the order they appear in the window from left to right.
 
I've been getting great success using Capcut to trim the resulting videos, and using the slimmed down (1s) video to get Wan 3 to "continue" the video. But the quality degrades if you join them too much.

It depends on what you want to do with the reference video, sometimes there's over-fitting problems as well. You have to be extra clear with your prompts.
 
I wrote the Dzine guide thread last year that was pretty active before SimpCity's site update axed it. Venice is worth it just alone for Wan 3.0. I'm getting full 30 second penetrative sex clips. It understands proper realistic movements, does sound, can do scene cuts, and has amazing facial/body consistency with proper references. Pricing is better on Venice than Dzine for Wan 3.0. You can get the same results on Dzine, and the interface is better in my opinion, but pricing is insane. I'll try to put out a guide with pics.
 
Hi, I do remember you. Personally, I’m using a similar approach to what I did with Dzine, except that Wan 3.0 handles vaginal penetration perfectly. With Dzine, you needed a starting image where penetration had already occurred. So, with Wan 3.0, you can have someone undress and then penetrate them. However, I have to do it in 480p; in 720p, it acts up. Is it the same for you?
 
Looks like Wan 3.0 is at capacity at the moment... is this normal for new and popular models? Does anyone know if these resolve automatically after some time?
 
One thing to check is the picture/video you're trying to upload. It will give you this error if the AI recognizes notable celebrities.
 
Tried character replacement in wan 3.0 with video and image reference and the woman in scene looked the same and didn't replace her with the reference image. Anyone get it to work recently?
 
It worked for me only when I cropped the original woman’s head out of the reference video
 
Prompts that have had success for me, credits to stefankirov1210 for giving a great basis.

Recreate every motion from video1 precisely - same timing, body language, poses, gestures, and professional adult scene interactions. The woman's face, body, hair from @Image1 becomes the performer executing the scene, completely replace the woman in the video.

Recreate the scene in video1 but with the woman in @Image1 instead.

Not every sort of pose will work and big tits don't always scale down to smaller cup sizes. Blowjob scenes retain some influence of the original performers face as well. I don't think I'll dip much further into replacement since I'm enjoying 30 second image referenced clips far more, but you can get some neat stuff with video replacement. I can't wait to see where the tech goes in another year, its moved so fast over the past year.

It's gonna suck when the 30% off deal goes away, read somewhere it ends tomorrow? $2 a vid is gonna hurt. Also, Dzine has Hailuo MiniMax H3 15 second vids at 768p for free for the entire week. It's ass at reference sex scenes (don't try it), but may dabble with it a bit for other things. Overall the model is eh compared to Wan 3.0.
 
Ideally, no one should buy them once the price goes back to normal :D That might force them to keep the 30% discount :p It's better to have a lower margin and sell a lot than a higher margin and sell nothing at all.
 
Yeah it sucks, was almost $2 for a 30 second clip at 480p. They put it on sale again but only at 20% off, so like $1.50 a video now. Though, the reduced load has lowered wait times to 5 minutes max per video.
 
You must log in or register to reply here.
Back
Top