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

Some observations on Minimax Enhanced:
It straight up does not understand anal. It can make a butthole, but it doesn't think anything should ever go in there.
Good news though is that motion is the best on Minimax. Characters are more lively than in Seedance. Also, Minimax does pole dancing vids better than even when Grok 1.5 was completely uncensored.
 
Did Venice just increase the price on token top ups? Was it always 25 bucks for 2,500? I cancelled my sub for awhile but uncensored Seedream V5 pro is pretty cheap and the free Qwen Edit uncensored does a decent job at turning 9:16 images into 16:9.
 
It’s a shame the price doesn’t drop more when you buy one, or that there isn’t a preferential rate for subscribers.
 
Venice doesn't make something uncensored it either is or it isn't from the development team especially the non-open source models
 
I know that obviously. They are usually the only major platform to provide the model uncensored, that's why I said in that way.
 
SD 2.5 is live on Venice.
Good news, it is not that expensive compared to SD 2.0. 480p/15s video costs around the same.
480p/30s video costs less than 3 dollars.
 
anyone else getting content moderation errors? It seems like if starting image is nsfw everything's ok, but otherwise I'm getting moderated, is that intended?
 
I have had no issues with generating NSFW I2V so far other than it just not looking that great. No moderation issues
 
if you attach another image and that one is nsfw then no moderation comes up, just edit the other one
 
What's the issue with how it looks? Is it not a step up from v2 for NSFW content?
 
From what I have done with this version and the prior one it does not quite understand NSFW as well as WAN 2.7 (and Grok of course) do. Visually it looks fine enough, but it does not know what to do with prompts as well.
It struggled a lot with attempting to make it generate some blowjob videos, it did not understand what that meant
 
How exactly are you prompting in line? I'm looking at the official guide and can't find an example anywhere. I'm having bleedthrough issues or sometimes the model just spits the reference video back out.
 
Not sure what you mean in terms of "in line" but I just edit the prompts in notepad and paste them into the textbox in venice's video studio.
 
I've been running some tests with SD 2.5 and heres what I've found:
1. Visually its better than 2.0, it has better physics, spacial sense, motion, ect. Videos just look more realistic.
2. The audio isn't too great, and maybe its because I whipped up my testing prompts wrong or hastily. But from the handful of gens I have done so far the audio sounded nothing like the person, where as with 2.0 it was damn near perfect every time.
3. Its more expensive and takes longer to output a video. a 480p 15 second video on 2.0 costs about $1.50 on venice, while the same video costs around $2.30, with 30 second videos being around $3.50. Generations usually take around 8-12 minutes as opposed to 4-5.
4. You shouldn't use prompts formatted for 2.0, you should instead have grok or an ai update them specifically for 2.5 and link 2.5 documentation (i would love to link it but i can't)
Unfortunately I cant share any of my tests here. But I ran a test where I took a 2.0 prompt that gave a damn good output with 2.0, and gave 2.5 the exact same prompt, assets, parameters, ect.
2.5 with the 2.0 prompt completely botched the audio, taking tone and voice notations literally and having the woman say them aloud, however the visual aspect looked more realistic than 2.0.
Then I gave grok a link to the 2.5 prompting documentation and told it to optimize the 2.0 prompt for 2.5, and the output was a bit better, however the audio was a bit closer to 2.5 but still slightly lacking, but overall looking even better than the previous attempt. So do with that as you will.
 
Based on the reviews I've seen so far, the Grok uncensored incident still prevails over Seedance 2.5. Such a shame because I really wanted to mess with the 30-second-length stuff, but there's no point in spending money on something that can't produce results like the way Grok did.
 
I just tested a video with Seedance 2.0 R2V and then with 2.5 R2V (using the same prompt); I can confirm the audio is terrible, and the face rendering is all over the place. Personally, I'm sticking with 2.0—it's better.
 
You must log in or register to reply here.
Back
Top