Heads up to anyone using the new #WanX 2.7 to generate #AIArt or #AIVideo, the order of the reference images you provide matters. If you provide two humans, the first image you provide will ALWAYS be on the left and the second will ALWAYS be on the right. That's good, but they never explain that.
I am three gazes at this image away from actually making the lockers in the locker room of my lifeguard show this narrow just for laughs. #WanX #AIArt
#WanX has a "Project" mode (the one with the timeline) and a "Generate" mode. Based on my experiments, there appears to be a flaw where attempting to run a generation while in "Project" mode locks up your one and only gen slot. Avoid the hassle by taking 5-secs to switch to "Generate" and back.
I have to rethink one of my primary characters because the #AIVideo model #WanX has no idea what "tapping her foot" means. This is yet another roadblock in my march to greatness.
I have to admit, #WanX 2.6 is fucking nailing the reference-to-image #AIArt stills I need to get my shots ready for my #AIVideo series #MarqueeCove. The censorship on their public site is horrifying, but I have other 2.6 providers for that.
You know how sometimes in crime dramas they have a "reenactment" of the crime using holograms or whatever? #WanX has apparently never heard of such a thing. That is cry-laugh #AIVideo right there. The audio makes it funnier. Note the high heels. Some day they'll fix this shit and I'll be so sad.
It would appear that the bonus credits we have been receiving during the #WanX 2.6 #AIVideo testing period has concluded. We're back to 10 credits per day.
I like 15-seconds a lot... except it's not for everything. Sometimes you only need 12 seconds. But we'll fix it in post. #WanX #AIVideo
I just logged back into #WanX and they gave me another 165 credits for today to help test their new 2.6 model. So.... ca-fucking-ching #AIVideo peeps => create.wan.video
I will give #WanX credit for one very important thing that for some reason most of the services in the #GenerativeAI industry have completely forgotten: when you put out a new #AIArt or #AIVideo model, hand out a handful of credits so we can come have a look.
(2/2) Here, I'll give you a hint. Over and over again it's triplets. Every single #WanX 2.6 #AIArt output I've gotten so far replicates the same humans for a crowd. This is bad. Wan has long been at the cutting edge of AI video. They are clearly not investing as much in image gen. Save your credits.
(1/2) Notice anything unusual about these #WanX 2.6 text-to-image outputs?
#PromptShare "Female lifeguards wearing red-one-piece-swimsuits on a Northern-California-cove beach running in a high-action shot toward the beach at noon on a sunny day."
Image on the left is the #AIArt reference I just gave to #WanX 2.6 along with two of my characters (max 3 references). Output on right. Extra bank of lockers isn't so awful, but door fail, ceiling fail, floor fail, two "Denise" characters, three "Monica" characters, and still no D-cups.
#WanX 2.6 includes upgrades to their #AIArt generator--which was previously crippled by their intense desire for censorship. That failure continues. This was the best file I got out of their reference-to-image. That is not "Monica" on the left and all of the lockers are wrong. Jesus says no D-cups.
Heads up to the #AIVideo crowd: #WanX 2.6 is now live. Upgrades to reference-to-video, "multi-shot" from a single prompt, and apparently their obsession with fish-eye lenses has finally paid off.
#MarqueeCove Bro. Fuck yeah. #Lifeguard #Rescue #AIVideo #WanX
If you don't say anything--and I mean ANYTHING--about "woman", "female", "bathing suit", "happiness", or "Eve", #WanX 2.1 will give you the #AIVideo you need.
The #AIVideo crowd may be interested to know that if your original video reference has sunglasses, the #WanX video-to-video ("Avatar" => "Photo Animate") model will close the eyes on the output--which is objectively hysterical.
Not that I recommend the #WanX website for any #AIVideo anymore given their insane UI design and censorship (it was a miracle I got them to take a woman in a swimsuit), however I wanted you to see what their new "avatar" system looks like. My ready-made audio and reference were used.
In my opinion, the best #AIArt Web site is #Dezgo. I use it almost daily. They just added two new pay-as-you-go #AIVideo features: "text-to-video" and "image-to-video". I think "text-to-video" is a waste of time, but to each their own. That #WanX 2.1 allows NSFW. dezgo.com/app/image2vi...
Apparently, the official #WanX #AIVideo web site will not accept any woman in a bathing suit, including this one. I guess women have the evil nipples and like spitting fruit seeds in the face of God so they're out. You can have all beach-side men you'd like. #Sexism #GenerativeAI
Whatever, #WanX. I'm sure this probably lost something in translation, but when the only things you feel "uninspired" about are women in bikinis, it's hard not to take that personally. #AIVideo
I have more questions than answers from #WanX. They claim you can generate a video "instantly" using credits. What? Also, it appears start/end keyframes are going to be pay-walled soon.
I want you to focus not on the content of this #AIVideo, but rather the quality. This is a sample output from #FloraFauna using the #WanX video model. I provided the original image at 1600x900. It generated a final output of 2.3MB, 1280x720, 3635059 bitrate (high), and 16FPS--good enough for stills.
I got the turn-around I needed by putting him behind her. Also, that is "coyly".
#WanX #PromptShare "Man waves and speaks. Female coyly turns and waves back at him."
This is it: the greatest #AIVideo moment I have had to date. This is the first-ever debut output of my own personal #WanX 2.1 deploy. It cost me 48 minutes of my life and $1.15, but it's the most sturdy 1080p proof-of-concept Docker container I've ever done. I am absolutely fucking jazzed right now.
