GrandVisionsAI

AI video artist. The goal is for you to not be able to tell.

[GrandVisionsAI @GrandVisionsAI]

This is seriously impressive from @Alibaba_Wan. On my very first attempt, Wan 3.0 produced a more believable acting performance than I achieved across multiple Seedance 2.5 generations, while costing less than half as much.

I’ve tested Seedance 2.5 close to 100 times, specifically focusing on acting, and saw little improvement regardless of how I prompted it. Its acting ceiling feels surprisingly low and, in my opinion, may even be worse than a well-prompted 2.0, which could produce genuinely convincing performances.

I’d love to be proven wrong. If you have examples of strong acting from 2.5, please share them, because I haven’t found any yet.

Excellent work, @Alibaba_Wan team 👏. There is serious potential here.

#Wan3 #Seedance #AIFilmmaking

[GrandVisionsAI @GrandVisionsAI]

Hey man, not sure how many generations you are running but I'd suggest running multiple and trying different prompts, cause wan 3.0 is much better than this, it's the best model for acting (examples on my page).

It's also worth noting Wan has prompt enhancement enabled by default, so it would be worth trying it with prompt enhancement off aswell, if you haven't.

[GrandVisionsAI @GrandVisionsAI]

@eattheethos This is the main issue I have with Seedance 2.5, it looks real but the acting performance is off (especially voices).

Try wan 3.0, you'd be surprised what it can do, I have some examples on my page comparing it to seedance 2.5 and it's miles ahead acting wise.

[GrandVisionsAI @GrandVisionsAI]

@eeelsi @mightyking I'd say the opposite, Wan 3.0 has given me the best acting performances of any model, by far.

[GrandVisionsAI @GrandVisionsAI]

@aplifere @Alibaba_Wan @TopviewAIJP @TopviewAIhq Thanks, glad you enjoyed it!

[GrandVisionsAI @GrandVisionsAI]

Another incredible example of what Wan 3.0 can do, this time with anime.

Huge appreciation to the @Alibaba_Wan team for building something this amazing. I’d love the opportunity to work with the team in any capacity, help push the model even further, and showcase what it’s truly capable of. 🙏

And since many people have asked, I’m using Wan 3.0 through @TopviewAIhq @TopviewAIJP. My plan includes free 720p generations. It can be slow, but it’s reliable, and I consistently get at least a few videos generated each day. Not sponsored by the way.

I’d love to work in 1080p and create longer form content, but that’s beyond my budget at the moment. Hopefully soon!

[GrandVisionsAI @GrandVisionsAI]

@JWallaceParker Thanks!

[GrandVisionsAI @GrandVisionsAI]

My entry for Topview AI’s 30-Second Challenge.

I’m not seeing many dialogue scenes in this challenge, which makes sense. The visuals are what most people care about, after all.

And even in the dialogue scenes I have seen, there’s often this huge emphasis on making everything LOOK cinematic. Color grading, upscaling, extra polish, pushing the image as far as possible. All of that visual polish to keep your eyes busy enough that you don’t notice the wooden acting underneath.

I guess that works for most people. It just doesn’t do much for me.

Wan 3.0, you know the drill. Raw 720p. No color grading. No upscaling. I’m trying to make people feel something here, and for me, a strong performance trumps even the most visually spectacular Seedance video you could ever make.

To my Japanese friends, please let me know how the Japanese sounds. It seems pretty good to me... でも、私は日本人じゃない 😅

And everyone else, let me know what you think as well!

@TopviewAIhq @TopviewAIJP #Topview30sChallenge
#TopviewAI

[TopviewAI @TopviewAIhq]

Think you can make an impact in just 30 seconds?

The Topview 30s Challenge is here! 🚀

Create. Experiment. Get creative.

Show us what you can do and turn 30 seconds into something unforgettable.

👉 Join the challenge: https://t.co/JDNAiVUd2S

#Topview30sChallenge

[GrandVisionsAI @GrandVisionsAI]

@SteveEricsson1 @Alibaba_Wan @TopviewAIJP @TopviewAIhq Try out Wan 3.0 on the unlimited queue and see how it goes. This is from their Discord.

[GrandVisionsAI @GrandVisionsAI]

@PixVerse 👀👀👀

[PixVerse @PixVerse]

We're giving away 6,000,000 credits.
One mission: create.

Fuel your next AI film with PixLight.
Start creating. Start your story.

Follow + RT + Reply = 150 Creds & Access to 6M Creds in DMs(72H ONLY)

[GrandVisionsAI @GrandVisionsAI]

@Chizzydigital @Alibaba_Wan I provided a reference for the character on the right, but only a description for the character on the left.

Depending on your description you sometimes get characters that are...inspired by real works, happens in every video model.

[GrandVisionsAI @GrandVisionsAI]

This is seriously impressive from @Alibaba_Wan. On my very first attempt, Wan 3.0 produced a more believable acting performance than I achieved across multiple Seedance 2.5 generations, while costing less than half as much.

I’ve tested Seedance 2.5 close to 100 times, specifically focusing on acting, and saw little improvement regardless of how I prompted it. Its acting ceiling feels surprisingly low and, in my opinion, may even be worse than a well-prompted 2.0, which could produce genuinely convincing performances.

I’d love to be proven wrong. If you have examples of strong acting from 2.5, please share them, because I haven’t found any yet.

Excellent work, @Alibaba_Wan team 👏. There is serious potential here.

#Wan3 #Seedance #AIFilmmaking

[GrandVisionsAI @GrandVisionsAI]

@Liftmodez @Alibaba_Wan 🤣 it's his cousin.

[GrandVisionsAI @GrandVisionsAI]

@SteveEricsson1 @Alibaba_Wan @TopviewAIJP @TopviewAIhq No it's not, they have separate queues, I usually get 4+ videos of Wan 3.0.

[GrandVisionsAI @GrandVisionsAI]

@vgonearth @Alibaba_Wan Im still experimenting with prompting. So I can't say for sure you'll get the best results doing what I did, simpler prompts may be better.

[GrandVisionsAI @GrandVisionsAI]

This is a part of the prompt I used, same one from SD 2.5 where I explicitly state the pauses (necessary for 2.5).

Osric is located on the right side of the frame, slightly further from the lens. He wears damp chainmail with a dark, worn leather gorget at the neck. He is squatting on his heels, maintaining the same eye level as Hakon, appearing at a clearly visible, near-frontal three-quarter angle. Osric's sword remains sheathed on his left side at all times; he never touches it. Both faces must be clearly visible. The background includes churned-up mud, leaning and cracked wooden palisades, pale smoke drifting continuously toward the right side of the frame, and a colorless dawn sky. Osric looks into Hakon's eyes and speaks first. His tone is flat and procedural, using this offer as leverage, not an act of kindness: OSRIC: “They want a name... Give me one and I can keep you alive until noon.” A brief, natural pause occurs at the ellipsis. Hakon lets the silence linger for a moment. He recognizes the familiar manner of speaking beneath Osric's authoritative tone and replies with weary, old humor rather than mockery: HAKON: “You still say it the same way... Like you’re asking for salt.” Osric keeps his mouth still while listening. The shared memory hits him before he replies. Osric tries to reject the memory, but his procedural facade cracks. He does not raise his voice or become harsher. The first word carries regret and pain, sounding more like a plea than a command. He pauses authentically at the ellipsis before completing the sentence: OSRIC: “Don’t... We’re past that.”
After speaking, he immediately tries to seal his emotions back up. Hakon thinks briefly, then quietly delivers the message, without cruelty or vindictiveness. HAKON: “She asked for you....At the end..... Not for me—for you.”

[GrandVisionsAI @GrandVisionsAI]

@Tony_Anjuna @SamJWasserman Thanks for sharing. I usually use blender mcp with claude but it's a real pain in the ass cause I'm a noob at blender. Gonna try this next time.

[GrandVisionsAI @GrandVisionsAI]

Yes, but consistency will always be an issue.

For example, in a shot where your character’s face appears small and/or unclear, the upscaler has no reliable reference for what the character should actually look like. It can only make its best approximation, producing a sharper, more detailed face that will not fully match the character’s original appearance.

[GrandVisionsAI @GrandVisionsAI]

Another incredible example of what Wan 3.0 can do, this time with anime.

Huge appreciation to the @Alibaba_Wan team for building something this amazing. I’d love the opportunity to work with the team in any capacity, help push the model even further, and showcase what it’s truly capable of. 🙏

And since many people have asked, I’m using Wan 3.0 through @TopviewAIhq @TopviewAIJP. My plan includes free 720p generations. It can be slow, but it’s reliable, and I consistently get at least a few videos generated each day. Not sponsored by the way.

I’d love to work in 1080p and create longer form content, but that’s beyond my budget at the moment. Hopefully soon!

[GrandVisionsAI @GrandVisionsAI]

@Jenny_MommaLion @Alibaba_Wan Thanks for sharing!

[GrandVisionsAI @GrandVisionsAI]

@StevieMac03 Thanks for the tip. Not sure if anyone else has mentioned it but, SAM is pretty good at this (you can isolate almost anything).
https://t.co/Vzv5JiETPy

[GrandVisionsAI @GrandVisionsAI]

@re_skob @Gossip_Goblin

[GrandVisionsAI @GrandVisionsAI]

@realrebelai @InvestorCrack I wanted to experiment with Flux 3 but its too expensive. How does the open version compare quality wise, and what are the hardware requirements like?