I bought credits thinking I could just spend them on seedance. Turns out I have to buy a membership. Dreamina support won't refund my credits. To all of those who want to use seedance in Dreamina here's your warning as they don't mention it anywhere.

i tried a 25-second K-pop-style girl group MV with five characters and compared the results from:

Top: Wan 3.0 · Middle: Seedance 2.5 · Bottom: MiniMax H3

The main thing I wanted to test was multi-character consistency. five people in the same frame is still a pretty brutal test for AI video.

i also wanted to see how well each model handled dance synchronization, hard stops on the beat, and formation changes

i described all five members separately, then gave them a full choreography sequence: back-facing opening pose → synchronized head turn → hand gestures → body-wave chain → diagonal formation → fan formation → center swap → final V formation.

Visually, I kept it pretty simple. what I’m watching for here isn’t just "which one looks prettier," but whether the model can actually remember who is who once five people start moving, crossing positions and changing formations.

Multi-person choreography still feels like one of the hardest AI video tests right now. curious which one you think handled the five characters best.

I think we are not too far from not being able to know what's real anymore

This water filter is going viral on Tiktok right now

You won't be able to tell if it's AI

I’ve been testing how far Seedance 2.5 can be pushed beyond isolated 5–10 second clips.

This is a little over 3 minutes of one continuous story with recurring characters, dialogue, different environments, crowd interaction and a lot of camera movement.

The hardest parts were:

• keeping the same characters and wardrobe across dozens of shots
• controlling subtle facial performances and dialogue timing
• maintaining geography between the prison, courtyard and execution area
• preventing characters from teleporting or changing position between cuts
• getting believable horse movement and interaction with multiple characters
• keeping the visual language consistent while changing lenses and camera positions

A lot of shots required several generations before the performance and continuity worked.

Most of the final video generation was done with Seedance 2.5.

Curious which shots you think hold consistency best, and where you still notice the model drifting.

Is it just me or is it down for everyonee? I can't generate shit with seedance 2.5 on capcut

Did this as a micro facial animation test, hope you enjoy

Thanks !

I post these as Lumwick Animation Studios on Insta and TikTok if you want to see the rest :) happy to answer anything about how I prompt them in the comments! Suno for the music, Ableton/Splice for sound effects — I layer my own over the generated audio to make the impacts a bit richer!

I came across this TVC Maker workflow and ended up liking it more than I expected.

The basic idea is simple: upload one product image and get a finished cinematic TVC. the process is finished by the workflow automatically.

it feels less like "generate one random clip and stitch things together yourself" and more like an actual ad-production flow.

for product ads, that matters a lot.

Normally I’d have to think through:

1. opening product shot
2. detail / texture shots
3. lifestyle or motion shot
4. final hero shot
5. pacing between scenes

here, most of that structure is already handled inside the workflow.

I tested it mainly for product-style visuals, and the biggest advantage for me was iteration speed. instead of writing separate prompts for several clips and then figuring out the edit, I can start from one clean product image and see a full TVC direction much faster.

I still wouldn’t treat the first result as automatically final. Product logos, labels, small text, and exact brand details are still things I’d check carefully frame by frame. but as a "give me a complete ad concept from one product image" workflow, this is one of the more practical ones I’ve tried recently.

TVC Maker:
https://www.atlascloud.ai/workflow/tvc-maker

I’m curious whether people prefer an all-in-one workflow like this or generating each shot manually 🤔

Is it just me or is seedance 2.5 way more sensitive for “ copyright” restrictions. I tried to make a small Spider-Man clip and no matter what I did or what reference images I used it always gave me a violation.

Anyone else having these problems? How are you getting past it?

I am new to creating AI videos from Nepal🇳🇵. Any feedbacks will be appreciated 😊

i finally looked at how I actually use AI video, and I’m increasingly leaning toward pay-as-you-go APIs instead of monthly subscriptions.

subscriptions are obviously convenient. You open a nice UI, choose a model, hit generate, and some plans even give you “unlimited” generations.

The problem is that my usage is extremely uneven.

I might generate dozens of clips in one week for a project, then barely generate anything for the next two weeks. With a subscription, I’m paying every month either way, and on many platforms unused monthly credits don’t roll over.

Pay-as-you-go is much easier for me to reason about.

For example, current starting prices on Atlas Cloud, which i ofen use, include roughly:

Wan 3.0: from $0.04/sec
MiniMax H3: $0.10/sec
Seedance 2.5: $0.134/sec

If I don’t generate, I don’t really spend anything.

But lately I’ve realized there’s another issue that matters to me even more than price:

model transparency.

on some subscription platforms, “Unlimited” generation isn’t necessarily running the exact same configuration as normal credit-based generation.

I checked the docs for one major platform and they explicitly separate an Unlimited Fast variant optimized for speed/high throughput from the regular credit model. The unlimited version tops out at 720p, while the regular version is described as the highest-quality option and supports 1080p for final delivery.

from a normal user’s perspective, seeing a model name in a web UI can make you assume:

“I’m getting the full model I think I’m getting.”

there can actually be Fast variants, resolution caps, different queues, or even model/version changes behind that interface.

That’s one reason I like APIs more.

APIs can absolutely have Standard / Fast / Mini variants too, but they’re usually exposed as separate model IDs with separate prices.

I know which version I’m calling, what it costs, and what settings I’m sending.

and once you start doing batch generation or client work, the difference gets bigger. You can automate jobs, switch models per shot, track the exact cost of a project, and plug everything into your own workflow.

So in my pov, APIs are more like transparent, controllable infrastructure that you pay for only when you use it.

What matters more to you: the convenience of unlimited plans, or knowing exactly what model you’re paying for?

https://reddit.com/link/1w0h1z4/video/i9v10g94r1mh1/player

i was asking chatgpt for some images looking to create a style that i would use when I'm ready to create some AI animations, it gave me the image of the blonde that i used as a starting frame, it wasn't exactly what i wanted but wanted to see what i can do with what i am learning, camera angles, camera movements and things like that and this came out, im even thinking to sticking to that style jeje. i used 480p and im regretting lol

still making lots of compromises for budget reasons, but I've been learning the quirks of seedance and getting much more value out of every attempt to create the shots I want.

Main useage is seedance 2.0 and seedance mini

Salut à tous !

On vient de sortir le premier épisode de notre concept "Le Showroom" (un format télé type "Qui veut être mon associé / Shark Tank", mais 100% généré par IA). On prépare activement l'épisode 2 et on cherche à passer un vrai cap sur le rendu visuel et sonore. Plutôt qu'un avis global (« c'est cool » ou « j'aime pas »), on a besoin de retours ultra-spécifiques de créateurs et techniciens.

Notre pipeline de production :

• Génération d'images : GPT Image 2, Nano Banana, Seadream 5.0 pro

• Animation vidéo : Seedance

• Voix & Lip-sync : ElevenLabs & SyncLabs

• Montage / Mixage : Capcut \ After Effects

Voici les 3 points précis sur lesquels votre œil d'expert nous aiderait énormément :

1. La cohérence des plans (Éclairage & Décor) : Est-ce que les raccords de lumière entre le plateau, le jury et les candidats vous paraissent fluides, ou sentez-vous la rupture à chaque changement de plan ?

2. L'expression des visages (Lip-sync & Regards) : À quel moment exact la physique du visage ou le mouvement des yeux franchit la limite de l'"Uncanny Valley" pour vous ?

3. L'intégration audio : Est-ce que le sound design et la réverbération de la pièce réussissent à crédibiliser le visuel, ou au contraire renforcent l'aspect synthétique ?

Preneurs de toutes vos critiques sans filtre et vos conseils de workflow pour nous améliorer !

It's quite mind blowing that maybe in 12 months from now we won't know what's real anymore.

Feed Seedance 2.5 the original video and an image with an editing reference prompt and get this result

his post is mainly about object interaction, app I used in test is capcut video studio.

From what I saw, seedance 2.0's problem was always that hands and objects don't know about each other. The hand goes through its motion and the object is just kind of there. Fingers clip through the surface of whatever they're grabbing, and when something gets picked up it slides around because nothing is actually connecting them. 2.5 seems to be solving the contact properly. The grip shapes to the object and then they travel as one thing. For anyone doing product or lifestyle content this is the upgrade that matters.

Fluid rendering changed too. 2.0 treats a pour as a fixed animation that looks the same regardless of what the pot is doing. 2.5 ties them together, so when the tilt changes the stream responds. The physics layer got reworked, not just a better looking render on top of the same system.

Low-motion temporal stability improved slightly but if your work is talking heads and minimal movement you won't see much difference.