Seedance V2 dropped this week, and it is the moment AI video quietly turned into AI editing.
On a recent episode of Greg Isenberg’s podcast, Sirio — founder of the AI creative platform Enhancor — walked Greg through six use cases he has already shipped on top of the model. The conversation was not another “look how cool this is” reel. It was a practical breakdown of how a creator or founder can build an ads engine, a content machine, or an AI-influencer studio on top of this single model.
If you schedule social content for a brand, run e-commerce ads, or ship UGC at scale, the takeaways matter. And if you are the person who has to turn all that footage into a posting calendar across Instagram, TikTok, YouTube, LinkedIn, and X, there is a very specific way to plug Seedance into a workflow that does not break you.
Here is what Sirio showed, what it means for the next 12 months of social content, and how to turn it into a distribution system.
The headline: Seedance is the first real multi-input video model
Every generator before this one played the same game. Give it a prompt. Maybe a first frame. Maybe a last frame. You got a clip, and you hoped the camera moved somewhere sensible.
Seedance V2 accepts multiple inputs at once — up to two images, two videos, and an audio file in a single generation. In the demo below, Sirio feeds the model a green-screen clip of two characters and tells it, in plain English, to replace both characters and swap the background. The motion is preserved. The camera work is preserved. The two new characters walk into the same scene and do the same thing.
That is the capability Greg and Sirio kept coming back to during the episode: Seedance is not a generator. It is a video editor that happens to be driven by natural language. Character swap, background swap, text preservation, ad translation, template population — all of it runs from a prompt and a small pile of reference material.
Sirio put it bluntly in the conversation: “Seedance 2, it’s not only a video generator, it is a video editor. That’s how I see it.”
That distinction is the whole story. Once you stop thinking “I need to generate a video from scratch,” and start thinking “I need to edit this existing asset with references,” everything downstream gets easier — including the part where you have to post ten variations of it across channels before Friday.
Use case 1: Virtual try-on in -30°C Montreal
Sirio filmed himself walking around Montreal in shorts. The temperature was minus thirty. Instead of reshooting in a jacket, he fed the clip to Seedance along with a reference image of a winter outfit and asked the model to dress him, and — for fun — to have a bear walk by.
What makes this more than a party trick:
- His face stays exactly the same. Sirio, who builds AI models for a living, said he would not have clocked it as AI if he had not generated it himself.
- The boots, the pattern on the pants, and the cut of the outfit all match the reference image.
- The model tracks the bear with its eyes and head as it walks through the frame.
The commercial version of this is obvious. An e-commerce brand with a single model shoot can now produce dozens of clean variations — new outfits, new environments, new props — from one source clip. The motion stays consistent, the identity stays consistent, the brand asset library gets 10x bigger without another studio day.
For anyone running a UGC pipeline, this is the new base layer. You shoot once. You generate twenty. You then schedule each variation to a different channel, a different audience, a different region.
Use case 2: Ad translation without reshooting the ad
This one is the most underrated of the six demos. Sirio takes a Chinese-language ad for a pair of glasses. A model walks into frame, talks about the product, winks, taps the frame of her glasses, and turns away. It’s a beautifully shot spot.
He feeds Seedance the original video, a reference image of a completely different model, and a prompt asking it to swap the talent and translate the voiceover to English. A minute later:
The wink is the same. The hand motion on the glasses is the same. The camera blur is the same. The only things that changed are the face, the outfit, and the language.
This is A/B testing at its theoretical limit. Hold motion, framing, camera movement, and product constant. Vary the demographic, the language, and the creative tone. Measure which combination converts, then double down on the winners.
For media buyers, the math gets frightening. A creative team that used to produce four variations a week for Meta and TikTok can now produce forty — all with identical motion grammar so the creative test is actually clean. That scale is also where most teams break, because suddenly there are forty clips that all need captions, cuts, thumbnails, and a posting schedule across five channels. That is where automation has to step in, and it is the exact problem Postiz was built for.
Use case 3: AI influencers that don’t feel like AI influencers
Sirio’s most impressive demo was a 10-second clip of a woman holding a drink and doing a taste test. The source was a Nano Banana Pro image. The motion and the voice were generated inside Seedance with a single, extremely long prompt.
Greg reacted to it in real time: “I have goosebumps on that one. That looked real.”
The key Sirio shared for lip-sync and emotional realism is specific enough to screenshot:
Do not prompt emotions by saying “the character is sad” or “the character is happy.” You have to describe the muscle movements, the transitions in emotion, the transitions in tone, the transitions in body language.
In other words: “her cheeks rise slightly, her brows soften at the outer edge, then her lips part into a small, surprised half-smile” beats “she looks happy” every single time.
You can put anything inside quotation marks in the prompt and the character will say it on camera, matched to mouth movement. Sirio also pointed out that the product in her hand — the text on the can, specifically — did not warp, wrap, or re-letter across frames. Text persistence has been the quiet killer of every AI UGC attempt to date. Seedance cleaned it up.
If you run a faceless channel, a brand shop account, or a creator-persona ad account, this is the new floor. The bar for “that’s clearly AI” just moved up.
Use case 4: 3D product templates and filling in the middle of a video
Two more uses Sirio demoed, quickly, because they matter for the same reason:
Template-based 3D product shots. Take a neutral 3D-render template of a package on a studio background. Feed it a reference image of your branding. Seedance pastes the texture onto the package, keeps the lighting, keeps the camera move, and returns a fully-branded product animation. Anyone who has priced a 3D motion package for a product drop knows what this collapses.
Video extension — including filling gaps. Seedance can extend the tail of a clip (generate what happens next for a few extra seconds) and fill in the middle between two existing clips. For editors, the “I wish this clip were three seconds longer” problem is solved. For ad teams, it means a 6-second asset can be stretched to 15 without losing continuity.
Both of these are boring on the demo page and enormous in the production pipeline.
The prompting playbook
Across every demo, the same three rules kept surfacing:
- Be verbose. Seedance rewards specificity in a way most models punish. Short, punchy prompts work on Kling 3. Seedance wants detail — especially for identity, motion continuity, and transitions. Sirio drafts his own prompt first, then runs it through Claude Opus 4.6 to tighten and expand it before feeding it to the model.
- Your reference image is the whole game. “Everything starts with a very good source reference image,” Sirio said. The model mimics taste from what you feed it, the way a smart intern would. A weak reference gives you a weak video. Spend real time here.
- Describe muscle movement, not emotion. Covered above, but worth repeating: the realism of a lip-synced AI character is carried almost entirely by the specificity of the facial description.
If you only take one thing into your next prompt, make it the second one. A great reference image turns a mediocre prompt into a good video. A bad reference image cannot be saved by any prompt.
Is Seedance the only model that matters now?
No — and this was the part of the conversation most tutorials skip.
Greg asked Sirio directly: should we just make this the default?
Sirio’s answer was nuanced. Seedance is the state-of-the-art default for editing and multi-input generation. But Kling 3 still wins on cinematic feel — when you need a short clip to look like it came out of a film camera, it holds up better. Veo is still strong in specific categories. And Enhancor V4 — Sirio’s own fine-tuned model — is better at talking-head realism and low-fidelity creator-style content where the cinematic look would actually hurt.
The real answer for a creator or ads team: Seedance becomes the daily driver. The other models become specialist tools you reach for when Seedance’s “look” is the wrong look.
Price also matters. Seedance V2 is fast (roughly sixty seconds per generation in the demos) and generates up to 720p today. A 1080p version is on the roadmap, and once that ships, the gap with other models closes further.
What happens to Adobe
Greg closed the episode with the question everyone building in this space is quietly asking: Adobe is a $106 billion company that has owned creative software for twenty years. What happens to it?
Sirio’s read was measured. Adobe stays relevant because creative professionals still need precision — frame-by-frame edits, 8K output, color grading, the post-production work that agents can’t fully own yet. The opportunity for Adobe is to stop trying to be the place where content is generated, and lean into being the place where content is finished. Let the AI platforms generate. Let Adobe handle the final polish, ideally with an agentic editor layered on top of its classic tools.
In other words: the stack splits. Generation moves to tools like Seedance, Enhancor, and Nano Banana. Polish stays with Adobe — or whoever builds the agentic version of Adobe first.
Where this leaves the people who actually have to post it
Here is the part of the story most AI-video content skips: generating the clip is the easy half of the job now.
The hard half is what happens after. One Seedance generation becomes:
- A 9:16 cut for Instagram Reels and TikTok
- A 16:9 cut for YouTube and LinkedIn
- A 1:1 cut for the Meta feed
- A shortened teaser for X
- A longer cut for Pinterest and Threads
- A repurpose for a Reddit post in the right subreddit
Across five language variations. Across three product SKUs. Across two creator personas. Scheduled to the right times, in the right time zones, tagged with the right first comment, pinned in the right order.
A creative team that used to produce four variants a week is now going to produce forty. That scale breaks human scheduling workflows very fast.
That is the seam Postiz is built for. Upload the Seedance output once, and ship it to 28+ channels — Instagram, TikTok, YouTube, LinkedIn, X, Reddit, Facebook, Pinterest, Threads, Bluesky, and the long tail — from one calendar. Attach AI-generated captions. Route posts through team approval. Track performance per channel per variant. And, if you build on top of it, drive the entire posting pipeline with the Postiz API, MCP, or the AI agent so a new Seedance generation triggers an entire multi-platform schedule automatically.
The creators who win the next 12 months of AI video are not the ones generating the prettiest clip. They are the ones with the shortest path from “generation finished” to “live on every channel.”
The takeaway
- Seedance V2 is the first widely-available multi-input video model, and it is better thought of as an AI video editor than a generator.
- The six demos Sirio walked Greg through — character swap, virtual try-on, ad translation, 3D template branding, video extension, and AI-influencer lip sync — are each a full product category on their own.
- Strong reference images and highly specific prompts (including muscle-level emotional description) are the two biggest quality levers.
- Seedance becomes the default; Kling, Veo, and specialist models stay in the toolbox for specific looks.
- Adobe survives, but as the post-production layer — not the generation layer.
- The bottleneck shifts from generation to distribution. Whoever solves that ships the most content and wins the algorithm.
Try this workflow
If you want to start shipping AI video across every channel this week:
- Pick one of the six use cases Sirio demoed. Start with virtual try-on or ad translation — they are the fastest to ROI.
- Build a reference library. Three clean product images, two brand-tone voice samples, one locked-down source clip. This is the asset you will reuse for every future generation.
- Write your prompts twice. Draft once yourself, then run it through Claude Opus 4.6 with the instruction “optimize this prompt for a multi-input video-editing model.”
- Generate, then treat each clip like a base asset. Cut it into 9:16, 1:1, and 16:9 versions.
- Load the variants into Postiz, schedule them across every channel that matches the creative, and let the analytics tell you which variant to double down on.
Seedance V2 gave creators a hundred new things to make. The ones who get paid will be the ones who figure out how to ship all hundred without breaking.
Ready to stop posting one channel at a time? Start scheduling your AI content across 28+ social platforms with Postiz. One upload, one calendar, every channel — and it plugs straight into whatever you are generating with Seedance, Nano Banana, Kling, or Veo.
