Contents
0%Hey!
Seedance 2.0, the latest AI video model, has just been released. You can now turn a single product photo into a complete video ad in one shot. It delivers A-roll, B-roll, and voiceovers seamlessly, while giving you absolute control and 100% product consistency.
See It In Action
Let's start with this. I gave Seedance just two product images of this supplement brand and let it do its thing.
Gruns bottle
Gruns gummies referenceGruns bottleGruns gummies referenceThe scene creativity is wild. It opens with the raw product shot, then cuts to B-roll of a woman at the office, then doing yoga. It genuinely feels like a real UGC video.
What's really cool is the second input image. I gave it a photo of what the actual Gruns gummy bears look like. In the video, the gummies show up as real green bears, which is something a video model would almost never figure out on its own. That's the power of Omni-Reference mode. You give it context, and it actually uses it.
What Makes Seedance 2.0 Different
Here's a quick look at how Seedance 2.0 compares to Sora 2 for UGC. The short version: more control, longer generations, and a fraction of the cost.
| Feature | Seedance 2.0 | Sora 2 Pro |
|---|---|---|
| Image inputs | Up to 9 images | 1 (start frame only) |
| Video inputs | Up to 3 clips | None |
| Audio inputs | Up to 3 files | None |
| Max generation | 15 seconds | 12 seconds (lose 1-2s to start frame) |
| Edit & extend | Yes | No |
| Lipsync & voice | Built-in, best in class | Not available |
| Sound effects & music | Built-in | Not available |
| Cost | ~$11/month | ~$200/month |
Generating UGC with Seedance 2.0 feels way less like playing a slot machine and much more like you're actually directing your own scenes. Omni-Reference mode (up to 9 images + 3 videos + 3 audio files as input) gives you real control. And you can edit and extend scenes after they're done.
Now let me walk you through all the different UGC formats I've tested.
Freestyle UGC
In this section, you'll see what happens when you give Seedance minimal direction and let it improvise. Just product images in, full UGC video out.
Gym Supplements
Single product image. No detailed prompt. Seedance just improvised the entire gym workout scene.
Gym supplement productGym supplement productWhat makes freestyle mode so interesting is Seedance's ability to invent a scene that actually matches the product. It doesn't just slap the product into a random context. It understands what the product is and builds a story around it.
For comparison, I ran the same products in Sora 2 Pro. The main issue was the aspect ratio, which is a known limitation when using the Start Frame. The video itself also felt too polished, more like a commercial than authentic UGC.
Controlled UGC
This is where things get really interesting. Instead of letting Seedance improvise, you give it specific directions on the location, action, and how you want the video to play out. Here's what that looks like across a few different niches.
Beauty / Makeup (Lipstick)
I gave the model the actual lipstick product image and wrote a very detailed prompt about it being a peel-off lipstick.
Lipstick productLipstick productThe fact that it understood "peel-off lipstick" and nailed the application AND the peel is insane. This is where I started to realize Seedance actually reads and follows the prompt closely.
Sora 2's version was okay on the application part. But it completely failed on the peel-off. And because the max generation is 12 seconds with the first second lost to the start frame, every Sora 2 video ends up feeling rushed.
Supplements (PINC)
For this one, I gave it specific instructions on the location and the exact action I wanted the actor to perform.
PINC supplementPINC supplementThe prompt adherence on this one is great. It followed every detail I gave it. High realism, lipsyncing on point, and the voice sounds natural.
Sora 2's version was still realistic but gave off a studio vibe instead of authentic UGC. The product sizing was also a bit off.
Beauty / Skincare
Skincare productSkincare productThe skincare niche is a great test because the product interaction has to look natural. Nobody wants to watch someone awkwardly apply a cream. Seedance nails the hand movements and the way she holds the product.
Street Interviews
This format has been blowing up in the UGC niche. I wanted to see if Seedance could handle natural back-and-forth dialogue, camera movement, and voice pacing.
Host (Starpop mascot)Host (Starpop mascot)The control and accuracy here is on another level. The camera work, the voice pacing, the way the people actually talk. It all feels real.
But the most impressive part? The prompt adherence on the dialogue. I gave it a back-and-forth script between the interviewer and the guest, and Seedance followed it perfectly. Every other model I've tested mixes up who says what. They never properly follow the dialogue prompt. Seedance just gets it right.
Try-On & Unboxing
This is the ultimate Omni-Reference test. I gave Seedance 6 input images for this one. Product packaging, clothing items, the environment, and specific angles I wanted.
Skims jacket
Outfit reference 1
Outfit reference 2
Outfit reference 3
Outfit reference 4
Outfit reference 5Skims jacketOutfit reference 1Outfit reference 2Outfit reference 3Outfit reference 4Outfit reference 5It used all six images to create an unboxing video that actually looks real. The more context you give Seedance through Omni-Reference, the less you're leaving up to chance. And that's the whole point. You're not hoping it generates something good. You're telling it exactly what to create.
Tips & Tricks
After weeks of testing Seedance 2.0 for UGC, here are the three biggest things that will save you time and credits.
Be Specific with Omni-Reference
Seedance improvises really well on its own. You saw that with the freestyle examples above.
But if you want a specific result, take the time to give it detailed references. The difference is massive. For the unboxing video, I gave it 6 input images. The product packaging, the clothing items, the environment, the angles. And it used every single one.
The more context you give through Omni-Reference, the less you're leaving up to chance.
Use Blurry Images for Character Variety
You'll notice pretty quickly that Seedance tends to generate characters that look similar. And you can't give it a close-up photo of a real person because that usually gets blocked.
But there's a middle ground. If you give it an image of a person that's blurry or unclear, just enough to suggest the features, it creates a much more unique character consistently. If you're making multiple UGC videos and need the actors to look different from each other, this is the move.
Use Edit and Extend for Character Consistency
This is genuinely a game-changer and the biggest advantage over every other model right now.
Let's say you generated a UGC video and you're happy with it. Great character, great scene, but you need more footage. With any other model, you're done. You'd have to start over.
With Seedance 2.0, you take that video, feed it back in, and say "extend this and make the actor continue saying..." whatever you need. It keeps the same character, same environment, same voice. Even Veo 3's extension couldn't maintain voice consistency. Seedance nails both.
Next Steps
You can access Seedance 2.0 through Jianying.com right now. It will also be available on other platforms soon, including Starpop.AI.
All the testing I've been doing is being turned into ready-to-use templates on Starpop. Those will be available soon.
Want the full guide with every prompt, every input, and all the reference files I used? Join our Discord (link below) and you'll find a Notion page with everything. You can also send me your product and I'll generate AI UGC videos for you, completely free.
A lot of the examples in this post actually came from people in our Discord who sent me their product images. I ran them in Seedance 2.0 for free, and now they can use the videos however they want. If you want the same, just join and drop your request.