A frame from our 1080p launch teaser, generated by Seedance 2.5 from a founder photo and a product shot.
Launch day is the moment a young company has to explain itself in a few seconds. A good launch video does that better than any landing page: it shows the founder, the product and the problem it solves, and it gives every post, email and pitch deck something to point to. It is also one of the first things founders cut when money is tight. A small agency will quote thousands of dollars for a minute of finished video, and filming it yourself, in a cluttered workshop with a phone, rarely looks like the confident launch you had in mind.
We wanted to know how much of that gap a video generation model can close today. So we invented a hardware startup called Sprout, which makes a smart plant pot that tells you when to water, and gave ourselves only three images to work with: a photo of the founder in his workshop, a product photo of the pot, and a screenshot of its app. From those, we used Seedance 2.5 to generate a launch film, a speaking founder statement, product and app clips, versions at three resolutions, a lip-synced voiceover, an edit and an extension.
Nine clips cost $21.17. A realistic launch package, as we show at the end, would cost about twelve dollars. Here is what each step produced, including the parts that did not work.
The starting point: three images
Most founders have more than this on their phone, but we deliberately kept it minimal. All three images were created for this test with image models, standing in for the photos a real startup would already have.
The three references: the founder, the product and the app.
Seedance 2.5, model ID seedance-2.5, can use images in two ways. As a first frame, an image becomes the opening of the clip and the model animates from there. As a reference image, it describes a subject that the model then places into a new scene. Seedance 2.5 accepts up to 30 reference images, 10 reference videos and 10 audio clips in one request, and generates up to 30 seconds at 480p, 720p or 1080p, with sound by default. Requests are asynchronous; our clips took three to six minutes each.
Test 1: a 20-second launch film from all three references
We started with the most ambitious request: a short story using all three images.
A 20-second launch video for a smart plant pot startup. The founder from the reference stands in his workshop, picks up the white smart pot with the basil plant, its LED ring glows green, then he shows the phone app screen from the reference displaying the soil moisture, and finally places the pot on a sunny kitchen counter. Warm, authentic, documentary style, light acoustic music.
Test 1: three of the four story beats came through; the app screen did not.
The film followed the story in order, and it looked like real documentary footage. The founder placed the pot on his workbench, the camera moved to a close-up of the pot with its green ring glowing next to a phone, and the final shot showed the pot on a sunlit kitchen counter. The founder and the product both matched their references: same face, same black T-shirt, same white pot with basil and a thin green ring of light.
The weak point was the app. The phone appeared exactly when the prompt said it should, but its screen was black. The app screenshot we supplied as a reference was never shown. A model placing a recognisable person and product into a scene is one thing; placing a specific user interface onto a screen inside that scene is still harder. The 20-second film cost $4.11, or $0.206 per second at 720p.
Test 2: the founder speaks
A launch video needs a human voice. We used the founder photo as the first frame and wrote his line into the prompt.
The founder looks at the camera and says: “We built Sprout because I killed every plant I ever owned. Now the pot tells you when to water.” He smiles. Workshop ambience.
Test 2: the founder delivered his line with natural expressions.
We transcribed the clip with a speech-recognition model, and the transcript matched the script exactly: “We built Sprout because I killed every plant I ever owned. Now the pot tells you when to water.” The founder’s expression changed as he spoke, from animated to a smile at the end, and his appearance and workshop matched the photo. The ten-second clip cost $2.06. You can hear it in the walkthrough video at the end of this article.
Test 3: the product and the app
Next, two short clips that any product launch needs: a hero shot of the product and a demonstration of the app. Both used the images as first frames.
A finger taps the green “Water now” button on the phone screen and the moisture reading rises smoothly from 42% to 68%.
Test 3: the product push-in, and the app animation with a brief wrong value in the middle.
The product clip was a slow, clean push-in on the pot with the basil moving slightly, exactly the kind of shot that sits under a headline. It cost $1.65 for eight seconds.
The app clip was more interesting. A finger tapped the “Water now” button, the button reacted, and by the end the reading showed 68 percent, as requested. The model even updated “Next watering in 2 days” to “4 days”, which is logical after watering. But in the middle of the animation the number briefly read 33 percent, moving the wrong way before rising. Viewers would barely notice at full speed, but for a product demo that claims accuracy, it matters. Check animated numbers frame by frame, or animate them yourself in an editor. The six-second clip cost $1.24.
Test 4: 480p, 720p or 1080p?
Resolution changes the price a lot, so we generated the same ten-second teaser, the founder holding up the pot, at 480p and at 1080p.
Test 4: at 480p the basil plant disappeared from the pot.
The difference was not only sharpness. At 480p, which cost $0.92, the founder held up an empty white pot: the basil plant from the reference had disappeared. At 1080p, which cost $5.10, the plant and the glowing ring were both present and sharp. One pair of clips is not proof that low resolution always loses details, but it was a clear warning. At $0.51 per second, 1080p is expensive for a startup budget; 720p, at about $0.21 per second, kept the product intact in every clip we made at that resolution and is the setting we would use.
Test 5: lip sync to your own voiceover
Many founders would rather write and record their own voiceover, or use a consistent synthetic voice across a whole campaign. Seedance 2.5 accepts audio as a reference, so we tested it. We generated a short voiceover with a text-to-speech model, “Sprout is the plant pot that tells you when to water. Pre-orders open on Monday.”, 6.6 seconds long, and asked for an eight-second clip of the founder speaking it.
The founder from the reference image stands in his workshop holding the white smart plant pot and speaks the words of the reference audio directly to the camera, natural lip sync.
Test 5: the founder speaks the supplied voiceover, holding the product.
The first sentence came through clearly and in sync: the transcript read “Sprout is the plant pot that tells you when to water.” The second sentence was missing. The clip was eight seconds long, and with the model’s usual pause at the start, 6.6 seconds of speech did not fit. The fix is simple: make the clip at least two seconds longer than the audio. The founder held the pot throughout and spoke to the camera. Judge the lip sync for yourself in the walkthrough video below. This clip cost $1.65.
Test 6: editing and extending
Once you have clips, it is tempting to adjust them rather than regenerate. We tried two adjustments.
Test 6: the edit changed the colour but dropped the light ring; the extension drifted from the references.
The edit asked Seedance to change the white pot in the 1080p teaser to matte black while keeping the founder, the basil, the green LED ring and all motion the same. The founder, the plant, the workshop and every movement were preserved, and the pot became black. But the LED ring, the product’s most distinctive feature, disappeared. The edit returned ten seconds at 720p and cost $2.47.
The extension asked for six more seconds after the speaking clip, with the founder putting the pot down and showing the app. The result had two problems. Seedance returned only the new six seconds, not the original clip plus the extension, so you need to join them yourself. And the new segment drifted from the references: the founder was the same, but the basil had become a different plant in a pot without a light ring, and the phone showed an invented app with a dial and a different layout. The extension cost $1.97, the most expensive per second of anything we generated, because it is billed for the input it reads as well as the output.
What it costs
Billed cost per job and per second.
Before you plan a launch budget, check the current Seedance 2.5 API pricing for the resolution you need, because it drives almost everything. In our account, 720p cost about $0.21 per second whether we used text, a first frame, references or an audio reference. 480p was less than half that, and 1080p about two and a half times as much. Edits and extensions cost more per second than new clips.
A launch package at 720p: about a minute of footage for $12.36.
Our recommended package is a ten-second founder statement in your own words, an eight-second product hero shot, a six-second app demo, a twenty-second story film from references, and budget for two retakes. That is about a minute of footage for $12.36. Cut it together with captions, your logo and a music track in any free editor, and you have a launch video for less than the price of a lunch meeting.
Lessons for founders
- Start from real photos. The founder and the product stayed recognisable in every reference and first-frame clip.
- Write the founder’s words yourself. The spoken line was reproduced exactly; it should be true and in your own voice.
- Use 720p. At 480p the product lost its plant; 1080p was more than twice the price for detail most viewers will not see on a phone.
- Do not expect your UI to appear on screen. The app reference never showed up inside the story film; record a real screen capture for the demo.
- Check animated numbers. The moisture reading briefly went the wrong way.
- Make lip-sync clips longer than the audio. Two extra seconds would have saved our second sentence.
- Prefer new clips to extensions. The extension cost the most per second and drifted from the references.
- Be honest about what is generated. Investors and customers will meet the real founder and the real product; the video should match them.
The walkthrough video
Walkthrough, including the founder’s generated voice and the lip-synced voiceover (about 3 minutes).
Assembling the launch video
None of the clips in this test is a finished launch video on its own. The finished video is an edit, and the edit is where a founder’s judgement matters most. Here is how we would put the package together.
Start with the founder statement, because people connect with people before products. Follow it with the product hero shot under a one-line caption that states the problem, then the app demo, which should be a real screen recording rather than a generated one, and finish with the story film and an end card with the launch date and URL. Thirty to forty-five seconds is enough for social platforms; a minute works on a landing page.
Three technical steps make generated clips feel like one film. First, trim the silence: our speaking clips started with a pause of half a second to a second and a half. Second, normalise the audio, because generated clips arrive at different levels; a single loudness pass to about -16 LUFS for web video evens them out. Third, add captions. Most social video is watched muted, and captions also make the founder’s words searchable. All of this can be done with free tools; we assembled the walkthrough video for this article with ffmpeg, using commands such as these:
# trim the first second of a clip
ffmpeg -ss 1.0 -i founder.mp4 -c:v libx264 -c:a aac founder-trim.mp4
# normalise loudness for web playback
ffmpeg -i founder-trim.mp4 -af loudnorm=I=-16:TP=-1.5:LRA=11 -c:v copy founder-norm.mp4
# join clips listed in a text file, one “file ‘name.mp4′” per line
ffmpeg -f concat -safe 0 -i clips.txt -c:v libx264 -crf 22 -c:a aac launch.mp4
Finally, keep a record of what was generated. If an investor or a journalist asks whether the product in the video is real, the honest answer should be easy: the product and the founder are real, the footage was generated from photos of them, and the app demo is a real recording.
What this test could not tell us
We generated one take of most clips, so we cannot say how often a given prompt succeeds. A real launch would justify two or three takes of each shot, which we included in the budget. We tested one founder and one product; a product with fine text on its packaging, or a founder whose face is partly hidden, might behave differently. And we judged lip sync by transcription and from extracted frames, not with a formal measure.
Appendix: the prompts we used
For anyone who wants to repeat these tests, here are the prompts as we sent them, grouped by clip. Parameters such as duration and resolution are listed after each prompt.
s2-launch-20 (model: seedance-2.5, size: 16:9, resolution: 720p, duration: 20)
A 20-second launch video for a smart plant pot startup. The founder from the reference stands in his workshop, picks up the white smart pot with the basil plant, its LED ring glows green, then he shows the phone app screen from the reference displaying the soil moisture, and finally places the pot on a sunny kitchen counter. Warm, authentic, documentary style, light acoustic music.
s2-founder-speaks (model: seedance-2.5, size: adaptive, resolution: 720p, duration: 10)
The founder looks at the camera and says: “We built Sprout because I killed every plant I ever owned. Now the pot tells you when to water.” He smiles. Workshop ambience.
s2-product-i2v (model: seedance-2.5, size: adaptive, resolution: 720p, duration: 8)
Slow push in on the smart pot as the green LED ring pulses once and a drop of water falls onto the soil, soft morning kitchen sounds.
s2-app-i2v (model: seedance-2.5, size: adaptive, resolution: 720p, duration: 6)
A finger taps the green “Water now” button on the phone screen and the moisture reading rises smoothly from 42% to 68%.
s2-launch-480 (model: seedance-2.5, size: 16:9, resolution: 480p, duration: 10)
A 10-second launch teaser for a smart plant pot startup: the founder from the reference holds up the white smart pot with the glowing green LED ring and smiles at the camera in his workshop.
s2-audioref (model: seedance-2.5, size: 16:9, resolution: 720p, duration: 8)
The founder from the reference image stands in his workshop holding the white smart plant pot and speaks the words of the reference audio directly to the camera, natural lip sync.
s2-edit-black (model: seedance-2.5, size: adaptive, resolution: 720p)
Edit the video: change the white plant pot to a matte black ceramic pot, keep the founder, the basil plant, the green LED ring and all motion the same.
s2-extend (model: seedance-2.5, size: adaptive, resolution: 720p, duration: 6)
Extend the video forward: the founder puts the pot down on the workbench, picks up his phone and shows the app to the camera with a smile.
The bottom line
Seedance 2.5 can turn three ordinary images into most of what a startup launch video needs: a founder who speaks your words, a product that looks like your product, and a short story that ties them together, at about 21 cents per second of 720p footage. It is not yet a substitute for showing your real interface, and edits and extensions can quietly change details that matter. Used for the parts it does well, and checked carefully, it puts a convincing launch video within reach of any founder with a laptop and a free afternoon. If you want to try it, or compare Seedance 2.5 with other video models through a single account and API key, all of them are available on APIMart.ai.
***
