AI Summary
5 min readSeedance 2.0: Make 100 AI Ads in 33 Mins
The Multi-Input Breakthrough
Seedance 2.0 launched as what its users describe as the first AI model that allows for multi-input generation. Instead of feeding a single image or text prompt, you can now combine up to two images, two videos, and an audio file — and the model merges them into one coherent output based on a natural language prompt. The host and his guest, Sirio, frame this not as a video generator but as a video editor. The distinction matters: you are not starting from scratch; you are compositing existing assets and telling the model what to keep and what to change.
The demo that opens the episode makes this concrete. Sirio shows a green-screen video of two people, entirely AI-generated, and then replaces both characters and the background simultaneously. He tags each input in the prompt — character one, character two, background image — and the model preserves the original motion while swapping everything else. The generation takes about sixty seconds. The motion control, he emphasizes, is controlled through natural language. You are not keyframing or masking; you are describing what should stay the same and what should change.
The Reference Image Principle
Continue reading the full summary in the app — free to try.
Read Full Summary →Free • No credit card required
Never miss an episode of The Startup Ideas Podcast
Get every new episode summarized in your inbox — free, ~5 minutes to read.
No spam. Unsubscribe anytime.
What you'll learn
- 1 (00:00) **Episode Introduction & C Dance 2.0 Overview** - Host Greg introduces guest Sirio and frames the episode as a practical guide to building businesses with C Dance 2.0, the world's greatest AI creative model.
- 2 (03:30) **Multi-Input Demo: Replacing Characters & Backgrounds** - Sirio demonstrates replacing two characters and a background in a green-screen video using multi-input tagging.
- 3 (06:10) **Prompting Strategy for C Dance 2.0** - Sirio shares best practices for writing effective prompts, emphasizing specificity and the use of LLMs like Claude.
- 4 (08:04) **Importance of High-Quality Source References** - Sirio explains that great output starts with great input images or videos that match your creative vision.
- 5 (09:48) **Virtual Try-On Demo: Replacing Outfits in Video** - Sirio shows a video of himself in Montreal, where AI replaces his outfit and adds a bear walking by, all while preserving his identity.
- 6 (13:12) **Translation & Character Replacement for Ads** - Sirio demonstrates translating a Chinese ad into English while replacing the model with a different person, keeping all motion identical.
- 7 (16:04) **Product Packaging Replacement in Video** - Sirio shows how to replace a generic 3D-rendered package with a branded image, keeping all other video elements consistent.
+ Full timestamped outline available in the app
Show Notes
In this episode I sit down with my friend Sirio, one of the most creative AI minds I know, to break down Seedance V2. Sirio walks us through the exact use cases, prompts, and tactics he's using to build on top of this model inside his platform Enhancor, covering multi-input generation, virtual try-ons, ad translation, AI influencers with lip sync, video extension, and 3D product template replacement. I wanted this to go beyond the "look how cool this is" tutorials and focus on how creators and founders can actually build businesses, run ads, and produce creative assets with it. By the end, you'll have a practical playbook for Seedance V2 and a clear view of where it fits alongside other models like Kling 3, Veo, and fine-tuned options.
Timestamp
00:00 – Intro
02:22 – Demo 1: Replacing Characters and Background in a Green Screen Scene
08:03 – Prompting Tactics and Optimize Prompts
09:45 – Demo 2: Virtual Try-On in Montreal (Minus 30 Degrees)
13:05 – Demo 3: Ad Translation and Character Replacement (Chinese to English)
16:02 – Demo 4: 3D Product Template with Brand Texture Swap
18:40 – Demo 5: Video Extension and Filling in the Middle
20:55 – Demo 6: AI Influencers and Prompting Realistic Emotion
29:31 – What Happens to Adobe Over the Next Five Years
Key Points
- Seedance V2 is the first widely available video model to support true multi-input generation — up to two images, two videos, and an audio file combined in a single prompt.
- Treat Seedance V2 as a video editor, not just a generator: character swap, background swap, text preservation, ad translation, and template population all work from natural-language prompts.
- Seedance rewards highly specific prompts; I pair my own draft with Claude Opus 4.6 to optimize prompts for vision models.
- Strong source reference images remain the single biggest quality lever — the model mimics taste from what you feed it.
- For AI influencers and lip sync, describe muscle movements and emotional transitions rather than simply labeling an emotion like "sad" or "happy."
- Seedance V2 is the current default for editing and generating video, yet other models (Kling 3 for cinematic feel, Enhancer V4 for talking-head realism) still win on specific use cases.
The #1 tool to find startup ideas/trends - https://www.ideabrowser.com
More from this podcast
The Startup Ideas Podcast →