Uploaded June 2026 | Updated November 2026, 5 days ago
► Full written breakdown: theoreticallymedia.com/blog/seedance-audio-first-seedaudio
Seedance isn't cheap — and 2.5 won't be cheaper — so the worst feeling is burning credits on a gen that drops the one line you actually wrote. ByteDance's new SeedAudio 1.0 is the cheap fix: not as a music model (it's no Suno), but as an audio reference that locks your dialogue, performance, and ambiance before you pay for the expensive video pass.
It's basically the animation workflow — audio first, then the visual. I walk through the whole thing: running it on Fal, why dumping every reference into one Omni ref causes context overload (music bleed, repeated lines, mush-mouth), the first-frame fix via Nano Banana Pro, how to timestamp your audio (drop the MP3 into Claude / Gemini / GPT and have it write a timed Seedance prompt), voice references for consistency, long-take editing, and the one prompt rule that keeps it on the rails. Plus the honest caveats — the model's still finicky, and on a simple one-liner the vanilla gen was honestly fine.
Tests: a high-fantasy audio drama, the Sniper Woman (built in Midjourney 8.2 preview), and a new Tues Sinclair & Malloy noir detective scene that gave us some of the best acting we've pulled in one of these.
Also — camera control without Blender: Martini's new step-into-set + camera-motion timeline (keyframes, lens, motion sliders) gets you Blender-style control with no install. Not a sponsored segment — I've done paid Martini work before, but this one's just a feature I liked.
Big credit to Tom Likes Robots for the back-and-forth dialogue experiment that pointed at where this really shines.
🛠 Tools shown:
SeedAudio 1.0 (Fal): fal.ai/models/bytedance/seed-audio-1.0
Midjourney 8.2 (preview): midjourney.com
Flora: refer.flora.ai/theoretically-media
Martini · https://www.martini.film/theoretically-media-june
Kling Omni Director Look:
youtu.be/s40M0IvBEpM
Claude MCP to Blender: youtu.be/6YO6y_2SSro
Hey, why not subscribe to my Newsletter?
theoreticallymedia.com/newsletter
0:00 A New Seed Model — For Your Ears
0:34 What SeedAudio 1.0 Actually Is (Not Music)
1:20 First Look on Fal: A Fantasy Audio Drama
2:41 The Prompt That Flopped
3:05 The Real Trick: Audio as a Seedance Reference
3:13 Sniper Woman + a Mini Midjourney 8.2 Review
4:52 Where It Shines: Back-and-Forth Dialogue
6:16 The Detective Office (Tues & Malloy)
7:15 Fixing the Mess: Context Overload + Nano Banana Pro
8:31 Timestamp Your Audio
9:15 Long Takes & Light Editing
9:57 Voices, Consistency & Settings
11:10 The Animated Scene + a Key Prompt Tip
12:26 Why This Actually Matters
13:04 - The FBI Agent In a Diner Benchmark
13:43 What It Costs on Fal
14:17 Skipping Blender: Camera Control Without the Headache
15:57 Adding Real Camera Motion in Martini
17:40 The Camera-Control Wave Is Coming
18:08 Sound & Vision: Wrap
► Full written breakdown: theoreticallymedia.com/blog/seedance-audio-first-seedaudio
Seedance isn't cheap — and 2.5 won't be cheaper — so the worst feeling is burning credits on a gen that drops the one line you actually wrote. ByteDance's new SeedAudio 1.0 is the cheap fix: not as a music model (it's no Suno), but as an audio reference that locks your dialogue, performance, and ambiance before you pay for the expensive video pass.
It's basically the animation workflow — audio first, then the visual. I walk through the whole thing: running it on Fal, why dumping every reference into one Omni ref causes context overload (music bleed, repeated lines, mush-mouth), the first-frame fix via Nano Banana Pro, how to timestamp your audio (drop the MP3 into Claude / Gemini / GPT and have it write a timed Seedance prompt), voice references for consistency, long-take editing, and the one prompt rule that keeps it on the rails. Plus the honest caveats — the model's still finicky, and on a simple one-liner the vanilla gen was honestly fine.
Tests: a high-fantasy audio drama, the Sniper Woman (built in Midjourney 8.2 preview), and a new Tues Sinclair & Malloy noir detective scene that gave us some of the best acting we've pulled in one of these.
Also — camera control without Blender: Martini's new step-into-set + camera-motion timeline (keyframes, lens, motion sliders) gets you Blender-style control with no install. Not a sponsored segment — I've done paid Martini work before, but this one's just a feature I liked.
Big credit to Tom Likes Robots for the back-and-forth dialogue experiment that pointed at where this really shines.
🛠 Tools shown:
SeedAudio 1.0 (Fal): fal.ai/models/bytedance/seed-audio-1.0
Midjourney 8.2 (preview): midjourney.com
Flora: refer.flora.ai/theoretically-media
Martini · https://www.martini.film/theoretically-media-june
Kling Omni Director Look:
youtu.be/s40M0IvBEpM
Claude MCP to Blender: youtu.be/6YO6y_2SSro
Hey, why not subscribe to my Newsletter?
theoreticallymedia.com/newsletter
0:00 A New Seed Model — For Your Ears
0:34 What SeedAudio 1.0 Actually Is (Not Music)
1:20 First Look on Fal: A Fantasy Audio Drama
2:41 The Prompt That Flopped
3:05 The Real Trick: Audio as a Seedance Reference
3:13 Sniper Woman + a Mini Midjourney 8.2 Review
4:52 Where It Shines: Back-and-Forth Dialogue
6:16 The Detective Office (Tues & Malloy)
7:15 Fixing the Mess: Context Overload + Nano Banana Pro
8:31 Timestamp Your Audio
9:15 Long Takes & Light Editing
9:57 Voices, Consistency & Settings
11:10 The Animated Scene + a Key Prompt Tip
12:26 Why This Actually Matters
13:04 - The FBI Agent In a Diner Benchmark
13:43 What It Costs on Fal
14:17 Skipping Blender: Camera Control Without the Headache
15:57 Adding Real Camera Motion in Martini
17:40 The Camera-Control Wave Is Coming
18:08 Sound & Vision: Wrap










