No training
Multi-reference inference — no LoRA, no fine-tuning, no compute wait.
Product · Soul
Soul is the fastest path to identity-consistent generation. Save 2–8 photos of any character — yourself, a brand spokesperson, a fictional persona — and place them in any scene with one prompt.
Capabilities
Multi-reference inference — no LoRA, no fine-tuning, no compute wait.
Save once, generate forever. Characters persist across sessions.
Face, body proportions, signature features stay consistent.
Works with Nano Banana 2 (image), Veo 3.1, Kling, Sora 2 (video).
Layer cinematic camera moves on every Cast generation.
Keep personal characters private. Share community casts publicly.
Samples
Every clip below was rendered through Prototipal — no curation, no cherry-picked external content.
Related Apps
FAQ
Multi-reference inference is faster (no training step), cheaper (no compute charges), and matches LoRA quality for most use cases. We may add LoRA training later for power users.
4-6 from different angles works best — front, three-quarter left/right, profile, smiling and neutral. Clean backgrounds preserve identity better.
Yes — toggle the public flag on any character. Future versions will surface a /soul-community feed of shared characters.
Image cast: 2 credits · Video cast: from 4.5 credits/sec
Open the Soul library