One photo in a suit goes in.
A dark warlord comes out.
The exact Seedance 2 image-to-video prompt behind the Dark Ascension reel — plus the start/end keyframe trick that locks your face at the payoff instead of letting it drift into a stranger. Watch it, copy it, drop your own photo in. No email, no gatekeeping.
Seedance 2 · 9:16 · one continuous shot · made with Jeremy Knox AI
[SCENE] A crumbling black-stone temple interior — broken pillars, sparse red torchlight, falling ash, ember glow, deep shadow, volumetric crimson god-rays cutting the haze. Same architectural bones throughout, one unbroken space. [SUBJECT] A single person, TRUE accurate likeness to the reference photos, beginning in a fitted black business suit, seen from a three-quarter REAR angle with their face already partially visible in lost-profile. Exactly ONE figure, ONE head, ONE face in frame at all times. [ACTION] They stand motionless, three-quarter to camera; the black stone floor cracks open in crimson fissures beneath them; they levitate ~2ft as crimson energy spirals tight and coiled close to the body, and the black suit DISSOLVES outward into billowing dark robes like ink blooming in water (abstract dissolve, robes stay attached to the body). The head never spins, never turns from back to front, and never pulls a hood across the face — the single face stays continuously present the whole time; the reveal to full-front is achieved ONLY by the CAMERA orbiting around them, not by them turning. Ambient crimson electrical arcs crackle around the hands, body and the air as environmental discharge — never emitted from fingertips as a power; at the climax a hard speed-ramp into slow-mo lands on the face as the eyes ignite pure red and a single straight red plasma saber SNAP-IGNITES raised in the right hand, the blade's first flash under-lighting the whole face; dark stone spires rise as the camera cranes back. No spoken words, no on-screen text. [CAMERA] Arri Alexa Mini, 35mm then 28mm then 50mm then 24mm. ONE continuous take — low ground-level push-in from three-quarter behind; a single unbroken CAMERA ORBIT that travels around them from low-behind to chest-level full-front (the camera moves, the subject does not turn), rising to a HIGH OVER-SHOULDER three-quarter angle (never true overhead); a dive down the crimson energy column with a hard speed-ramp into deep slow-mo onto an extreme face close-up; a fast crane pull-back and up. [STYLE] Dark epic cinematic, high-contrast crimson-on-black, volumetric haze, single-source plasma under-lighting at the ignite with torch/ambient fill so the face stays readable, photoreal, film grain. [AUDIO — SFX only, no dialogue] sub-bass drone; stone-crack rip; rising crimson-energy hum; levitation whoosh; ambient crimson-lightning crackle; robe billow flutter; ~0.7s ducked near-silence right before the climax; saber-ignite plasma SNAP + light-burst as the loudest hit; eye-ignite tonal sting; choir slam; spires-rising grind; ember-hiss tail. [SETTINGS] 9:16, 1080x1920, 720p source upscaled to 1080, ~12s, 24fps. RULES: ONE continuous orbiting shot, follow logic, NO random cuts. Exactly ONE character, ONE head, ONE face — held continuously, never duplicated, never morphed; the face is revealed by camera orbit only, the head never turns on the neck. The saber hilt is a generic hilt in the right hand; the blade is a single straight red plasma beam and is the light source at ignition; the robes are generic dark robes attached to the body; the ink-dissolve suit-to-robe and the rise create the motion, never a static held stance; crimson arcs are ambient environmental discharge, not a fingertip power. NEGATIVE PROMPTS — exclude: a second face, a second head, two-headed figure, a face on the back of the head, duplicate or morphing face, head spinning 180 degrees, extra people, second figure, on-screen text, watermark, branded logos, helmet, any branded-franchise iconography or insignia, melted hands, extra fingers or limbs, floating disconnected props.
#seedance2 #aivideo #imagetovideo #aifilmmaking #promptengineeringTwo keyframes. Seedance fills the rest.
You don't hope one render morphs a businessman into a warlord. You give Seedance a start frame and an end frame, and it generates the entire transformation between them. The end frame locks the payoff — composition, lighting, and your likeness — so the face never drifts into someone else at the exact moment people are watching it.

START FRAME — suit, ¾ rear

END FRAME — the locked payoff
- 01Generate two keyframes from your real photo (any identity-preserving image model): a start in your suit, and an end as the exact warlord shot you want to land on. Frames are cheap — iterate them until both look like you.
- 02Feed both into Seedance 2 image-to-video as
startandendframes with the prompt above. 9:16, 720p, 12s. - 03Upscale the keeper to 1080. That's the whole reel — one continuous shot, your face locked at the climax.
My first version grew a second face on the back of his head mid-turn — pure nightmare fuel. The cause: a start frame facing fully away and an end frame facing fully forward. That's a 180° flip, and the model "solves" it by hallucinating a face onto the back of the skull as it rotates.
The fix is in the prompt above, and it's the part most people miss: shoot the start frame at a three-quarter rear (a sliver of the face already visible), then make the reveal camera-driven — "the camera orbits around them, the head never turns." One continuous face for the model to track, no morph. If you ever see a two-headed frame, that's why — and that line is the cure.
This is one prompt. I teach the whole pipeline.
How I run AI as an operating system — from a single prompt to an unattended fleet that ships content, code, and trades. 350+ lessons in the Academy.
Explore the Academy