SHOWCASE · A RECORDED AGENT WORKFLOW

A portrait in.
A spokesperson out.

It started with one identity portrait. An agent gave her an outfit, a character sheet and a voice, then put a product in her hand and let her speak for it. Five steps, the same face all the way through, and every image and clip on this page is the real output.

STILL HER

First portrait. Final frame. Same face.

Look at the eyes, the auburn bob, the beauty mark. On the left, the portrait the agent started from. On the right, a frame of the video five generations later. One agent, one MCP connection, one character from start to finish.

5 stepsone face, start to finish

Recorded 1 Oct 2026.

The original character portrait: auburn bob, beauty markA frame of the final video after she turns back to camera
The starting portraitThe final video frame

TURN THE SOUND ON

Picture, voice and sound.
In one take.

A night ride through wet neon streets. A line spoken straight down the lens. A cable clicking into a glowing port. Each clip came out of a single generation exactly as you hear it, motion and sound together. Press play.

A rider on a sport motorcycle through a wet, neon-lit back street at night

MiniMax H3 reference5 s · with sound

Macro: a gloved hand plugs a cable into a white network switch, the port lights up

Seedance 2.06 s · with sound

In the rain, she lifts the helmet visor and speaks to camera

FLUX 3 keyframes5 s · with sound

Close-up of a woman at a desk lit by monitors, delivering a line to camera

MiniMax H3 reference5 s · with sound

A masked figure walks out of a green glow carrying a network switch to the bench

Seedance 2.010 s · with sound

A rider stands beside her motorcycle on a lamplit street at night

FLUX 35 s · with sound

HOW THE AGENT DID IT

Five steps. Every prompt, every model.

The agent passed each result to the next step by its job reference, so nothing was downloaded or uploaded again. Step through the run, or copy a prompt and make your own version.

01 / A REAL MCP WORKFLOW

One character. Across generations.

A fresh model call has no memory of the last one. This recorded workflow carries identity forward with prompt_guide instructions and completed job references.

5 steps

five successful jobs

Recorded 1 Oct 2026

Original generated face lock with auburn bob and beauty markFinal MiniMax H3 video frame after the character turns back to camera
01 · Original face lock04 · Final video frame

From the original face lock to the final video frame: the character’s face, eye colour, auburn bob and beauty mark remain recognizable.

Playback of saved outputs, not live generation. To create your own, sign in with Google and create an MCP bearer token.

Face lock → outfit → character sheet + video

The sheet and video both use the outfit’s job_ref. The video includes speech; the music bed is generated separately. No image download or re-upload between calls.

No file inputSaved output
Generated face lock: auburn bob, grey-green eyes and a beauty mark

OBSERVED IN THIS RUN

This first plate establishes the character’s face, hair, eyes and beauty mark for the following generations.

STEP 1 / 5 · 1 CREDIT

Face lock

flux-2-pro

A 3:4 identity plate: forehead to upper chest, arms down, hands out of frame. The 12-point face-lock grammar fixes bone structure, hair, makeup and identity markers on a flat 18% gray field with shadowless lighting.

Prompt grammar: prompt_guide · face_lock

Recorded run · parameter excerpt
{
  "model": "flux-2-pro",
  "params": {
    "image_size": "portrait_4_3"
  }
}

RETURNED JOB_REF

job_NeOlT6PjpchlYwtG1KvV5w
Read & copy the full recorded prompt and arguments
Full recorded run arguments
{
  "model": "flux-2-pro",
  "params": {
    "prompt": "A clean cinema-character-reference 3:4 headshot, framed from forehead to upper chest with the face filling most of the frame. A Northern European woman in her early thirties, lean athletic build, fair skin with a cool neutral undertone and a matte finish. Dark auburn hair in a deep copper-brown, cut to a blunt chin-length bob with a centre parting tucked behind both ears, fine baby hairs at the hairline. Strong angular features: high wide cheekbones, a square clean jawline, a straight narrow nose, thin upper lip with a fuller lower lip. Wide-set grey-green eyes, level outer corners. Straight dark brows, natural thickness, low arch. Bare skin register, no visible makeup, no liner, no lash drama. A small flat beauty mark below the outer corner of the left eye and no other facial markings: no freckles, no piercings, no tattoos, no scars. Both arms down at her sides, hands not visible, nothing in frame but head, neck, shoulders and upper chest. Plain black thin-strap camisole, no jewellery, no logos, no graphics. Body squared to camera, head level, neutral relaxed expression, eyes to camera, lips closed. BACKDROP: a single flat 18% neutral gray field, one uniform value at every pixel, corner to corner, no seam line, no gradient, no hotspot, no vignette, no falloff. It is a flat colour field, NOT a photographed backdrop: no surface, no floor, no wall, no corner, no horizon. LIGHT: completely flat shadowless illumination, one enormous soft frontal source at camera position, matched equal fill from camera-left, camera-right, above and below, so both sides read at exactly the same brightness. No key-and-fill ratio, no modelling, no shadow side, no nose shadow, no under-chin shadow, no rim light, no hair light, no kicker, no specular hotspot. Extremely low contrast, even, milky, catalogue-flat. NO SHADOW OUTSIDE THE SUBJECT: no cast shadow, no contact shadow, no drop shadow, no ambient occlusion where the body meets the background, no darkening behind the shoulders, no halo, no edge darkening. Shading exists only on the subject and stops cleanly at the silhouette. No light bleed, no spill, no glow, no bounce, no reflected colour cast on the background. Skin reads matte and velvety with light natural sheen on the high points only, no oily shine, real fine even pore texture, real peach fuzz at the jaw and hairline, subsurface scattering reading as semi-translucent biology. Skin renders at its true tone, never darkened, never pale, never washed-out, never cool-shifted by the background. Never plastic, never doll-skin, never waxy. Even sharpness edge to edge. No depth-of-field falloff, no bokeh, no background blur, no lens vignette, no distortion, no flare, no bloom, no chromatic aberration, no haze. Photographed on a 50mm prime. Photographed, not generated. No text, no letters, no numbers, no watermarks anywhere in the image.",
    "image_size": "portrait_4_3"
  }
}

Pass the full argument object to the MCP run tool. To reproduce the workflow, use your own returned job references; these references identify the recorded jobs.

1 of 5

Recorded 1 Oct 2026 in one agent session: each step fed the next by its job reference.

Read the full SingularityByte experiment ↗

YOUR TURN

A product. A character.
A still that moves.

A product photo turned into an outdoor lifestyle scene

Put your product anywhere.

One product photo becomes the scenes your launch, your season and your feed are waiting for.

See the before and after →
The auburn-haired character from the recorded workflow

Build a character once.

Start with a face, dress her, then bring her back for the next scene, the next clip, the next ad.

Follow her five steps →

Want more? The gallery on the agents page and the blog are full of real generations, prompts included.