FastH3 OmniRef in ComfyUI: Reference to Video Workflow

FastH3 OmniRef; eight steps, Euler/simple, sigma shifts 12/3, VSA 10%. 3:4 portrait at 1 MP; six-second duration control, 24 fps. Seed 402832748317658. Saved output: 896 × 1184, 158 frames, 6.583 seconds. Load Audio was added to the working copy and connected to the standalone audio-reference socket with the audio VAE. The downloadable base graph leaves this socket unconnected.

subject_definitions:
<Subject 1> is the physical humanoid robot in <Picture 1>: an ivory ceramic face assembled from separate panels with visible screws, a scratched brushed-metal cranial plate, blue optical eyes, dark articulated metallic lips and jaw, exposed polished neck pistons, cables and shoulder bearings. <Picture 1> supplies robot identity, materials and workshop appearance, not a timed first frame. <Audio 1> is the natural recorded male narration reference for the robot's vocal timbre, relaxed cadence and clear English pronunciation.

summary:
[reference generation] A photorealistic close-up of the same physical robot speaking one complete sentence directly to camera in its softly blurred workshop. The robot's articulated mouth moves with the speech, with restrained head and eye movement. Use <Audio 1> as a voice and delivery reference while generating the performance and soundtrack.

retention_analysis:
<Picture 1>: fully_preserved mechanical identity, ivory facial plates, brushed metal, blue optical eyes, visible fasteners, exposed neck actuators and shoulders; partially_preserved workshop background and close portrait framing.
<Audio 1>: attribute_transfer natural male vocal timbre, unhurried reading cadence and intelligible pronunciation; the reference is audio conditioning rather than an externally dubbed soundtrack.

detailed_description:
Photoreal live-action footage of a physically manufactured robot, with believable optical reflections, mechanical tolerances and weight. Soft workshop daylight reveals small scratches, ceramic edges and brushed-metal grain without stylization.
[Shot 1] The continuous six-second shot opens on <Subject 1>, centered from upper chest to above the cranial plate, facing the camera at eye level. Its ivory ceramic cheeks and forehead retain their panel seams and recessed screws, the exposed neck cylinders remain aligned with the chest, and the blue optical eyes maintain a calm direct gaze. The same softly defocused workshop from <Picture 1> remains behind it, with bright windows on the left and neutral industrial equipment in the distance. The camera stays locked, with no zoom or cut.
The robot begins speaking promptly in the natural male reading voice referenced by <Audio 1>. (S1) <d>[English] They lived with their mother in a sandbank, underneath the root of a very big fir-tree.</d> Deliver the complete sentence clearly at the reference's relaxed but fluent pace, preserving the short pause after sandbank and finishing before the shot ends. The lower jaw pivots subtly on its mechanical joint, and the dark segmented lips articulate small, precise speech shapes while the ceramic cheek plates retain their rigid form. Keep all teeth and mouth components spatially consistent as the mouth opens and closes. A slight head inclination accompanies the middle of the sentence; the neck pistons move only enough to support that small gesture. One brief eyelid movement and tiny eye adjustment add an attentive physical presence. After the final word, the jaw settles and the robot holds its gaze for the remaining moment. Keep the robot clearly mechanical throughout, with no human skin, beard, morphing face, extra subject, subtitle or on-screen text.

overall_soundscape:
One clear natural male speaking voice (S1), conditioned by <Audio 1>, close and dry, with the full quoted sentence audible above very quiet workshop room tone. No electronic voice filter, exaggerated robotic effect, competing speech, music or loud mechanical noise.

non_diegetic_music:
N/A