MiniMax H3 · LoRA · three controlled runs

We tested a Sydney Sweeney H3 video LoRA.

One reviewed reference image. Three five-second prompts. The same adapter, resolution, safety setting, and inference route in every run.

Every video below is AI-generated and unofficial. Sydney Sweeney did not create, perform in, endorse, or supply a voice for these tests.

Frozen input

AI-generated result

Model route

H3 Spicy Image to Video

Frozen settings

5 seconds · 768P · LoRA 1.0

Provider cost

$0.375 each · $1.125 total

fal list pricing checked September 28, 2026

The outputs

Three motions from one image

We kept the input fixed to make the motion requests easier to compare. These are selected editorial tests, not a blind benchmark or proof of how the adapter performs across identities, scenes, or longer videos.

01 · AI-generated

Dialogue and lip sync

The face stays recognizable while the model adds blinking, a smile, speech, and quiet room audio. Mouth motion creates the largest visible identity change of the three tests.

Exact prompt and proof

Sydney Sweeney. Live-action handheld mirror selfie. She looks into the mirror, gives a small natural smile, and says in a friendly conversational voice: <d>[English] This is a five second character video test.</d> Natural blinking and subtle posture shifts. Quiet room tone, no music, no subtitles.

Request:: 01a0ebd7-b5b5-78d2-8aa0-b7057b6dff02
SHA-256:: f1cefd5a735c1428818554159106ac0427ca7fa8978ffc1e6f3a61bf1e39ae6b

02 · AI-generated

One-hand wave

The requested wave arrives clearly and returns to a relaxed smile. The moving hand is soft at peak motion, but the face and background remain stable.

Exact prompt and proof

Sydney Sweeney. Live-action mirror selfie. She lowers the phone slightly, looks toward her reflection, gives one relaxed wave, then smiles. Natural breathing, subtle hair motion, and stable anatomy. Quiet room tone, soft fabric movement, no dialogue, no music, no subtitles.

Request:: 01a0ebd7-b5b4-7661-9608-b4baa2766353
SHA-256:: a6366853121db06300181e32d23173a6db0ff7e2446a38df041e99c083fd6452

03 · AI-generated

Slow push-in and hair tuck

This is the cleanest identity-preservation example. The crop tightens, the free hand moves into frame, and the expression changes without a scene reset.

Exact prompt and proof

Sydney Sweeney. Live-action mirror selfie with a slow handheld push-in. She shifts her weight gently, turns one shoulder toward the mirror, tucks one loose strand of hair behind her ear, and settles into a restrained smile. Realistic micro-movements, stable framing, quiet room ambience, no dialogue, no music, no subtitles.

Request:: 01a0ebd7-b5ae-77d3-9d33-dadd872dbb54
SHA-256:: 75aec9b93504d9430ce80555b94a7e976eafc4bf019e95823e6d2e7e8d7417cc

What we learned

Good close-up stability, with familiar video-model tradeoffs.

Identity: all three results keep the subject recognizable through a restrained five-second motion.

Dialogue: lip sync works, but the wider mouth shapes produce more visible facial drift than the silent clips.

Hands: the wave reads clearly, although motion blur hides some finger detail at peak movement.

Prompt adherence: the hair tuck and push-in are the cleanest match. The wave prompt invents a phone-lowering action that is less meaningful with this crop.

Scope: the test uses one flattering close-up. It says nothing yet about full-body motion, complex interaction, cuts, or reference-to-video with two characters.

Newer H3 templates to try with your character

These saved H3 Max templates add selfie expressions, restaurant gestures, outdoor fashion and seated motion. They use separate characters and inputs from the three LoRA tests above. Choose a scene and inspect its preview before using it.

Ponytail Smile and Pout

A close portrait moves from a smile to a playful pout, with a fixed camera and a ponytail gesture.

The expression sequence is more pronounced than a still portrait; hands enter and leave the crop.

Dinner Table Hand Gestures and Smile

A seated restaurant portrait adds small hand gestures, a head tilt and a smile around a table of shared dishes.

Hand movement varies from the saved motion description; table details can shift between frames.

Golden-Hour Golf Course Reveal

A golf-course outfit clip follows a hair adjustment and a turn toward a golf bag in warm evening light.

The glove and finger shapes vary during the hair adjustment; inspect moving props.

Subway Seat Turn and Smile

A seated subway portrait shifts posture and turns toward the camera before smiling.

Background signage is garbled. Use this for pose and expression, not readable text.

Watch all 12 H3 template examples →

How the test was run

Each request used the same square reference, the same converted H3 adapter at scale 1.0, 768P output, a five-second duration, fast prompt expansion, and fal's safety checker. Only the motion prompt changed.

H3 generated the dialogue clip's sound directly with the video. We did not provide or clone Sydney Sweeney's voice. The two silent prompts requested room ambience and no music.

Before these editorial runs, Remix.Camera production telemetry contained zero customer Sydney H3 jobs. These three direct controlled tests are construction and visual-review evidence—not customer adoption evidence.

Use real-person likenesses responsibly

Do not present synthetic footage as authentic, imply participation or endorsement, or use someone's likeness commercially without the rights and consent your use requires. Label AI-generated media clearly and keep source, model, prompt, and output records.

Make a video with your character See more H3 examples