Showcase

Showcase (gfx1151)

Clips generated on a Strix Halo iGPU with this HIP port. Posters and MP4s live in the repo under assets/showcase/.

The titled project ident at the top of the README is an ffmpeg overlay of a 864×480 run (--steps 20 --layers 50 --reuse 1). The ident draft below is untitled 512×288 model output.

Clip	Kind	Knobs	File
Fox	T2VA	512² · 22f · --steps 20 --layers 50 --reuse 1	t2va-fox.mp4
Linux 35 tribute	T2VA, untitled	864×480 · 56f · --layers 50 --reuse 1 · --seed 1991	linux35.mp4
Project ident draft	T2VA, untitled	512×288 · 56f · --layers 45 --reuse 2 · --seed 1991	h3-hip-ident-draft-raw.mp4
AMD developer community	Ref2VA, untitled	864×480 · 56f · --layers 50 --reuse 1 · --seed 2026	amd-developer-community-raw.mp4

--seconds 2.33 is 56 frames at 24 fps.

H3 does not render readable captions. Titles such as Linux 35 or h3-hip.c on some clips were added with ffmpeg after generate.

Reproduce

Full prompts are in the README. Short form:

MODEL=/path/to/MiniMax-H3

# Fox
./h3 -d "$MODEL" -p "A red fox walks through fresh snow." \
  --width 512 --height 512 --frames 22 --steps 20 --layers 50 --reuse 1 \
  -o assets/showcase/t2va-fox.mp4

# Linux 35 (untitled)
./h3 -d "$MODEL" --width 864 --height 480 --seconds 2.33 \
  --steps 20 --layers 50 --reuse 1 --seed 1991 \
  -o assets/showcase/linux35.mp4
  # prompt: penguin on a snowy ridge, amber terminal, teal/amber grade (see README)

# Ident draft (untitled)
./h3 -d "$MODEL" --width 512 --height 288 --frames 56 \
  --steps 20 --layers 45 --reuse 2 --seed 1991 \
  -o assets/showcase/h3-hip-ident-draft-raw.mp4
  # prompt: cinematic HIP/GPU ident (see README)

# AMD developer community (untitled Ref2VA)
./h3 -d "$MODEL" \
  --ref-image assets/showcase/refs/helios-shirt.png \
  --ref-image assets/showcase/refs/run-open-shirt.png \
  --ref-image-size max \
  --width 864 --height 480 --seconds 2.33 \
  --steps 20 --layers 50 --reuse 1 --seed 2026 \
  -o assets/showcase/amd-developer-community-raw.mp4

Ref2VA needs MiniMax-H3/Ref2VA/. On this HIP path, encoding each still takes on the order of five minutes (video VAE encoder conv).