Four models. One local AI workbench.

What the project does

Qwen 3.8 27B runs locally with Q4_K_M weights and a vision projector. A retained code trial produced a working Python merge function; the saved output and a browser visualization are available in the exhibit. This is recorded work, not an available inference service.

Qwen Image 2.1 research studies explore a 1024-pixel still life and a native 2048-pixel character with transparency. These are retained local evaluation outputs under the Qwen Research License, not an offer of commercially licensed generation. MiniMax H3 uses the local FL2VA pruned INT8 workflow; The Line of Light combines generated shots with authored direction, editing and sound.

The portfolio plays saved audio and explains the workflow. It does not send generation requests to a live model. Runtime throughput and production service availability are not claimed.

Music work uses ACE-Step 1.5, with original lyrics, selected takes and mastered recordings kept as distinct steps. The three accepted English vocal recordings can be played in the Music exhibit.

Selected capabilities

• Qwen 3.8 27B: local Q4_K_M weights and F16 vision projector
• Qwen Image 2.1: local image generation with ComfyUI
• MiniMax H3: local FL2VA INT8 video generation and film production records
• ACE-Step 1.5: original lyrics, selected takes and finished recordings
• Saved outputs on this site; no live inference

Four workflows. Visible results.

Explore the four outputs in the studio

Qwen 3.8 27B · language

Saved Python output merges two sorted lists while preserving duplicates. The retained checks passed.

Qwen Image 2.1 · image

Still-life and character studies from local evaluation. Qwen Research License; commercial model use requires a separate license.

MiniMax H3 · video

H3-generated shots with human direction, edit and sound.

ACE-Step 1.5 · music

Listen to the three original English vocal recordings

Explore the work

• Inside the local AI workbench