Let Gemma write your prompts: MiniMax H3 image-to-video pipeline with 2K upscale
EzkaProductions · reddit · 2026-10-13
The author built a two-stage video generation pipeline with Gemma as a prompt enhancer, solving the struggle of writing good MiniMax H3 prompts:
- Feed a one-line idea ("a woman reads a letter by the window and starts crying") and Gemma expands it into a detailed cinematic prompt, in any input language
- Gemma also generates the video title and filename automatically to keep outputs organized
- Image-to-video with native synced audio; 2K upscale via SeedVR2 preserving original audio and fps
- Two-stage latent upscale with 3/4/5-step options for speed vs quality; optional last-frame control
Tested on an RTX 5090 laptop (24GB VRAM, 64GB RAM); a 15-second clip takes 10 minutes. The free workflow is linked in the video tutorial description.
More from Multimodal
- Midjourney Style Share: Gothic Skeleton Art via --sref 3041751636 — tisch_eins · 2026-10-13
- Ghazals written with Claude and sung by Suno, in Urdu, Devanagari and English — vishalmisra · 2026-10-13
- MiniMax H3 ref2v bug: dance moves reproduced perfectly but shuffled on the timeline — zxfwsn · 2026-10-13
- Grok Bot turns any topic into cinematic explainer videos with ElevenLabs narration — XFreeze · 2026-10-13
- AI-animated kids' pilot The Brass Acorn pairs Pixar-style visuals with an original song — Warblewick · 2026-10-13
- AI-generated image turns Wikipedia's puzzle globe into a real soccer ball — Delahuntagram · 2026-10-13