Filmmaker Shares 5 Hard-Won Lessons From Making a Star Trek vs Star Wars Fan Film With MiniMax H3 in ComfyUI
Glad-Hat-5094 · reddit · 2026-09-11
A Reddit user shares the workflow behind a Star Trek vs Star Wars fan film made with MiniMax H3 inside ComfyUI, a follow-up to an earlier 6-minute TNG fan film. The pipeline: generate starting images, use H3 to produce individual shots, then assemble in Adobe Premiere Pro — building the film from many short generations rather than whole scenes at once.
Key lessons learned:
- Generation length affects dialogue delivery: too short and characters race through lines unnaturally fast; too long and delivery drags. For 10–15 second dialogue shots the author tests several durations (10s, 11s, 12s…) at low resolution first, then commits to a high-resolution render once pacing sounds right.
- Higher resolution genuinely helps: much of the film was generated around 2MP / Full HD. Faces look noticeably less like "AI video." Low resolution is reserved for testing prompts, motion, dialogue and duration; high resolution is for final shots.
- Reference images beat massive prompts: if character placement, set, lighting, camera angle and composition are already correct in the starting image, H3 has far less room to wander. The starting image is treated as the visual authority for the shot.
- Lock the camera explicitly: for shots based on existing Star Trek compositions, the author specifies camera distance, character scale, framing, background and character position, and states the camera must remain completely stationary — otherwise the model tends to slowly push in or recompose.
- The post ends with a note about not mentioning certain characters (text truncated).
More from Multimodal
- Creator makes 2D electro-pop anime music video with just a prompt using MiniMax H3 — Hailuo_AI · 2026-09-11
- Street View to driving footage: GPT Astra fetches images, MiniMax H3 turns them into dashcam video — Hailuo_AI · 2026-09-11
- Single-author ECCV 2026 paper makes rolling shutter correction practical — ducha_aiki · 2026-09-11
- AI digital human covers Japanese classic so realistically viewers can't tell — JourneymanChina · 2026-09-11
- ComfyUI Style Explorer Adds LoRA Preview Catalog and Sharing — neonsparksuk · 2026-09-11
- 4 favorite Midjourney V6.1 --sref style codes, ready to copy — michaelrabone · 2026-09-11