Narrative AI video on a single RTX 3090: 720p Wan2GP workflow with 2x2 storyboard references
Inevitable_Emu2722 · reddit · 2026-10-14
A Redditor produced The Prompting Zone, a black-and-white Twilight Zone parody (episode: The Last Human), using Wan2GP with MiniMax H3 Ref2VA Pruned 20B on a single RTX 3090 24GB at 720p, 20 steps — roughly 30 minutes per 5-10 second clip.
Workflow: write a full script with dialogue, split into short scenes, generate a 2×2 grid reference image of four camera compositions per scene as a rough storyboard, then feed it with prompts describing camera moves, actions, and exact dialogue.
Lessons learned:
- The model sometimes misreads the 2×2 grid as a single four-panel video instead of sequential shots
- Character consistency is the biggest headache — faces drift between generations, requiring repeated descriptions of characters, clothing, and environments in nearly every prompt
- Readable on-screen UI text is unreliable; portraits and simple icons work much better
- Black-and-white styling helps consistency but doesn't solve continuity
The author plans to eventually automate the pipeline from script to storyboard, video prompts, and final assembly.
Related event: Creator builds narrative AI short film on a single RTX 3090 with Wan2GP(2 posts)→
More from Multimodal
- Stunning Midjourney Generations: Cosmic Light Tunnels in Warm Tones — ciguleva · 2026-10-14
- Redditor combines H3, UVR and YuE2-3B into a full AI music video pipeline — Delicious-Farmer-234 · 2026-10-14
- AI-generated NERDVANA tee mashes Nirvana's smiley with a dead Finder icon — Delahuntagram · 2026-10-14
- Alibaba's Representation Guidance: plain pixel-diffusion Transformers beat bigger models — bdsqlsz · 2026-10-14
- The Prompting Zone Debuts Episode 01: The Last Human — Inevitable_Emu2722 · 2026-10-14
- NVIDIA's PhyProbe scores physical consistency of AI-generated videos, beating prior evaluators — nvidia · 2026-10-14