Self-Trained Video Pipeline: 8 Minutes to Generate 15s of 2K Footage
Kyrannio · x · 2026-09-05
- Indie developer Kyrannio details his self-trained video generation pipeline: built on h3 with his own VAE plus several denoising/decoder components rather than an off-the-shelf model.
- Outputs are decent and relatively fast, but generating 15 seconds at 2K still takes about 8 minutes — a realistic benchmark for personally trained video models.
Related event: Indie dev's self-trained video pipeline renders 2K 15s clip in 8 minutes(2 posts)→
More from Multimodal
- 8 measured lessons from building a 2h manhwa recap in ComfyUI, plus 3 unsolved problems — Sensitive-Wealth5801 · 2026-09-05
- ChatGPT-6 Astra builds a detailed stylized Japanese corner shop in Blender from one prompt — chrisfirst · 2026-09-05
- GPT-6 Astra one-shots a 3D game in 45 minutes, with image gen as the graphics trick — yacineMTB · 2026-09-05
- MiniMax H3 long-video coherence caps out around 20 seconds in community testing — Tight_Organization54 · 2026-09-05
- AI-generated Chopin-style waltz praised as one of the most inspired AI music pieces yet — PTenigma · 2026-09-05
- Researcher Recreates Washington Square Arch in 3D with Astra, 100x Better Than Six Months Ago — OfirPress · 2026-09-05