H3 long-video degradation workaround: a second noise-injection refine stage in latent space
xyzdist · reddit · 2026-09-23
Reddit user xyzdist shares a new test tackling HunyuanVideo (H3)'s long-video quality degradation. The approach: a minimal workflow with native motion-context plus a second refinement stage that injects noise and clean conditioning into latent space to stop degradation, then blends the latent back.
The test video combines 10×8s segments using a cheap 10-step base + 4-step refine setup. The refine side degrades notably less, but contrast still worsens over time and background elements occasionally vanish. The author mentions tapered noise and selfLift as possible next steps and links earlier posts with workflow details.
More from Multimodal
- Generating accurate 3D building models from 150 planning PDFs with new Gemini model — fofrAI · 2026-11-11
- Stanford's DyCraft Uses Agentic Decomposition to Cut Physical Errors in Long-Horizon 4D Scene Generation — StanfordUniversity · 2026-11-11
- Midjourney SREF pick: --2523103199 for loose pencil-sketch portraits — tisch_eins · 2026-11-11
- H3 Image Tips: 16MP via ComfyUI Workflow Plus Detailed Prompts Yields Best Results — gatortux · 2026-11-11
- Gemini-generated endless loop of quantum cryostats self-assembling — hapticdata · 2026-11-11
- Dad brings his daughter's crayon drawings to life with AI for $1.75, and it's unsettling — Illustrious_Ad_4292 · 2026-11-11