Video Generation Test: ref2vid Loses Character Facial Details vs img2vid
Shadow-Games-1909 · reddit · 2026-08-10
A creator shared hands-on experience with a video generation model (likely H3). When generating long continuous scenes, the ref2vid mode handled character positioning better and reduced background mismatches compared to img2vid using first and last frames.
However, the author noted a significant quality drop in ref2vid, particularly the severe loss of character facial details. Even with a 4K upscaler, the visual quality falls far short of the original static image. The creator questions whether reference-based generation inevitably sacrifices this much image quality.
Related event: MiniMax H3 Ref2Vid Quality Issues Spark Debate(4 posts)→
More from Multimodal
- Kling 4.0 fully available, shows markedly better motion coherence in fast-action scenes — alifcoder · 2026-10-16
- MJev-Spatial: Single-Pass Classification Beats Autoregressive VLMs on Spatial Reasoning — mindlab-research · 2026-10-16
- H-DDCM: Hadamard Codebooks Make Diffusion-Based Image and Video Compression Scalable — Technion · 2026-10-16
- ComfyUI v0.41.0 ships with Flux3 image model and MiniMax-H3 LightVAE support — Fcking_Chuck · 2026-10-16
- AI-generated short film 'The Dragon Rider' glides cinematically above the clouds — jijo6686 · 2026-10-16
- Realistic candid Midjourney portrait prompt: natural skin, lived-in bedroom light — tisch_eins · 2026-10-16