MiniMax H3 Hybrid Model: Merging First Frame and Reference Images
Tokey_TheBear · reddit · 2026-08-17
A user shared a hybrid model approach for MiniMax H3 (merging FL2VA and REF2VA) to solve the limitation where official models cannot simultaneously maintain high-quality first-frame locking and use extra reference images.
Key Solution:
- Uses fl2va as the base for quality while incorporating transformer blocks from ref2va to enable reference conditioning.
- The b20-49 variant is preferred for high reference faithfulness (e.g., glyphs, specific stills), while b30-49 offers slightly higher raw quality.
Use Cases:
- Pattern A (First frame + Glyph Ref): Locks the opening scene location while using a second image (e.g., a logo) purely as a style/symbol reference without turning it into a video frame.
- Pattern B (Timed Hard Cuts): Enables precise hard cuts to different still images at 0s, 3s, and 6s within a single clip, overcoming the stock FL2VA limitation of only supporting start and end frames.
Implementation requires specific prompt formatting (e.g., fullypreserved) and ComfyUI wiring using the Reference-to-Video workflow.
Related event: MiniMax H3 Gains Hybrid Mode with Per-Segment Generation(3 posts)→
More from Multimodal
- USTC paper uses speculative sampling to cut latency in interactive video generation — ustc · 2026-10-16
- 800,000 Gaussian splats run in WebGPU; Casado sees near-infinite scaling — generativist · 2026-10-16
- Nano Banana 2.1 image generation now live in Perplexity Computer for all users — ChrisUniverse · 2026-10-16
- Demo shows precise camera-path control in 3D scenes from a single input image — XRarchitect · 2026-10-16
- Train Your Own LTX IC-LoRA With Grok Doing the Heavy Lifting — PurzBeats · 2026-10-16
- Hugging Face launches Open TTS Leaderboard: objective metrics to rank 8K+ open TTS models — dl_weekly · 2026-10-16