MiniMax H3 review: native 4096x1536 image editing in 1 minute, workflow included
-Ellary- · reddit · 2026-09-28
The author tests MiniMax H3 REF with Fizgig ComfyUI nodes: as a true Ref2Img/Img2Img editing model, it natively outputs 4096x1536 panoramas in 1 minute at 18 steps on an RTX 5060 Ti 16GB.\n\nHighlights: strong prompt/spatial understanding, no LLM-processed prompts needed per image, imitates many styles from a reference, solid world knowledge, sharp output with few artifacts, and faster than Qwen 2.1 with 2+ image references. ComfyUI workflows and prompts for every example are linked in the post.
More from Multimodal
- Agentic filmmaking experiment: GPT-6 Astra edits, Grok Imagine renders, Suno v6 scores — TinfoilTricorn · 2026-09-29
- Fizgig 6.6 adds edit-LoRA training for Qwen Image 2.1, learns grading from 40 pairs — shootthesound · 2026-09-29
- Deforum keeps going and going: an endless morphing chair animation — makeitrad1 · 2026-09-29
- GPT Image 2.5 vs open-source Krea 2 Turbo: same-prompt test highlights OpenAI's heavy censorship — Due_Research9042 · 2026-09-29
- Hugging Face release: Yue2-based quiet-storm model for 90s Jazz-Soul vibes — -becausereasons- · 2026-09-29
- One prompt, six shots: multi-shot video generation shows cinematic chops — minchoi · 2026-09-29