MiniMax H3 Image and FIBO 2 tested in ComfyUI with 14 free workflows
pixaromadesign · reddit · 2026-11-18
A hands-on ComfyUI episode tests two new image models with 14 free workflows (7 each):
- MiniMax H3 Image: text-to-image up to 4K, built-in prompt expander, text-based editing, image combination, and box/mask inpainting.
- FIBO 2 (by Bria AI): trained on long structured JSON prompts; a helper model converts short ideas into readable JSON. Offers Turbo and Dev variants plus editing, combining, and masked edits.
The author benchmarks both against earlier models with identical prompts: Krea 2, MOSS and P-H3 LoRA for T2I; P-H3 LoRA, FLUX 3 Klein and Qwen Image 2.1 for editing. Exact prompts ship next to the workflows.
More from Multimodal
- SenseNova-Loop-9B, a looped-transformer multimodal model, trends on Hugging Face — sensenova · 2026-11-18
- Kling Image 4.0 arrives with 'Tone' feature for consistent style and palette — gizakdag · 2026-11-18
- Higgsfield's new AI film: 'The hardest part is making it not look like AI' — CurieuxExplorer · 2026-11-18
- ComfyUI makes the MiniMax H3 video VAE up to 2.2x faster on Nvidia GPUs — NerdyRodent · 2026-11-17
- LMMs-Lab: Multimodality Is a Precondition for Next-Level Intelligence — liuziwei7 · 2026-11-17
- Kling Image 4.0 real-world tests: pet shots to wedding ad visuals look real — LudovicCreator · 2026-11-17