MiniMax H3 Open Weights
Official site ↗
Free — you supply the GPU
+ Strongest self-hostable video model right now
− Heavy hardware requirements
MODEL TOOL
Video generation：Budget
Text/image-to-video — ranked against the Artificial Analysis Video Arena.
1 · What are you doing?
2 · Budget tier
Recommended for · Video generation · Budget
Open weights you can self-host — free iteration for drafts and bulk clips.
Official site ↗
Wan2.2（Apache 2.0）
Official site ↗
Open weights — free
+ Last-gen flagship weights with mature ComfyUI workflows; some Wan 2.7 models are open too
− Well behind Wan 3.0 (the line went closed-API from 2.5 on)
LTX-2.5 Fast / Pro
Official site ↗
Open weights — free
+ Extremely fast — near-real-time drafts
− Preview-grade quality only
Video generationMiniMax H3 Open Weights / Wan2.2（Apache 2.0） / LTX-2.5 Fast / ProImage generationFLUX.2 dev / Z-Image Turbo / Qwen-ImageVoice & TTSIndexTTS-2.5 / CosyVoice 2 / KokoroAI avatarHunyuanVideo-Avatar / EchoMimic3D generationTRELLIS.2 / Hunyuan3D 2.1 / Pixal3DMusic generationMiniMax Music 3.0 / Stable Audio 3 Medium / ACE-Step 1.5
How to read this
Three tiers: flagship for quality-critical work, balanced for daily production, budget for bulk and experiments. Names track the current mainstream options — the landscape moves fast, so treat the tier logic (not the exact model list) as the durable part.
Rankings and prices checked — leaderboards move weekly; always confirm pricing on the official site before committing.
Sources: Artificial Analysis · LMArena · Crafiq 3D · Taxon3D