Work with specific model types

mistral.rs handles more than text chat. Each model category has its own request shape and conventions.

• Use vision and video input: image and video on Qwen3-VL and Gemma 4.
• Set up video input: FFmpeg, supported formats, CLI, and HTTP video examples.
• Use image generation models: diffusion models like FLUX.
• Use speech models: Voxtral for STT, Dia for TTS.
• Use embedding models: EmbeddingGemma, Qwen3-Embedding.

Model-family walkthroughs:

• Text model walkthroughs: Qwen3 thinking, SmolLM3, DeepSeek, GLM, GPT-OSS, and MoE notes.
• Vision model walkthroughs: Qwen-VL, Gemma, Llama, Mistral, Phi, Idefics, LLaVA, and MiniCPM-O.

Full model and modality list: supported models reference.