Generative Models¶

Free models for generating text, images, audio, and code.

Text Generation (LLMs)¶

Run locally via Ollama¶

Ollama makes it easy to run LLMs locally:

ollama run llama3.1      # Meta Llama 3.1 8B
ollama run mistral       # Mistral 7B
ollama run gemma2        # Google Gemma 2 9B
ollama run phi3          # Microsoft Phi-3
ollama run qwen2.5       # Alibaba Qwen 2.5 7B
ollama run codellama     # Code Llama 7B

Run in browser via WebLLM¶

WebLLM runs full LLMs in the browser via WebGPU:

Model	Size	VRAM	Speed	Link
Llama 3.1 8B	4.3 GB (q4)	6 GB	~30 tok/s	webllm.mlc.ai
Mistral 7B	3.8 GB (q4)	5 GB	~35 tok/s	webllm.mlc.ai
Phi-3 Mini	2.2 GB (q4)	3 GB	~45 tok/s	webllm.mlc.ai
Gemma 2 2B	1.4 GB (q4)	2 GB	~50 tok/s	webllm.mlc.ai
TinyLlama 1.1B	0.6 GB (q4)	1 GB	~60 tok/s	webllm.mlc.ai

Image Generation¶

Model	Type	License	VRAM	Link
SDXL	Diffusion	CreativeML Open	8 GB	HF
SD 3 Medium	Diffusion	Stability Community	12 GB	HF
FLUX.1 Schnell	Flow matching	Apache-2.0	12 GB	HF
Kandinsky 3	Diffusion	Apache-2.0	8 GB	HF
PixArt-alpha	Diffusion (fast)	Apache-2.0	8 GB	HF

Code Generation¶

Model	Parameters	License	Best For	Link
CodeLlama 7B	7B	Llama 2	General coding	HF
DeepSeek Coder v2	16B	MIT	Code completion	HF
StarCoder2	3B–15B	BigCode Open	Multi-language code	HF
Qwen2.5 Coder	7B	Apache-2.0	Code + instruction	HF

Audio Generation¶

Model	Task	License	Link
Bark	Text-to-speech (multilingual)	MIT	HF
MusicGen	Text-to-music	CC-BY-NC	HF
AudioCraft	Music + sound effects	CC-BY-NC	github.com/facebookresearch
Coqui TTS	Text-to-speech	MPL-2.0	github.com/coqui-ai
Piper	Fast TTS (CPU)	MIT	github.com/rhasspy/piper

Video Generation¶

Model	Type	License	Link
CogVideoX	Text-to-video	Apache-2.0	HF
AnimateDiff	Image-to-video	Apache-2.0	HF
Stable Video Diffusion	Image-to-video	Stability Community	HF