Cookbook
Recipes for video, image, audio, and world generation.
Pick a model, choose your hardware, and copy the command. Every recipe starts from checked-in YAML, so the settings you see are the settings that run.
Before you pick a device, check weights and quantization for measured checkpoint sizes and the FastH3 precision variants.
FastVideo FastH3
FastH3 Omni v1
Video and audio from a prompt plus ordered image, video, and audio references.
FastH3 4-Step V1
FastH3 V1 on one DGX Spark with offloaded encoders and lazy module loading.
FastH3 8-Step V2
Distilled MiniMax-H3 video generation with synchronized audio.
FastH3 Trim
The smaller FastH3 model for eight-step video generation with synchronized audio.
New
Open-H3-Regenerate-2K
Regenerate a 768P MiniMax-H3 draft at 2560x1440 with a three-step refiner.
MiniMax H3
Generate synchronized video and stereo audio from a structured text prompt with the full MiniMax H3 checkpoint.
Python SDK · CLI
Pika H3 Flex
Pika's post-trained MiniMax-H3: synchronized video and stereo audio from text in six forwards, at 480p or 768p, 5-15 s, in BF16 or as FastVideo's NVFP4 QAD release on Blackwell, with optional FP4 attention.
Python SDK · CLI
FastH3 Omni V1
Run the FastH3 Omni BF16 release on Apple Silicon: compose it with the NVFP4 text-encoder stack, convert the DiT to INT6 with VSA, and decode with the LightVAE encoder and the FastH3 NVFP4 decoder.
Python SDK · CLI · MLX
FastWan2.1 1.3B
A compact, distilled Wan model for text-to-video generation.
Wan2.1 I2V 14B
Turn a reference image into video with Wan2.1.
Wan2.2 A14B
The maintained high-capacity Wan2.2 text-to-video example with CPU offload settings encoded in its checked-in Python source.
Python SDK · CLI · Serving
Wan2.2 TI2V 5B
Use one maintained 5B checkpoint for text-to-video or add an image input to switch the same recipe to image-to-video.
Python SDK · CLI · Serving
FastMetal 1.3B
Run the released FastMetal 1.3B QAD checkpoint through FastVideo's native Apple Silicon MLX path.
Python SDK · CLI · MLX
FastMetal 5B
Run the released Wan2.2 5B FastMetal checkpoint as MLX T2V with MLX DiT denoising and MLX TAEHV decode. The CUDA Wan2.2 TI2V 5B recipe is the image-capable path.
Python SDK · CLI · MLX
FastMetal 14B
Run the released 14B FastMetal QAD checkpoint through the same Apple Silicon MLX entrypoint as the 1.3B release.
Python SDK · CLI · MLX
LTX2 Distilled
Generate video with incremental delivery over WebSocket.
LTX-2 base (1088p)
Base LTX-2 text-to-video at 1088x1920 using FastVideo default sampling for LTX2 base. The example loads a community Diffusers mirror of the base checkpoint; registered aliases include Lightricks/LTX-2 and FastVideo/LTX2-Diffusers.
Python SDK · CLI
Kandinsky 5.0 T2V Lite SFT
Kandinsky 5.0 text-to-video (Lite SFT variant) from the maintained example; alternative Lite/Pro checkpoints are listed in the source.
Python SDK · CLI
Kandinsky 5.0 I2V Pro distilled
Animate an input image with Kandinsky 5.0 I2V Pro (distilled) on a single GPU.
Python SDK · CLI
Kandinsky 6 Pro TI2VA
Generate a five-second, 512x768 video with synchronized audio from text, or set IMAGE_PATH in the maintained example to condition on an image.
Python SDK · CLI
Kandinsky 6 Pro pi-Flow
Generate a five-second video with synchronized audio using the distilled pi-Flow checkpoint in ten inference steps and guidance 1.0.
Python SDK · CLI
Kandinsky 6 Lite TI2VA
Generate a five-second, 512x768 video with synchronized audio from text or an image with the 3.2B Lite checkpoint in fifty inference steps and guidance 5.0.
Python SDK · CLI
Kandinsky 6 Lite pi-Flow
Generate a five-second video with synchronized audio using the 3.2B Lite distilled pi-Flow checkpoint in ten inference steps and guidance 1.0.
Python SDK · CLI
Kandinsky 6 VSR
Upscale an existing clip by 2.25x with the four-step flow-matching VSR checkpoint while preserving its source audio.
Python SDK · CLI
Kandinsky 6 VSR distilled
Upscale an existing clip by 2.25x with the distilled two-step VSR checkpoint while preserving its source audio.
Python SDK · CLI
FLUX.2 Klein 4B
Generate an image in four denoising steps with the distilled FLUX.2 Klein checkpoint.
Python SDK · CLI
FLUX.2 dev
Full FLUX.2 dev text-to-image with embedded guidance and the Mistral3 text encoder, from its maintained example.
Python SDK · CLI
FLUX.1 dev
FLUX.1 dev text-to-image through the Diffusers-backed pipeline. The example defaults to a local weights directory, so this recipe passes the Hugging Face ID explicitly.
Python SDK · CLI
Stable Audio Open 1.0
Six-second text-to-audio generation with Stable Audio Open 1.0 from its maintained example; duration and steps are documented knobs in the source.
Python SDK · CLI
Stable Audio Open Small
The smaller Stable Audio Open variant with its own shorter training window, from its maintained example.
Python SDK · CLI