Text to Video
Video generation pipelines built on LTX 2.5 and MiniMax H3 Turbo GGUF quants.
Setup
Video models use a frame-aware latent that carries a temporal dimension. The sampler steps over the full sequence rather than a single frame.
Load the diffusion model and text encoder with the GGUF loader nodes from comfyui-gguf-reboot. The loader nodes give the best performance.
Models and formats
| Model | Repository | Recommended formats |
|---|---|---|
| LTX 2.5 | molbal/LTX-2.5-GGUF | Q8_0, Q8_CR, Q4_K_M |
| MiniMax H3 Turbo | molbal/MiniMax-H3-Turbo-GGUF | Q8_0, FP8, Q4_K_M |
Loading instructions
- Download the GGUF files from Hugging Face.
- Load the diffusion model and text encoder with the GGUF loader nodes.
- Use a frame-aware latent with the desired frame count and resolution.
- Configure the sampler for the model's step count and schedule.
- Decode frames with the VAE and assemble them into a video.
VRAM with video
Video latents multiply memory per frame. On 16GB cards use Q8_0 or the U16G mixed build. See Quant Formats.
Showcase
Video examples are on molbal's Civitai profile.