Pika H3 Flex: FL2VA
Animate between two still frames with Pika H3 Flex: the first image opens the shot and the second closes it, while the prompt describes the path between them. The two frames are locked to the start and the end of the clip, the adapter runs at strength 1.0 on the int8 MiniMax H3 FL2VA base with the lcm sampler and Pika's six-step sigma list, and the output size is taken from the first frame and snapped to a multiple of 32 automatically, so portrait sources stay portrait.
Minimum ComfyUI version: 0.44.0
Models and weights
- minimax_h3_audio_vae_fp32.safetensors ·
models/vae/ - minimax_h3_fl2va_pruned_int8_convrot.safetensors ·
models/diffusion_models/ - minimax_h3_video_vae_int8_convrot.safetensors ·
models/vae/ - pika_h3_flex_fl2va_v1_bf16.safetensors ·
models/loras/ - qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors ·
models/text_encoders/
ComfyUI nodes
- Basic Guider
- CLIPLoader
- Create Video
- Float
- Get Image Size
- Int
- KSamplerSelect
- LoadImage
- LoraLoaderModelOnly
- ManualSigmas
- Math Expression
- MiniMax H3 Image to Video
- RandomNoise
- SamplerCustomAdvanced
- Save Video
- UNETLoader
- VAE Decode Audio
- VAEDecode
- VAELoader
ORIGINAL REPOSITORY
Comfy-Org/workflow_templates ↗GitHub activity
- Stars
- 1,445
- Forks
- 270
- Open issues and PRs
- 98
- Repository size
- ~4,915,364 KB
- Default branch
- main
- Created
- Feb 28, 2025
- Repository updated
- Nov 5, 2026
- Last push
- Nov 5, 2026
- Language
- TypeScript
- License
- MIT
Repository-wide figures from GitHub, checked Nov 5, 2026. Stars are endorsements, not a Mega AI Hub quality rating. Repository size is GitHub’s estimate; a ZIP download may differ.
Resource information
- Deployment
- Local
- Publisher
- Comfy Org
- License
- MIT
- Last record update
- Nov 5, 2026