Add Pika H3 Flex LoRA adapters (fl2va + ref2va)
#70
by ComfyUI-Wiki - opened
Adds the two Pika H3 Flex adapters, converted to the ComfyUI layout:
- loras/pika_h3_flex_fl2va_v1_bf16.safetensors: text-to-video, first/last-frame, i2v continuation
- loras/pika_h3_flex_ref2va_v1_bf16.safetensors: reference-to-video
Rank-128 bf16 adapters on MiniMax H3 (416 tensors, 50 DiT blocks + token_refiner), from Pika / Mellis (metadata modified_by: Mellis, Inc.), released under the MiniMax H3 Community License.
Sampling: lcm, cfg 1.0, ManualSigmas 1.0, 0.972973, 0.923077, 0.8, 0.571429, 0.333333, 0.0 (6 steps), adapter strength 1.0. Do not stack a turbo LoRA on top.
Templates: Comfy-Org/workflow_templates video_pika_h3_flex_t2v / _fl2va / _ref2va.
ComfyUI-Wiki changed pull request status to open
ComfyUI-Wiki changed pull request status to merged