Fabio
c36510c21e
Some checks failed
Linux Build / Linux cpu (push) Has been cancelled
Linux Build / Linux cpu-clang (push) Has been cancelled
Linux Build / Linux vulkan (push) Has been cancelled
mac-build / macOS CPU (push) Has been cancelled
Nix Build / Nix (cpu) (push) Has been cancelled
Nix Build / Nix (vulkan) (push) Has been cancelled
windows-build / Windows CPU (push) Has been cancelled
42 lines
1.4 KiB
C++
42 lines
1.4 KiB
C++
#pragma once
|
|
|
|
#include "engine/framework/core/execution_context.h"
|
|
#include "engine/framework/assets/tensor_source.h"
|
|
#include <filesystem>
|
|
#include <functional>
|
|
#include <memory>
|
|
#include <vector>
|
|
|
|
namespace engine::models::latentsync {
|
|
|
|
struct LatentSyncPipelineSources {
|
|
std::shared_ptr<const assets::TensorSource> unet, vae, whisper, detector, landmarks;
|
|
std::filesystem::path detector_config, landmark_config;
|
|
};
|
|
|
|
struct LatentSyncGenerationOptions {
|
|
int64_t steps = 20;
|
|
int64_t cache_interval = 1;
|
|
float guidance_scale = 1.5F;
|
|
int64_t frames_per_chunk = 16;
|
|
double fps = 25;
|
|
uint64_t seed = 1247;
|
|
};
|
|
|
|
class LatentSyncPipeline {
|
|
public:
|
|
LatentSyncPipeline(core::ExecutionContext & execution, const LatentSyncPipelineSources & sources,
|
|
assets::TensorStorageType weight_type, bool vae_cudnn_f16 = false,
|
|
bool unet_cudnn_f16 = false, bool mem_saver = false);
|
|
~LatentSyncPipeline();
|
|
// Frames and callback pixels are RGB interleaved bytes; mask is RGB planar 512x512 in [0,1].
|
|
void run(const std::vector<std::vector<uint8_t>> & frames, int height, int width,
|
|
const std::vector<float> & audio_16khz, const std::vector<float> & mask,
|
|
const LatentSyncGenerationOptions & options,
|
|
const std::function<void(size_t, const std::vector<uint8_t> &)> & emit);
|
|
private:
|
|
struct Impl;
|
|
std::unique_ptr<Impl> impl_;
|
|
};
|
|
|
|
} // namespace engine::models::latentsync
|