feat: add Wan2.2 S2V (audio+img-to-video) support (#1925)

Co-authored-by: leejet <leejet714@gmail.com>
This commit is contained in:
George
2026-09-13 23:33:50 +08:00
committed by GitHub
co-authored by leejet
parent 4a7da26b73
commit 0bd72f075a
32 changed files with 1533 additions and 57 deletions
+19
View File
@@ -460,6 +460,11 @@ ArgOptions SDContextParams::get_options() {
"path to standalone LTX audio vae model",
0,
&audio_vae_path},
{"",
"--audio-encoder",
"path to wav2vec2 audio encoder model (Wan2.2 S2V)",
0,
&audio_encoder_path},
{"",
"--taesd",
"path to taesd. Using Tiny AutoEncoder for fast decoding (low quality)",
@@ -898,6 +903,7 @@ std::string SDContextParams::to_string() const {
<< " vae_path: \"" << vae_path << "\",\n"
<< " vae_format: \"" << vae_format << "\",\n"
<< " audio_vae_path: \"" << audio_vae_path << "\",\n"
<< " audio_encoder_path: \"" << audio_encoder_path << "\",\n"
<< " taesd_path: \"" << taesd_path << "\",\n"
<< " esrgan_path: \"" << esrgan_path << "\",\n"
<< " control_net_path: \"" << control_net_path << "\",\n"
@@ -963,6 +969,7 @@ sd_ctx_params_t SDContextParams::to_sd_ctx_params_t(bool taesd_preview) {
sd_ctx_params.embeddings_connectors_path = embeddings_connectors_path.c_str();
sd_ctx_params.vae_path = vae_path.c_str();
sd_ctx_params.audio_vae_path = audio_vae_path.c_str();
sd_ctx_params.audio_encoder_path = audio_encoder_path.c_str();
sd_ctx_params.taesd_path = taesd_path.c_str();
sd_ctx_params.control_net_path = control_net_path.c_str();
sd_ctx_params.ip_adapter_path = ip_adapter_path.c_str();
@@ -1515,6 +1522,14 @@ ArgOptions SDGenerationParams::get_options() {
return 1;
};
auto on_audio_arg = [&](int argc, const char** argv, int index) {
if (++index >= argc) {
return -1;
}
ref_audio_paths.push_back(argv[index]);
return 1;
};
auto on_cache_mode_arg = [&](int argc, const char** argv, int index) {
if (++index >= argc) {
return -1;
@@ -1704,6 +1719,10 @@ ArgOptions SDGenerationParams::get_options() {
"--ref-audio",
"standalone WAV reference for MiniMax-H3 Ref2VA (can be used multiple times)",
on_ref_audio_arg},
{"",
"--audio",
"driving audio track (Wan2.2 S2V; can be used once)",
on_audio_arg},
{"",
"--cache-mode",
"caching method: 'easycache' (DiT), 'ucache' (UNET), 'dbcache'/'taylorseer'/'cache-dit' (DiT block-level), 'spectrum' (UNET/DiT Chebyshev+Taylor forecasting)",