upstream-archive/stable-diffusion.cpp
1
0
Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-07-29 22:30:43 -05:00
Code Issues Packages Projects Releases Wiki Activity

feat: add hunyuan video 1.5 support (#1795)

Browse Source
This commit is contained in:
leejet
2026-07-18 22:30:10 +08:00
committed by GitHub
parent b290693977
commit ea4e566ccf
19 changed files with 1970 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files

71
src/name_conversion.cpp
View File

@@ -664,6 +664,72 @@ std::string convert_diffusers_dit_to_original_flux(std::string name) {
return name;
}
std::string convert_hunyuan_video_to_original_flux(std::string name) {
int num_layers = 54;
int num_single_layers = 0;
static std::unordered_map<std::string, std::string> hy_name_map;
if (hy_name_map.empty()) {
// --- double transformer blocks ---
for (int i = 0; i < num_layers; ++i) {
std::string block_prefix = "double_blocks." + std::to_string(i) + ".";
std::string dst_prefix = "double_blocks." + std::to_string(i) + ".";
hy_name_map[block_prefix + "img_mod.linear"] = dst_prefix + "img_mod.lin";
hy_name_map[block_prefix + "txt_mod.linear"] = dst_prefix + "txt_mod.lin";
// attn
hy_name_map[block_prefix + "img_attn_qkv"] = dst_prefix + "img_attn.qkv";
hy_name_map[block_prefix + "txt_attn_qkv"] = dst_prefix + "txt_attn.qkv";
// norm
hy_name_map[block_prefix + "img_attn_q_norm.weight"] = dst_prefix + "img_attn.norm.query_norm.scale";
hy_name_map[block_prefix + "img_attn_k_norm.weight"] = dst_prefix + "img_attn.norm.key_norm.scale";
hy_name_map[block_prefix + "txt_attn_q_norm.weight"] = dst_prefix + "txt_attn.norm.query_norm.scale";
hy_name_map[block_prefix + "txt_attn_k_norm.weight"] = dst_prefix + "txt_attn.norm.key_norm.scale";
// ff
hy_name_map[block_prefix + "img_mlp.fc1"] = dst_prefix + "img_mlp.0";
hy_name_map[block_prefix + "img_mlp.fc2"] = dst_prefix + "img_mlp.2";
hy_name_map[block_prefix + "txt_mlp.fc1"] = dst_prefix + "txt_mlp.0";
hy_name_map[block_prefix + "txt_mlp.fc2"] = dst_prefix + "txt_mlp.2";
// output projections
hy_name_map[block_prefix + "img_attn_proj"] = dst_prefix + "img_attn.proj";
hy_name_map[block_prefix + "txt_attn_proj"] = dst_prefix + "txt_attn.proj";
}
}
hy_name_map["time_in.mlp.0"] = "time_in.in_layer";
hy_name_map["time_in.mlp.2"] = "time_in.out_layer";
hy_name_map["time_r_in.mlp.0"] = "time_r_in.in_layer";
hy_name_map["time_r_in.mlp.2"] = "time_r_in.out_layer";
hy_name_map["vector_in.mlp.0"] = "vector_in.in_layer";
hy_name_map["vector_in.mlp.2"] = "vector_in.out_layer";
hy_name_map["guidance_in.mlp.0"] = "guidance_in.in_layer";
hy_name_map["guidance_in.mlp.2"] = "guidance_in.out_layer";
hy_name_map["txt_in.c_embedder.linear_1"] = "txt_in.c_embedder.in_layer";
hy_name_map["txt_in.c_embedder.linear_2"] = "txt_in.c_embedder.out_layer";
hy_name_map["txt_in.t_embedder.mlp.0"] = "txt_in.t_embedder.in_layer";
hy_name_map["txt_in.t_embedder.mlp.2"] = "txt_in.t_embedder.out_layer";
replace_with_prefix_map(name, hy_name_map);
static const std::vector<std::pair<std::string, std::string>> generic_name_map = {
{"_attn_qkv.", "_attn.qkv."},
{"_attn_proj.", "_attn.proj."},
{"mlp.fc1.", "mlp.0."},
{"mlp.fc2.", "mlp.2."},
{".modulation.linear.", ".modulation.lin."},
};
replace_with_name_map(name, generic_name_map);
return name;
}
std::string convert_diffusers_dit_to_original_lumina2(std::string name) {
int num_layers = 30;
int num_refiner_layers = 2;
@@ -807,6 +873,8 @@ std::string convert_diffusion_model_name(std::string name, std::string prefix, S
name = convert_diffusers_dit_to_original_sd3(name);
} else if (sd_version_is_flux(version) || sd_version_is_flux2(version) || sd_version_is_longcat(version) || sd_version_is_sefi_image(version)) {
name = convert_diffusers_dit_to_original_flux(name);
} else if (sd_version_is_hunyuan_video(version)) {
name = convert_hunyuan_video_to_original_flux(name);
} else if (sd_version_is_z_image(version)) {
name = convert_diffusers_dit_to_original_lumina2(name);
} else if (sd_version_is_anima(version)) {
@@ -980,6 +1048,9 @@ std::string convert_diffusers_to_original_wan_vae(std::string name) {
}
std::string convert_first_stage_model_name(std::string name, std::string prefix, SDVersion version) {
if (sd_version_is_hunyuan_video(version)) {
return name;
}
if (sd_version_uses_wan_vae(version)) {
return convert_diffusers_to_original_wan_vae(name);
}