Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-18 16:55:14 +02:00
Code Issues Packages Projects Releases Wiki Activity

sd: sync to master-642-3a8788c

Browse Source
This commit is contained in:
Wagner Bruna
2026-05-21 20:46:14 -03:00
parent a0413bdf55
commit 8427efb4c6
12 changed files with 859 additions and 440 deletions
Download Patch File Download Diff File Expand all files Collapse all files
otherarch/sdcpp/stable-diffusion.cpp
+17 -7
View File
@@ -155,7 +155,7 @@ public:
bool kcpp_lora_cache_populate = false;
std::string taesd_path;
sd_tiling_params_t vae_tiling_params = {false, false, 0, 0, 0.5f, 0, 0};
sd_tiling_params_t vae_tiling_params = {false, false, 0, 0, 0.5f, 0, 0, nullptr};
bool offload_params_to_cpu = false;
float max_vram = 0.f;
bool use_pmid = false;
@@ -2921,7 +2921,7 @@ void sd_img_gen_params_init(sd_img_gen_params_t* sd_img_gen_params) {
sd_img_gen_params->batch_count = 1;
sd_img_gen_params->control_strength = 0.9f;
sd_img_gen_params->pm_params = {nullptr, 0, nullptr, 20.f};
sd_img_gen_params->vae_tiling_params = {false, false, 0, 0, 0.5f, 0.0f, 0.0f};
sd_img_gen_params->vae_tiling_params = {false, false, 0, 0, 0.5f, 0.0f, 0.0f, nullptr};
sd_cache_params_init(&sd_img_gen_params->cache);
sd_hires_params_init(&sd_img_gen_params->hires);
}
@@ -2950,7 +2950,7 @@ char* sd_img_gen_params_to_str(const sd_img_gen_params_t* sd_img_gen_params) {
"increase_ref_index: %s\n"
"control_strength: %.2f\n"
"photo maker: {style_strength = %.2f, id_images_count = %d, id_embed_path = %s}\n"
"VAE tiling: %s (temporal=%s)\n"
"VAE tiling: %s (temporal=%s, extra_tiling_args=%s)\n"
"hires: {enabled=%s, upscaler=%s, model_path=%s, scale=%.2f, target=%dx%d, steps=%d, denoising_strength=%.2f}\n",
SAFE_STR(sd_img_gen_params->prompt),
SAFE_STR(sd_img_gen_params->negative_prompt),
@@ -2970,6 +2970,7 @@ char* sd_img_gen_params_to_str(const sd_img_gen_params_t* sd_img_gen_params) {
SAFE_STR(sd_img_gen_params->pm_params.id_embed_path),
BOOL_STR(sd_img_gen_params->vae_tiling_params.enabled),
BOOL_STR(sd_img_gen_params->vae_tiling_params.temporal_tiling),
SAFE_STR(sd_img_gen_params->vae_tiling_params.extra_tiling_args),
BOOL_STR(sd_img_gen_params->hires.enabled),
sd_hires_upscaler_name(sd_img_gen_params->hires.upscaler),
SAFE_STR(sd_img_gen_params->hires.model_path),
@@ -3007,7 +3008,7 @@ void sd_vid_gen_params_init(sd_vid_gen_params_t* sd_vid_gen_params) {
sd_vid_gen_params->fps = 16;
sd_vid_gen_params->moe_boundary = 0.875f;
sd_vid_gen_params->vace_strength = 1.f;
sd_vid_gen_params->vae_tiling_params = {false, false, 0, 0, 0.5f, 0.0f, 0.0f};
sd_vid_gen_params->vae_tiling_params = {false, false, 0, 0, 0.5f, 0.0f, 0.0f, nullptr};
sd_vid_gen_params->hires.enabled = false;
sd_vid_gen_params->hires.upscaler = SD_HIRES_UPSCALER_LATENT;
sd_vid_gen_params->hires.scale = 2.f;
@@ -5460,14 +5461,24 @@ SD_API bool generate_video(sd_ctx_t* sd_ctx,
sd_ctx->sd->diffusion_model->free_params_buffer();
}
int64_t latent_end = ggml_time_ms();
LOG_INFO("generating latent video completed, taking %.2fs", (latent_end - latent_start) * 1.0f / 1000);
sd_audio_t* generated_audio = nullptr;
if (sd_version_is_ltxav(sd_ctx->sd->version) &&
latents.audio_length > 0 &&
sd_ctx->sd->audio_vae_model != nullptr) {
int64_t audio_latent_decode_start = ggml_time_ms();
auto audio_latent = unpack_ltxav_audio_latent(final_latent,
latents.audio_length,
sd_ctx->sd->get_latent_channel());
if (!audio_latent.empty()) {
LOG_DEBUG("decode audio latent %dx%dx%dx%d",
(int)audio_latent.shape()[0],
(int)audio_latent.shape()[1],
(int)audio_latent.shape()[2],
(int)audio_latent.shape()[3]);
auto waveform = sd_ctx->sd->decode_ltx_audio_latent(audio_latent);
if (!waveform.empty()) {
generated_audio = waveform_to_sd_audio(sd_ctx->sd, waveform);
@@ -5475,6 +5486,8 @@ SD_API bool generate_video(sd_ctx_t* sd_ctx,
LOG_WARN("LTX audio latent decode failed; continuing with silent video output");
}
}
int64_t audio_latent_decode_end = ggml_time_ms();
LOG_INFO("decoding audio latent completed, taking %.2fs", (audio_latent_decode_end - audio_latent_decode_start) * 1.0f / 1000);
}
if (latents.video_conditioning_frame_count > 0) {
@@ -5487,9 +5500,6 @@ SD_API bool generate_video(sd_ctx_t* sd_ctx,
final_latent = sd::ops::slice(final_latent, 2, latents.ref_image_num, final_latent.shape()[2]);
}
int64_t latent_end = ggml_time_ms();
LOG_INFO("generating latent video completed, taking %.2fs", (latent_end - latent_start) * 1.0f / 1000);
auto result = decode_video_outputs(sd_ctx, latent_upscale_enabled ? hires_request : request, final_latent, num_frames_out);
if (result == nullptr) {
free_sd_audio(generated_audio);