mirror of
https://github.com/leejet/stable-diffusion.cpp.git
synced 2026-08-10 06:07:58 +00:00
Compare commits
No commits in common. "87a01773be23b996e38217a6a574c2de08ac560f" and "78124b64544d2e5d6cc1036e0c3a4db0b5973367" have entirely different histories.
87a01773be
...
78124b6454
@ -224,7 +224,6 @@ public:
|
||||
std::shared_ptr<ControlNet> control_net;
|
||||
std::shared_ptr<IPAdapter::IPAdapterRunner> ip_adapter;
|
||||
sd::Tensor<float> ip_adapter_tokens;
|
||||
sd::Tensor<float> ip_adapter_uncond_tokens;
|
||||
float ip_adapter_strength = 1.0f;
|
||||
std::vector<std::shared_ptr<GenerationExtension>> generation_extensions;
|
||||
std::vector<std::shared_ptr<LoraModel>> runtime_lora_models;
|
||||
@ -2101,9 +2100,8 @@ public:
|
||||
}
|
||||
|
||||
void compute_ip_adapter_tokens(const sd_image_t& image, float strength) {
|
||||
ip_adapter_tokens = {};
|
||||
ip_adapter_uncond_tokens = {};
|
||||
ip_adapter_strength = strength;
|
||||
ip_adapter_tokens = {};
|
||||
ip_adapter_strength = strength;
|
||||
if (ip_adapter == nullptr || clip_vision == nullptr || image.data == nullptr) {
|
||||
return;
|
||||
}
|
||||
@ -2113,19 +2111,10 @@ public:
|
||||
return;
|
||||
}
|
||||
ip_adapter_tokens = ip_adapter->compute(n_threads, embed);
|
||||
if (ip_adapter_tokens.empty()) {
|
||||
LOG_ERROR("IP-Adapter conditional image projection failed");
|
||||
return;
|
||||
if (!ip_adapter_tokens.empty()) {
|
||||
LOG_INFO("IP-Adapter: %lld image tokens, strength %.2f",
|
||||
(long long)ip_adapter_tokens.shape()[1], strength);
|
||||
}
|
||||
auto uncond_embed = sd::Tensor<float>::zeros_like(embed);
|
||||
ip_adapter_uncond_tokens = ip_adapter->compute(n_threads, uncond_embed);
|
||||
if (ip_adapter_uncond_tokens.empty()) {
|
||||
LOG_ERROR("IP-Adapter unconditional image projection failed");
|
||||
ip_adapter_tokens = {};
|
||||
return;
|
||||
}
|
||||
LOG_INFO("IP-Adapter: %lld image tokens, strength %.2f",
|
||||
(long long)ip_adapter_tokens.shape()[1], strength);
|
||||
}
|
||||
|
||||
std::vector<float> process_timesteps(const std::vector<float>& timesteps,
|
||||
@ -2617,7 +2606,7 @@ public:
|
||||
const sd::Tensor<float>* c_concat_override = nullptr,
|
||||
const std::vector<int>* local_skip_layers = nullptr,
|
||||
const std::vector<sd::Tensor<float>>* ref_latents_override = nullptr,
|
||||
bool use_uncond_ip = false) -> sd::Tensor<float> {
|
||||
bool apply_ip = true) -> sd::Tensor<float> {
|
||||
diffusion_params.context = condition.c_crossattn.empty() ? nullptr : &condition.c_crossattn;
|
||||
diffusion_params.c_concat = c_concat_override != nullptr ? c_concat_override : (condition.c_concat.empty() ? nullptr : &condition.c_concat);
|
||||
diffusion_params.y = condition.c_vector.empty() ? nullptr : &condition.c_vector;
|
||||
@ -2629,9 +2618,8 @@ public:
|
||||
nvf = static_cast<int>(noised_input.shape()[3]);
|
||||
}
|
||||
UNetDiffusionExtra unet_extra{nvf, &controls, control_strength};
|
||||
const auto& ip_tokens = use_uncond_ip ? ip_adapter_uncond_tokens : ip_adapter_tokens;
|
||||
if (!ip_tokens.empty()) {
|
||||
unet_extra.ip_context = &ip_tokens;
|
||||
if (apply_ip && !ip_adapter_tokens.empty()) {
|
||||
unet_extra.ip_context = &ip_adapter_tokens;
|
||||
unet_extra.ip_scale = ip_adapter_strength;
|
||||
}
|
||||
diffusion_params.extra = unet_extra;
|
||||
@ -2727,7 +2715,7 @@ public:
|
||||
uncond.c_concat.empty() ? nullptr : &uncond.c_concat,
|
||||
uncond_skip_layers,
|
||||
nullptr,
|
||||
true);
|
||||
false);
|
||||
if (uncond_out.empty()) {
|
||||
return {};
|
||||
}
|
||||
@ -2737,7 +2725,7 @@ public:
|
||||
img_uncond.c_concat.empty() ? nullptr : &img_uncond.c_concat,
|
||||
nullptr,
|
||||
uncond_without_ref_latents ? &empty_ref_latents : nullptr,
|
||||
true);
|
||||
false);
|
||||
if (img_uncond_out.empty()) {
|
||||
return {};
|
||||
}
|
||||
@ -3585,22 +3573,21 @@ char* sd_sample_params_to_str(const sd_sample_params_t* sample_params) {
|
||||
void sd_img_gen_params_init(sd_img_gen_params_t* sd_img_gen_params) {
|
||||
*sd_img_gen_params = {};
|
||||
sd_sample_params_init(&sd_img_gen_params->sample_params);
|
||||
sd_img_gen_params->clip_skip = -1;
|
||||
sd_img_gen_params->ref_images_count = 0;
|
||||
sd_img_gen_params->ref_image_args = "";
|
||||
sd_img_gen_params->width = 512;
|
||||
sd_img_gen_params->height = 512;
|
||||
sd_img_gen_params->strength = 0.75f;
|
||||
sd_img_gen_params->seed = -1;
|
||||
sd_img_gen_params->batch_count = 1;
|
||||
sd_img_gen_params->control_strength = 0.9f;
|
||||
sd_img_gen_params->ip_adapter_strength = 1.0f;
|
||||
sd_img_gen_params->qwen_image_layers = 3;
|
||||
sd_img_gen_params->circular_x = false;
|
||||
sd_img_gen_params->circular_y = false;
|
||||
sd_img_gen_params->pm_params = {nullptr, 0, nullptr, 20.f};
|
||||
sd_img_gen_params->pulid_params = {nullptr, 1.0f};
|
||||
sd_img_gen_params->vae_tiling_params = {false, false, 0, 0, 0.5f, 0.0f, 0.0f, nullptr};
|
||||
sd_img_gen_params->clip_skip = -1;
|
||||
sd_img_gen_params->ref_images_count = 0;
|
||||
sd_img_gen_params->ref_image_args = "";
|
||||
sd_img_gen_params->width = 512;
|
||||
sd_img_gen_params->height = 512;
|
||||
sd_img_gen_params->strength = 0.75f;
|
||||
sd_img_gen_params->seed = -1;
|
||||
sd_img_gen_params->batch_count = 1;
|
||||
sd_img_gen_params->control_strength = 0.9f;
|
||||
sd_img_gen_params->qwen_image_layers = 3;
|
||||
sd_img_gen_params->circular_x = false;
|
||||
sd_img_gen_params->circular_y = false;
|
||||
sd_img_gen_params->pm_params = {nullptr, 0, nullptr, 20.f};
|
||||
sd_img_gen_params->pulid_params = {nullptr, 1.0f};
|
||||
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);
|
||||
}
|
||||
@ -5789,7 +5776,7 @@ static std::optional<ImageGenerationLatents> prepare_video_generation_latents(sd
|
||||
auto encode_condition_frame = [&](const sd::Tensor<float>& image,
|
||||
int64_t latent_frame,
|
||||
const char* name) -> bool {
|
||||
auto encoded = sd_ctx->sd->encode_first_stage(image.unsqueeze(2));
|
||||
auto encoded = sd_ctx->sd->encode_first_stage(image);
|
||||
if (encoded.empty()) {
|
||||
LOG_ERROR("failed to encode Hunyuan Video %s conditioning frame", name);
|
||||
return false;
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user