DarthAffe/stable-diffusion.cpp
1
0 Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-11-11 11:26:34 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat: add P-H3-Image generation and editing support (#2272)

Browse Source
This commit is contained in:
leejet authored and GitHub committed 2026-11-04 23:24:51 +08:00
1 parent 984eb0686c
commit 3e92bf990a
18 files changed
+421 -41
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/pipeline/diffusion_engine.cpp
+12 -6
View File
@@ -778,7 +778,7 @@ bool StableDiffusionGGML::init_model_loader(ModelLoader& model_loader, ModelConf
if (strlen(SAFE_STR(sd_ctx_params->embeddings_connectors_path)) > 0) {
LOG_INFO("loading embeddings connectors from '%s'", sd_ctx_params->embeddings_connectors_path);
if (!load_model("embeddings_connectors", sd_ctx_params->embeddings_connectors_path)) {
if (!load_model("embeddings_connectors", sd_ctx_params->embeddings_connectors_path, "embeddings_connectors.")) {
LOG_WARN("loading embeddings connectors from '%s' failed", sd_ctx_params->embeddings_connectors_path);
}
}
@@ -1491,7 +1491,7 @@ bool StableDiffusionGGML::build_denoiser() {
}
case H3_AV_FLOW_PRED: {
LOG_INFO("running in MiniMax H3 AV FLOW mode");
denoiser = std::make_shared<H3AVFlowDenoiser>(default_flow_shift, 3.f, get_latent_channel());
denoiser = std::make_shared<H3AVFlowDenoiser>(default_flow_shift, version == VERSION_P_H3_IMAGE ? 1.f : 3.f, get_latent_channel());
break;
}
case MING_IMAGE_FLOW_PRED: {
@@ -2594,9 +2594,9 @@ sd::Tensor<float> StableDiffusionGGML::sample(const std::shared_ptr<DiffusionMod
condition.c_position_ids.empty() ? nullptr : &condition.c_position_ids,
condition.c_ref_audios.empty() ? nullptr : &condition.c_ref_audios,
condition.c_reference_blocks.empty() ? nullptr : &condition.c_reference_blocks,
audio_length,
version == VERSION_P_H3_IMAGE ? 2 : audio_length,
std::isfinite(active_flow_shift) ? active_flow_shift : 12.f,
3.f};
version == VERSION_P_H3_IMAGE ? 1.f : 3.f};
} else if (sd_version_is_ltxav(version)) {
diffusion_params.extra = LTXAVDiffusionExtra{
nullptr,
@@ -3007,7 +3007,10 @@ sd::Tensor<float> StableDiffusionGGML::decode_first_stage(const sd::Tensor<float
if (sd_version_is_pid(version) || sd_version_is_minit2i(version) || version == VERSION_IRIS) {
return sd::ops::clamp((x + 1.f) * 0.5f, 0.0f, 1.0f);
}
auto latents = first_stage_model->diffusion_to_vae_latents(x);
auto image_latents = version == VERSION_P_H3_IMAGE && x.dim() == 5
? sd::ops::slice(x, 3, 0, get_latent_channel())
: x;
auto latents = first_stage_model->diffusion_to_vae_latents(image_latents);
auto tiling_params = first_stage_model->resolve_tiling_params(vae_tiling_params);
const bool prefer_temporal_tiling = decode_video && latents.dim() == 5 && latents.shape()[2] > 1 &&
first_stage_model->can_temporal_tile_decode();
@@ -3021,6 +3024,9 @@ sd::Tensor<float> StableDiffusionGGML::decode_first_stage(const sd::Tensor<float
return {};
}
auto decoded = first_stage_model->decode(n_threads, latents, tiling_params, decode_video, circular_x, circular_y);
if (version == VERSION_P_H3_IMAGE && !decoded.empty() && decoded.dim() == 5) {
decoded = decoded.squeeze(2);
}
if (!decoded.empty() ||
!sd::backend_fit::prepare_vae_decode_retry_tiling(tiling_params, prefer_temporal_tiling,
first_stage_model->last_compute_status(),
@@ -3220,7 +3226,7 @@ RefImageParams StableDiffusionGGML::resolve_ref_image_params(const char* ref_ima
if (params.force_ref_timestep_zero && !sd_version_is_krea2(version)) {
LOG_WARN("force_ref_timestep_zero is only supported by Krea2 architecture for now");
}
if (sd_version_is_minimax_h3(version)) {
if (version == VERSION_MINIMAX_H3) {
// H3 consumes references as VAE latents in its own packed sequence rather than
// through the generic conditioner/DiT reference path, so ref_image_size is the
// only arg it acts on. Everything else is reported as ineffective rather than