Skip to content

fix: avoid segfault for pix2pix models with no reference images #766

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 2 commits into
base: master
Choose a base branch
from
+26 −5
Open

fix: avoid segfault for pix2pix models with no reference images #766

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
9c7cb44
fix: avoid segfault for pix2pix models with no reference images
wbruna Aug 14, 2025
9dae12f
fix: default to empty reference on pix2pix models to avoid segfault
wbruna Aug 15, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 26 additions & 5 deletions stable-diffusion.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -448,6 +448,10 @@ class StableDiffusionGGML {
diffusion_model->alloc_params_buffer();
diffusion_model->get_param_tensors(tensors);

if (sd_version_is_unet_edit(version)) {
vae_decode_only = false;
}

if (high_noise_diffusion_model) {
high_noise_diffusion_model->alloc_params_buffer();
high_noise_diffusion_model->get_param_tensors(tensors);
Expand Down Expand Up @@ -2319,19 +2323,36 @@ sd_image_t* generate_image(sd_ctx_t* sd_ctx, const sd_img_gen_params_t* sd_img_g
init_latent = generate_init_latent(sd_ctx, work_ctx, width, height);
}

sd_guidance_params_t guidance = sd_img_gen_params->sample_params.guidance;
std::vector<sd_image_t*> ref_images;
for (int i = 0; i < sd_img_gen_params->ref_images_count; i++) {
ref_images.push_back(&sd_img_gen_params->ref_images[i]);
}

std::vector<uint8_t> empty_image_data;
sd_image_t empty_image = {(uint32_t)width, (uint32_t)height, 3, nullptr};
if (ref_images.empty() && sd_version_is_unet_edit(sd_ctx->sd->version))
{
LOG_WARN("This model needs at least one reference image; using an empty reference");
empty_image_data.reserve(width * height * 3);
ref_images.push_back(&empty_image);
empty_image.data = empty_image_data.data();
guidance.img_cfg = 0.f;
}

if (sd_img_gen_params->ref_images_count > 0) {
LOG_INFO("EDIT mode");
}

std::vector<ggml_tensor*> ref_latents;
for (int i = 0; i < sd_img_gen_params->ref_images_count; i++) {
for (int i = 0; i < ref_images.size(); i++) {
ggml_tensor* img = ggml_new_tensor_4d(work_ctx,
GGML_TYPE_F32,
sd_img_gen_params->ref_images[i].width,
sd_img_gen_params->ref_images[i].height,
ref_images[i]->width,
ref_images[i]->height,
3,
1);
sd_image_to_tensor(sd_img_gen_params->ref_images[i].data, img);
sd_image_to_tensor(ref_images[i]->data, img);

ggml_tensor* latent = NULL;
if (sd_ctx->sd->use_tiny_autoencoder) {
Expand Down Expand Up @@ -2364,7 +2385,7 @@ sd_image_t* generate_image(sd_ctx_t* sd_ctx, const sd_img_gen_params_t* sd_img_g
SAFE_STR(sd_img_gen_params->prompt),
SAFE_STR(sd_img_gen_params->negative_prompt),
sd_img_gen_params->clip_skip,
sd_img_gen_params->sample_params.guidance,
guidance,
sd_img_gen_params->sample_params.eta,
width,
height,
Expand Down
Loading