DarthAffe/stable-diffusion.cpp
1
0
Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2025-12-12 13:28:37 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat: align the spatial size to the corresponding multiple (#1073)

Browse Source
This commit is contained in:
leejet 2025-12-10 23:15:08 +08:00 committed by GitHub
parent 1ac5a616de
commit 8823dc48bc
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 44 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
ggml_extend.hpp
View File

@ -60,6 +60,14 @@
#define SD_UNUSED(x) (void)(x) #define SD_UNUSED(x) (void)(x)
#endif #endif
__STATIC_INLINE__ int align_up_offset(int n, int multiple) {
return (multiple - n % multiple) % multiple;
}
__STATIC_INLINE__ int align_up(int n, int multiple) {
return n + align_up_offset(n, multiple);
}
__STATIC_INLINE__ void ggml_log_callback_default(ggml_log_level level, const char* text, void*) { __STATIC_INLINE__ void ggml_log_callback_default(ggml_log_level level, const char* text, void*) {
switch (level) { switch (level) {
case GGML_LOG_LEVEL_DEBUG: case GGML_LOG_LEVEL_DEBUG:

49
stable-diffusion.cpp
View File

@ -1898,6 +1898,18 @@ public:
return vae_scale_factor; return vae_scale_factor;
} }
int get_diffusion_model_down_factor() {
int down_factor = 8; // unet
if (sd_version_is_dit(version)) {
if (sd_version_is_wan(version)) {
down_factor = 2;
} else {
down_factor = 1;
}
}
return down_factor;
}
int get_latent_channel() { int get_latent_channel() {
int latent_channel = 4; int latent_channel = 4;
if (sd_version_is_dit(version)) { if (sd_version_is_dit(version)) {
@ -3133,22 +3145,19 @@ sd_image_t* generate_image(sd_ctx_t* sd_ctx, const sd_img_gen_params_t* sd_img_g
sd_ctx->sd->vae_tiling_params = sd_img_gen_params->vae_tiling_params; sd_ctx->sd->vae_tiling_params = sd_img_gen_params->vae_tiling_params;
int width = sd_img_gen_params->width; int width = sd_img_gen_params->width;
int height = sd_img_gen_params->height; int height = sd_img_gen_params->height;
int vae_scale_factor = sd_ctx->sd->get_vae_scale_factor(); int vae_scale_factor = sd_ctx->sd->get_vae_scale_factor();
if (sd_version_is_dit(sd_ctx->sd->version)) { int diffusion_model_down_factor = sd_ctx->sd->get_diffusion_model_down_factor();
if (width % 16 || height % 16) { int spatial_multiple = vae_scale_factor * diffusion_model_down_factor;
LOG_ERROR("Image dimensions must be must be a multiple of 16 on each axis for %s models. (Got %dx%d)",
model_version_to_str[sd_ctx->sd->version], int width_offset = align_up_offset(width, spatial_multiple);
width, int height_offset = align_up_offset(height, spatial_multiple);
height); if (width_offset > 0 || height_offset > 0) {
return nullptr; width += width_offset;
} height += height_offset;
} else if (width % 64 || height % 64) { LOG_WARN("align up %dx%d to %dx%d (multiple=%d)", sd_img_gen_params->width, sd_img_gen_params->height, width, height, spatial_multiple);
LOG_ERROR("Image dimensions must be must be a multiple of 64 on each axis for %s models. (Got %dx%d)",
model_version_to_str[sd_ctx->sd->version],
width,
height);
return nullptr;
} }
LOG_DEBUG("generate_image %dx%d", width, height); LOG_DEBUG("generate_image %dx%d", width, height);
if (sd_ctx == nullptr || sd_img_gen_params == nullptr) { if (sd_ctx == nullptr || sd_img_gen_params == nullptr) {
return nullptr; return nullptr;
@ -3422,9 +3431,19 @@ SD_API sd_image_t* generate_video(sd_ctx_t* sd_ctx, const sd_vid_gen_params_t* s
int frames = sd_vid_gen_params->video_frames; int frames = sd_vid_gen_params->video_frames;
frames = (frames - 1) / 4 * 4 + 1; frames = (frames - 1) / 4 * 4 + 1;
int sample_steps = sd_vid_gen_params->sample_params.sample_steps; int sample_steps = sd_vid_gen_params->sample_params.sample_steps;
LOG_INFO("generate_video %dx%dx%d", width, height, frames);
int vae_scale_factor = sd_ctx->sd->get_vae_scale_factor(); int vae_scale_factor = sd_ctx->sd->get_vae_scale_factor();
int diffusion_model_down_factor = sd_ctx->sd->get_diffusion_model_down_factor();
int spatial_multiple = vae_scale_factor * diffusion_model_down_factor;
int width_offset = align_up_offset(width, spatial_multiple);
int height_offset = align_up_offset(height, spatial_multiple);
if (width_offset > 0 || height_offset > 0) {
width += width_offset;
height += height_offset;
LOG_WARN("align up %dx%d to %dx%d (multiple=%d)", sd_vid_gen_params->width, sd_vid_gen_params->height, width, height, spatial_multiple);
}
LOG_INFO("generate_video %dx%dx%d", width, height, frames);
enum sample_method_t sample_method = sd_vid_gen_params->sample_params.sample_method; enum sample_method_t sample_method = sd_vid_gen_params->sample_params.sample_method;
if (sample_method == SAMPLE_METHOD_COUNT) { if (sample_method == SAMPLE_METHOD_COUNT) {