fix img2vid

Author: stduhpf

stable-diffusion.cpp

@@ -3633,7 +3633,8 @@ SD_API sd_image_t* generate_video(sd_ctx_t* sd_ctx, const sd_vid_gen_params_t* s
         denoise_mask = ggml_new_tensor_4d(work_ctx, GGML_TYPE_F32, init_latent->ne[0], init_latent->ne[1], init_latent->ne[2], 1);
         ggml_set_f32(denoise_mask, 1.f);
 
-        sd_ctx->sd->process_latent_out(init_latent);
+        if (!sd_ctx->sd->use_tiny_autoencoder)
+            sd_ctx->sd->process_latent_out(init_latent);
 
         ggml_ext_tensor_iter(init_image_latent, [&](ggml_tensor* t, int64_t i0, int64_t i1, int64_t i2, int64_t i3) {
             float value = ggml_ext_tensor_get_f32(t, i0, i1, i2, i3);
@@ -3643,7 +3644,8 @@ SD_API sd_image_t* generate_video(sd_ctx_t* sd_ctx, const sd_vid_gen_params_t* s
             }
         });
 
-        sd_ctx->sd->process_latent_in(init_latent);
+        if (!sd_ctx->sd->use_tiny_autoencoder)
+            sd_ctx->sd->process_latent_in(init_latent);
 
         int64_t t2 = ggml_time_ms();
         LOG_INFO("encode_first_stage completed, taking %" PRId64 " ms", t2 - t1);