diff --git a/video/out/opengl/utils.c b/video/out/opengl/utils.c
index 0bebe1263c..fa1131e815 100644
--- a/video/out/opengl/utils.c
+++ b/video/out/opengl/utils.c
@@ -481,6 +481,9 @@ struct gl_shader_cache {
     struct sc_uniform *uniforms;
     int num_uniforms;
 
+    // For checking that the user is calling gl_sc_reset() properly.
+    bool needs_reset;
+
     bool error_state; // true if an error occurred
 
     // temporary buffers (avoids frequent reallocations)
@@ -494,17 +497,26 @@ struct gl_shader_cache *gl_sc_create(GL *gl, struct mp_log *log)
         .gl = gl,
         .log = log,
     };
+    gl_sc_reset(sc);
     return sc;
 }
 
+// Reset the previous pass. This must be called after
+// Unbind all GL state managed by sc - the current program and texture units.
 void gl_sc_reset(struct gl_shader_cache *sc)
 {
+    GL *gl = sc->gl;
+
+    if (sc->needs_reset)
+        gl->UseProgram(0);
+
     sc->prelude_text.len = 0;
     sc->header_text.len = 0;
     sc->text.len = 0;
     for (int n = 0; n < sc->num_uniforms; n++)
         talloc_free(sc->uniforms[n].name);
     sc->num_uniforms = 0;
+    sc->needs_reset = false;
 }
 
 static void sc_flush_cache(struct gl_shader_cache *sc)
@@ -867,14 +879,20 @@ static GLuint create_program(struct gl_shader_cache *sc, const char *vertex,
 // 1. Generate vertex and fragment shaders from the fragment shader text added
 //    with gl_sc_add(). The generated shader program is cached (based on the
 //    text), so actual compilation happens only the first time.
-// 2. Update the uniforms set with gl_sc_uniform_*.
+// 2. Update the uniforms and textures set with gl_sc_uniform_*.
 // 3. Make the new shader program current (glUseProgram()).
-// 4. Reset the sc state and prepare for a new shader program. (All uniforms
+// After that, you render, and then you call gc_sc_reset(), which does:
+// 1. Unbind the program and all textures.
+// 2. Reset the sc state and prepare for a new shader program. (All uniforms
 //    and fragment operations needed for the next program have to be re-added.)
-void gl_sc_gen_shader_and_reset(struct gl_shader_cache *sc)
+void gl_sc_generate(struct gl_shader_cache *sc)
 {
     GL *gl = sc->gl;
 
+    // gl_sc_reset() must be called after ending the previous render process,
+    // and before starting a new one.
+    assert(!sc->needs_reset);
+
     assert(sc->vao);
 
     for (int n = 0; n < MP_ARRAY_SIZE(sc->tmp); n++)
@@ -997,7 +1015,7 @@ void gl_sc_gen_shader_and_reset(struct gl_shader_cache *sc)
     for (int n = 0; n < sc->num_uniforms; n++)
         update_uniform(gl, entry, &sc->uniforms[n], n);
 
-    gl_sc_reset(sc);
+    sc->needs_reset = true;
 }
 
 // Maximum number of simultaneous query objects to keep around. Reducing this
diff --git a/video/out/opengl/utils.h b/video/out/opengl/utils.h
index d81599a71f..ab3c13c915 100644
--- a/video/out/opengl/utils.h
+++ b/video/out/opengl/utils.h
@@ -167,8 +167,9 @@ void gl_sc_uniform_mat3(struct gl_shader_cache *sc, char *name,
                         bool transpose, GLfloat *v);
 void gl_sc_set_vao(struct gl_shader_cache *sc, struct gl_vao *vao);
 void gl_sc_enable_extension(struct gl_shader_cache *sc, char *name);
-void gl_sc_gen_shader_and_reset(struct gl_shader_cache *sc);
+void gl_sc_generate(struct gl_shader_cache *sc);
 void gl_sc_reset(struct gl_shader_cache *sc);
+void gl_sc_unbind(struct gl_shader_cache *sc);
 
 struct gl_timer;
 
diff --git a/video/out/opengl/video.c b/video/out/opengl/video.c
index 040dbb2f3d..b000b2e7ce 100644
--- a/video/out/opengl/video.c
+++ b/video/out/opengl/video.c
@@ -1026,10 +1026,11 @@ static void finish_pass_direct(struct gl_video *p, GLint fbo, int vp_w, int vp_h
 {
     GL *gl = p->gl;
     pass_prepare_src_tex(p);
+    gl_sc_generate(p->sc);
     gl->BindFramebuffer(GL_FRAMEBUFFER, fbo);
-    gl_sc_gen_shader_and_reset(p->sc);
     render_pass_quad(p, vp_w, vp_h, dst);
     gl->BindFramebuffer(GL_FRAMEBUFFER, 0);
+    gl_sc_reset(p->sc);
     memset(&p->pass_tex, 0, sizeof(p->pass_tex));
     p->pass_tex_num = 0;
 }
@@ -2335,8 +2336,9 @@ static void pass_draw_osd(struct gl_video *p, int draw_flags, double pts,
             pass_colormanage(p, csp_srgb, true);
         }
         gl_sc_set_vao(p->sc, mpgl_osd_get_vao(p->osd));
-        gl_sc_gen_shader_and_reset(p->sc);
+        gl_sc_generate(p->sc);
         mpgl_osd_draw_part(p->osd, vp_w, vp_h, n);
+        gl_sc_reset(p->sc);
     }
     gl_sc_set_vao(p->sc, &p->vao);
 }
diff --git a/video/out/vo_rpi.c b/video/out/vo_rpi.c
index c046928f38..acec865c6b 100644
--- a/video/out/vo_rpi.c
+++ b/video/out/vo_rpi.c
@@ -285,8 +285,9 @@ static void update_osd(struct vo *vo)
             abort();
         }
         gl_sc_set_vao(p->sc, mpgl_osd_get_vao(p->osd));
-        gl_sc_gen_shader_and_reset(p->sc);
+        gl_sc_generate(p->sc);
         mpgl_osd_draw_part(p->osd, p->osd_res.w, -p->osd_res.h, n);
+        gl_sc_reset(p->sc);
     }
 
     MP_STATS(vo, "stop rpi_osd");