diff --git a/api/onnx_web/diffusers/run.py b/api/onnx_web/diffusers/run.py
index e651e7f5..27e72a63 100644
--- a/api/onnx_web/diffusers/run.py
+++ b/api/onnx_web/diffusers/run.py
@@ -32,8 +32,9 @@ def run_txt2img_pipeline(
 ) -> None:
     # TODO: add to params
     highres_scale = 4
-    highres_steps = 15
-    highres_strength = 0.5
+    highres_steps = 25
+    highres_strength = 0.2
+    highres_steps_post = int((params.steps - highres_steps) / highres_strength)
 
     latents = get_latents_from_seed(params.seed, size, batch=params.batch)
 
@@ -87,8 +88,9 @@ def run_txt2img_pipeline(
 
     for image, output in zip(result.images, outputs):
         if highres_scale > 1:
+            highres_progress = ChainProgress.from_progress(progress)
             # load img2img pipeline once
-            highpipe = load_pipeline(
+            highres_pipe = load_pipeline(
                 server,
                 OnnxStableDiffusionImg2ImgPipeline,
                 params.model,
@@ -104,32 +106,32 @@ def run_txt2img_pipeline(
                 if params.lpw:
                     logger.debug("using LPW pipeline for img2img")
                     rng = torch.manual_seed(params.seed)
-                    result = highpipe.img2img(
+                    result = highres_pipe.img2img(
                         tile,
                         params.prompt,
                         generator=rng,
                         guidance_scale=params.cfg,
                         negative_prompt=params.negative_prompt,
                         num_images_per_prompt=1,
-                        num_inference_steps=params.steps - highres_steps,
+                        num_inference_steps=highres_steps_post,
                         strength=highres_strength,
                         eta=params.eta,
-                        callback=progress,
+                        callback=highres_progress,
                     )
                     return result.images[0]
                 else:
                     rng = np.random.RandomState(params.seed)
-                    result = highpipe(
+                    result = highres_pipe(
                         params.prompt,
                         tile,
                         generator=rng,
                         guidance_scale=params.cfg,
                         negative_prompt=params.negative_prompt,
                         num_images_per_prompt=1,
-                        num_inference_steps=params.steps - highres_steps,
+                        num_inference_steps=highres_steps_post,
                         strength=highres_strength,
                         eta=params.eta,
-                        callback=progress,
+                        callback=highres_progress,
                     )
                     return result.images[0]