updates

app.py

@@ -74,6 +74,8 @@ def encode_target_images(images, pipeline):
 
 @spaces.GPU(duration=120)
 def generate_image(text, img1, img2, img3, guidance_scale, inference_steps, seed, enable_cpu_offload=False, neg_prompt="", true_cfg=1.0, image_cfg=0.0):
+    if neg_prompt == "":
+        neg_prompt = "3d render, cartoon, low resolution, illustration, blurry, unrealistic"
     if enable_cpu_offload:
         pipeline.enable_sequential_cpu_offload()
     input_images = [img1, img2, img3]
@@ -101,7 +103,7 @@ def generate_image(text, img1, img2, img3, guidance_scale, inference_steps, seed
         num_inference_steps=inference_steps,
         height=512,
         width=numref * 512,
-        generator = torch.Generator(device="cpu").manual_seed(seed),
+        generator = torch.Generator(device="cuda").manual_seed(seed),
         joint_attention_kwargs={'shared_attn': True, 'num': numref},
         return_dict=False,
         negative_prompt=neg_prompt,

pipelines/flux_pipeline/pipeline.py

@@ -110,7 +110,6 @@ def normalized_guidance_image(neg_noise_pred, noise_pred, image_noise_pred, true
     pred_guided = image_noise_pred + image_cfg_scale * diff_img + true_cfg_scale *  diff_txt
     return pred_guided
 
-
 class SynCDFluxPipeline(FluxPipeline):
     
     model_cpu_offload_seq = "text_encoder->text_encoder_2->transformer->vae"